JungleLab experiments

Element tested: video hook only — the original Feb 2026 Master Wu Wei Chen hook versus a new anchor-standard hook rewrite of the same script; same character, same headline, same primary text, same destination, same CTA. Advertorial destination (17 Jetsetters). Sibling: Exp 106 (same control + V1 hook, quiz destination).

Status: Planned — prepared, dry-run PASS, not enqueued (PREPARED_NOT_ENQUEUED)

Family: wuwei-anchor-hook

Altitude: Rung 5 — hook (video opening)

Date: 2026-09-12

Avatar: The Cursed

Campaign ID: 120237762972300539

Campaign: 20260222 - TBF | Curses - Testing

Destination URL: https://discover.karmaitems.com/tbf-1206-curses-advertorial-ev-001

Destination shorthand: 17 Jetsetters (Cursed)

Creative type: Control = existing Meta video ID 3083132932075391 (live ad 120237763128430539 "20260222 - TBF Exotic Monk Avatar Messages | Master Wu Wei - 17 Jetsetters (Curses)", creative 2113620142520798; reused as existing_media, no re-upload). V1 = new SoulGem authority_talking_head render, 1.0x, output_captioned.mp4, sha256 a0d705842f93cfff2188e8b2386ac6251df7809cee9d810b72e2cbb97714a51e, 222.96 s (run tbf-wuwei-a-jetsetters-20260912).

Ad set name: 20260912 - EXP - TBF - Exp 105 Wu Wei Hook - 17 Jetsetters (Cursed)

Live ad set name: N/A — not launched; PREPARED_NOT_ENQUEUED

Ad set ID: N/A — not launched

Result: Pending — prepared 2026-09-12, not enqueued

Production authorization: Edward 2026-09-12 (session), verbatim: "Queue them alongside the original old control. Obviously get the destiantions right, including on the primary text and disable destination fan out". Batch built and dry-run PASS (mutation-free); enqueue is a separate step and has not been executed — no producer_package.json, no enqueue_result.json, no queue custody, no Meta mutation. Per fresh-pair doctrine the Control (the live original) also appears inside the fresh ad set, so the ad set carries exactly two ads: Control + V1.

Hypothesis: on the identical character, script body, headline, primary text, destination, and audience, the anchor-standard hook (names the curse-placer already in your life, then the slow drain) out-allocates the original Feb 2026 opening, because the Cursed avatar responds to a specific, personal cause named in the first seconds.

Prospect claim: The Cursed already suspects someone close is behind the run of bad luck; a hook that names that person and the slow leak earns the watch-through before the mechanism is explained.

Placement: problem-aware; the 17 Jetsetters advertorial is this avatar's live primary destination in the Curses - Testing campaign.

Novel variable: video hook only (Control = original Wu Wei video; V1 = new Wu Wei render with the anchor hook).

Hold constant:

• Character Master Wu Wei Chen; Control video is the live original (Meta video 3083132932075391).
• Copy: canonical Cursed Sheet row 5 (A5:Z5) headline "Wear It 2 Weeks and Watch Your Luck Turn" and primary text byte-exact as fetched (embedded CTA URL = the advertorial). No override, no swap.
• Destination https://discover.karmaitems.com/tbf-1206-curses-advertorial-ev-001 (byte-exact, no trailing slash); CTA learn_more; both ads carry it.
• Avatar The Cursed, campaign 120237762972300539, template ad set 120247531189150539 targeting (age keys stripped), page 109598535381020, pixel 839353417201669.
• AdLauncher canonical CBO floor (1500 cents, shared contract default since 2026-09-12 06:02 UTC).
• Destination fan-out disabled with operator approval (fixed phrase, approved_by Edward, instruction = the verbatim line above).

Primary test variants:

• Control — 20260912 - EXP - TBF - Exp 105 Control Wu Wei Original Hook - 17 Jetsetters (Cursed)

> existing Meta video 3083132932075391 (live ad 120237763128430539)

• V1 — 20260912 - EXP - TBF - Exp 105 V1 Wu Wei Anchor Hook - 17 Jetsetters (Cursed)

> new render tbf-wuwei-a-jetsetters-20260912, sha256 a0d705842f93cfff2188e8b2386ac6251df7809cee9d810b72e2cbb97714a51e

Decision logic:

• Rank the two arms by Meta allocation (winner = highest spend, last day).
• Arms under $50 lifetime spend are Meta rejections, not rivals, never re-run.
• If both arms starve there is no mature comparison and one fresh-pair confirmation may follow.
• If V1 wins, the anchor hook becomes the default Wu Wei opening for the advertorial destination; if Control wins, the original hook stays. Exp 106 decides the quiz destination independently.

Delivery risk: fresh two-arm CBO ad set in the same campaign; the Control video is also live in its original ad set, so early allocation can thin. Pre-registered NO-DELIVERY branches above. Never inject into the live original ad set.

Why it matters: decides whether the anchor-standard hook rewrite beats the original Feb 2026 opening on the Cursed avatar's advertorial destination before the same rewrite pattern is applied to other Exotic Monk speakers.

Spend contract: fresh CBO ad set in campaign 120237762972300539, AdLauncher canonical floor 1500 cents (shared contract default). Not a producer-selected override.

Production route: manual AdLauncher batch (no Trello): output/meta-uploads/tbf-exp105-wuwei-hook-a-jetsetters-20260912/ — mutation-free PRE-ENQUEUE package (manual_manifest.json + dry_run_request.json + dry_run_report.json; NO producer_package.json, NO enqueue_result.json, no queue custody, no Meta mutation). Script: products/tbf/scripts/exotic-monk/master-wu-wei-chen-20260912-a-jetsetters.txt (branch tbf/wu-wei-hook-rewrite-20260912).

Source: live ad 120237763128430539 in campaign 120237762972300539 (Control video 3083132932075391, read back 2026-09-12); canonical Cursed Sheet row 5 (A5:Z5) for headline and primary text.

Element tested: video hook only — the original Feb 2026 Master Wu Wei Chen hook versus a new anchor-standard hook rewrite of the same script (quiz close); same character, same headline, same primary text, same destination, same CTA. Quiz destination (Cursed Quiz).

Sibling: Exp 105 (same control + V1 hook, advertorial destination)

Status: Planned — prepared, dry-run PASS, not enqueued (PREPARED_NOT_ENQUEUED)

Family: wuwei-anchor-hook

Altitude: Rung 5 — hook (video opening)

Date: 2026-09-12

Avatar: The Cursed

Campaign ID: 120237762972300539

Campaign: 20260222 - TBF | Curses - Testing

Destination URL: https://discover.karmaitems.com/tbf-1308-crs-quiz-ev-001/

Destination shorthand: Cursed Quiz

Creative type: Control = existing Meta video ID 3083132932075391 (live ad 120237763128430539 "20260222 - TBF Exotic Monk Avatar Messages | Master Wu Wei - 17 Jetsetters (Curses)", creative 2113620142520798; reused as existing_media, no re-upload). V1 = new SoulGem authority_talking_head render, 1.0x, output_captioned.mp4, sha256 842d6c8a20edda50af8f3a68ea70da77eb81b84ff05591509a547f9524ea7abe, 228.04 s (run tbf-wuwei-b-quiz-20260912).

Ad set name: 20260912 - EXP - TBF - Exp 106 Wu Wei Hook - Cursed Quiz

Live ad set name: N/A — not launched; PREPARED_NOT_ENQUEUED

Ad set ID: N/A — not launched

Result: Pending — prepared 2026-09-12, not enqueued

Production authorization: Edward 2026-09-12 (session), verbatim: "Queue them alongside the original old control. Obviously get the destiantions right, including on the primary text and disable destination fan out". Batch built and dry-run PASS (mutation-free); enqueue is a separate step and has not been executed — no producer_package.json, no enqueue_result.json, no queue custody, no Meta mutation. Per fresh-pair doctrine the Control (the live original) also appears inside the fresh ad set, so the ad set carries exactly two ads: Control + V1.

Hypothesis: on the identical character, script body, headline, primary text, destination, and audience, the anchor-standard hook (names the curse-placer already in your life, then the slow drain) out-allocates the original Feb 2026 opening, because the Cursed avatar responds to a specific, personal cause named in the first seconds.

Prospect claim: The Cursed already suspects someone close is behind the run of bad luck; a hook that names that person and the slow leak earns the watch-through before the mechanism is explained.

Placement: problem-aware; the Cursed Quiz is this avatar's live secondary destination in the Curses - Testing campaign.

Novel variable: video hook only (Control = original Wu Wei video; V1 = new Wu Wei render with the anchor hook, quiz close).

Hold constant:

• Character Master Wu Wei Chen; Control video is the live original (Meta video 3083132932075391).
• Copy: canonical Cursed Sheet row 5 (A5:Z5) headline "Wear It 2 Weeks and Watch Your Luck Turn"; primary text = row 5 with the embedded CTA URL swapped to the quiz URL (thread_destination_override, allow_non_sheet_url, override reason quoting Edward's line).
• Destination https://discover.karmaitems.com/tbf-1308-crs-quiz-ev-001/ (byte-exact with trailing slash); CTA learn_more; both ads carry it.
• Avatar The Cursed, campaign 120237762972300539, template ad set 120247531189150539 targeting (age keys stripped), page 109598535381020, pixel 839353417201669.
• AdLauncher canonical CBO floor (1500 cents, shared contract default since 2026-09-12 06:02 UTC).
• Destination fan-out disabled with operator approval (fixed phrase, approved_by Edward, instruction = the verbatim line above).

Primary test variants:

• Control — 20260912 - EXP - TBF - Exp 106 Control Wu Wei Original Hook - Cursed Quiz

> existing Meta video 3083132932075391 (live ad 120237763128430539)

• V1 — 20260912 - EXP - TBF - Exp 106 V1 Wu Wei Anchor Hook - Cursed Quiz

> new render tbf-wuwei-b-quiz-20260912, sha256 842d6c8a20edda50af8f3a68ea70da77eb81b84ff05591509a547f9524ea7abe

Decision logic:

• Rank the two arms by Meta allocation (winner = highest spend, last day).
• Arms under $50 lifetime spend are Meta rejections, not rivals, never re-run.
• If both arms starve there is no mature comparison and one fresh-pair confirmation may follow.
• If V1 wins, the anchor hook becomes the default Wu Wei opening for the quiz destination; if Control wins, the original hook stays. Exp 105 decides the advertorial destination independently.

Delivery risk: fresh two-arm CBO ad set in the same campaign; the Control video is also live in its original ad set and in Exp 105, so early allocation can thin. Pre-registered NO-DELIVERY branches above. Never inject into the live original ad set.

Why it matters: decides whether the anchor-standard hook rewrite beats the original Feb 2026 opening on the Cursed avatar's quiz destination before the same rewrite pattern is applied to other Exotic Monk speakers.

Spend contract: fresh CBO ad set in campaign 120237762972300539, AdLauncher canonical floor 1500 cents (shared contract default). Not a producer-selected override.

Production route: manual AdLauncher batch (no Trello): output/meta-uploads/tbf-exp106-wuwei-hook-b-quiz-20260912/ — mutation-free PRE-ENQUEUE package (manual_manifest.json + dry_run_request.json + dry_run_report.json; NO producer_package.json, NO enqueue_result.json, no queue custody, no Meta mutation). Script: products/tbf/scripts/exotic-monk/master-wu-wei-chen-20260912-b-quiz.txt (branch tbf/wu-wei-hook-rewrite-20260912).

Source: live ad 120237763128430539 in campaign 120237762972300539 (Control video 3083132932075391, read back 2026-09-12); canonical Cursed Sheet row 5 (A5:Z5) for headline; primary text CTA URL swapped to the row's secondary (quiz) URL.

Element tested: Audio and visual production polish on the Champa Dorje authority talking-head — orchestral score bed (V1) and orchestral score bed plus ~20% realistic non-product B-roll cutaways (V2) versus the voiceover-only Control

Status: Complete

Result vs decision logic: Branch 1 (Delivered) fired: Control is the allocation winner among delivered arms ($65.98 lifetime), so voiceover-only — no orchestral bed — stays the default production treatment. Branch 2 (Partial no-delivery) fired: V1 ($38.29) and V2 ($14.12) both stayed under $50 lifetime and are never re-run. Branch 3 (Total no-delivery) not fired. Verdict: neither the orchestral score nor the score-plus-B-roll treatment earned a read; plain voiceover is the settled default for this monk authority talking-head.

Learnings:

• [family] Keep the Champa authority talking-head voiceover-only — the orchestral score bed and the score-plus-B-roll treatments never earned a read, so plain voiceover is the settled default for this format.
• [family] Do not add a motivational score or realistic cutaways to this monk authority format expecting a lift — the audience did not reward the extra production.
• Transfer: Cheapest test: repeat the score-versus-voiceover question on a non-authority surface (a dense-caption or UGC format) to learn whether the null is specific to the monk talking-head or production polish never moves this audience.

Family: champa-curse-qa-remake

Move: Production polish (Rung 8) — audio-treatment isolation on the Debt-Anxious champa-curse-qa remake; Edward-directed

Production authorization: Edward 2026-09-05 (chat): "run a junglelab experiment on this anchor testing if applying incredibly motivational orchestral movie [score] on a tibetan monk authority video will improve performance". The instruction fully specifies the lineup (scripts byte-identical; V1 differs only by the score bed), so it is recorded as the Checkpoint 1 clearance. Normal-cadence enqueue; no forced release. V1 review media hosted for Edward before release. V2 added at Edward's instruction 2026-09-05 (chat): "can we also ad some B-Roll, not of the product, realistic non-product B-Roll, should only cover about 20% of the video, distributed evenly. Let's make musc+B-Roll V2". Edward 2026-09-05 (chat): "the first b-roll cutaway has to start between 1second and 3 seoncd" — V2 recut accordingly (r2); the r1 cut (first cutaway at 15.8 s) was never released.

Date: 2026-09-05

Avatar: The Debt-Anxious (provisional)

Campaign ID: 120247651749480539

Campaign: 20260902 - TBF | Provisional Avatars - Testing

Destination shorthand: Champa Advertorial (Curses)

Destination URL: https://discover.karmaitems.com/tbf-1206-curses-advertorial-ev-001/

Ad set name: 20260905 - EXP - TBF - Exp 88 Orchestral Score - Champa Advertorial (Curses)

Live ad set name: 20260905 - P - EXP - TBF - Exp 88 Orchestral Score - Champa Advertorial (Curses)

Ad set ID: 120247712226430539

Live ad IDs:

• Control: 120247712234060539, 120247712235140539
• V1: 120247712235890539, 120247712237280539
• V2: 120247712238090539, 120247712239600539

AdLauncher request ID: produced-tbf-exp088-orchestral-score-20260905-r3-3b5fe258

AdLauncher queue state: live (publisher proof; register auto-synced)

AdLauncher request SHA-256: d359764e5cb4d1473c9dc8a227023720854d8e35c822d95a4d29c59292c606a3

Source observation: Swipe anchor champa-curse-qa (sw_b04ddeafb0; Authority UGC / Direct Pain Point monk Q&A) was remade for The Debt-Anxious through Kira Collab (adchain run tbf-debt-anxious-champa-curse-qa-remake-20260903) and rendered as SoulGem authority_talking_head (Champa Dorje, 576x704, 1.3x, burned captions, voiceover only — no music bed; SoulGem workdir soulgem/outputs/2026-09-03/champa-dorje-the-monk_162939). It went live 2026-09-05 as ad 120247701241660539 in the Remake Trio adset 120247701239190539 (same campaign). No allocation result exists yet, so this is not a decomposition of a proven winner: it tests whether the score improves this remake.

Hypothesis: A motivational orchestral film score under the monk's voiceover raises perceived production value and emotional stakes, so the identical message earns more allocation than the voiceover-only version.

Novel variable: Stacked production polish, two arms. V1 — presence of the orchestral motivational music bed (ElevenLabs Music /v1/music, prompt sha256 876ffeb8297f9a7b49e5dca544170e527d6d81078cd1257141a8554a126b4438, music.mp3 sha256 8b64980c11704353b74899e349337463e5c161bbf8bfaa935961394faa3299e6, mixed at 0.22 by soulgem.av_mux.mix_music, video stream copied). V2 — V1's exact audio (audio stream md5 604a8c34721a92af089c4eed47a52363, byte-identical to V1) plus seven 5-second realistic non-product B-roll cutaways (Hedra Kling V3 Standard T2V in SoulGem's realistic register: dim kitchen couple over bills, candlelit ritual table, wallet emptied on a kitchen table, stacked envelopes, late-night office, Tibetan plateau with prayer flags, sunrise doorway) hard-cut over the talking head at 2.6, 27.0, 51.5, 76.0, 100.5, 125.0 and 149.5 s on the 1.0x timeline (24.5 s apart), covering 19.9% of the runtime; the first cutaway opens at 2.0 s on the finished timeline, inside the hook; the CTA close stays on Champa; captions re-burned from the Control's own captions.ass and speed applied with the Control's own 1.3x function. V2 vs V1 isolates the B-roll; V1 vs Control isolates the score. Control carries neither.

Hold constant: Video stream byte-identical between Control and V1 (video_stream_md5 92ddca072078862cb4dd8360fdd8d15b; first frame md5 identical); V2 differs from that stream only inside its seven cutaway windows; script sha256 b09135589757147350d5187a102b748e8eb8c6bb132cad15b660d05cf2cb1b1e; Champa Dorje render and portrait (no re-render); voice AnUAVuhTZsESOSU2wluY; 576x704; 1.3x playback; burned neutral_box captions (caption sensecheck ok inherited with the identical video stream); duration 135.42 s; headline "It Could Be Bad Luck Around You, Not A Personal Failing"; frozen-angle primary text identical to the live Remake Trio; destination URL; CTA learn_more; campaign, template adset 120247694689160539 targeting, pixel 839353417201669, page 109598535381020. Parity evidence: output/adchain/tbf-exp088-orchestral-score-20260905/parity/parity_manifest.json, verifier verify.py (validation.json all_pass true). V2: PSNR versus Control 43.7 dB mean outside the cutaway windows and 13.1 dB inside (parity/v2/psnr_summary.json); cutaway plan output/adchain/tbf-exp088-orchestral-score-20260905/v2/cutaway_plan.json; V2 verifier verify_v2.py (validation_v2.json all_pass true).

Variant map:

• 20260905 - EXP - TBF - Exp 88 Control Orchestral Score - Champa Advertorial (Curses) — live Remake Trio video verbatim (sha256 21dfb2672fe371f7cfa69cacdbda7d79736964efe3583474124a57d1551acc77), fresh ad ID in the fresh adset, no re-render.
• 20260905 - EXP - TBF - Exp 88 V1 Orchestral Score - Champa Advertorial (Curses) — same bytes plus the orchestral score bed (sha256 29700c76b31ca38fa406f542111eccd9ecb9ae4af9d2ca00bd0468c0cb0ad1f7).
• 20260905 - EXP - TBF - Exp 88 V2 Orchestral Score - Champa Advertorial (Curses) — V1 audio byte-identical plus seven realistic non-product B-roll cutaways, 19.9% coverage (sha256 2dacea5b3920d51663fd7f35cea2e98b25e9837e3b56fedf208cd1d05d322e72) (r2 recut; first cutaway at 2.0 s).

Decision logic:

• Delivered: The allocation winner among delivered arms is the verdict. V2 wins — score plus B-roll cutaways become the default production treatment for this format (and V2 over V1 says the cutaways add on top of the score). V1 wins — the score alone is the treatment; cutaways do not add. Control wins — voiceover-only, no bed, stays the default.
• Partial no-delivery: Any arm under $50 lifetime spend is a Meta rejection, never re-run; the allocation winner among delivered arms is the verdict.
• Total no-delivery: If all three arms starve under $50 lifetime spend, record an early signal only and require one fresh-pair confirmation before changing doctrine.

Delivery risk: Fresh three-arm CBO adset in the ACTIVE Provisional Avatars Testing campaign. The arms are near-siblings (audio bed; audio bed plus 20% cutaways) — TBF 013 starvation shape; the partial-no-delivery branch above is the pre-registered verdict. The Control creative also runs live in the Remake Trio adset (120247701239190539) in the same campaign; that is a same-campaign sibling competing for CBO budget, recorded as external context, not a within-adset confound. AdLauncher fixed CBO floor: 2000 cents.

Transferable rule: If V2 wins: [style: authority_talking_head] TBF monk authority talking-heads carry a motivational orchestral score bed at 0.22 and ~20% evenly spaced realistic non-product B-roll cutaways by default. If V1 wins: [style: authority_talking_head] the score bed is the default; cutaways are not. If Control wins: [style: authority_talking_head] voiceover-only stays the default; neither treatment buys allocation. Lane note: Lane B isolation on an unproven remake at Edward's direction; first read on The Debt-Anxious, rule scoped to the format.

Production route: AdChain experiment mode — Control reuses the live SoulGem final; V1 is the canonical mix_music over the same bytes with one ElevenLabs Music bed (no re-render of Champa; V2 adds seven Hedra Kling T2V B-roll clips over the same footage). One three-media AdLauncher request (re-enqueued twice 2026-09-05: first when V2 was added, then when V2 was recut for the 1–3 s first-cutaway rule), normal release cadence, no forced release. Review media: https://tools.ambercollective.co/media/soulgem/tbf-exp088-orchestral-score/control.mp4 · https://tools.ambercollective.co/media/soulgem/tbf-exp088-orchestral-score/v1_orchestral_score.mp4 · https://tools.ambercollective.co/media/soulgem/tbf-exp088-orchestral-score/v2_music_broll.mp4 (cutaway contact sheet: .../contact_sheet.png)

Result: Control — top spend $6.42 on 2026-09-11 (latest day). Total adset spend $118.39 over 5 days. Ranking: Control $6.42 latest / $65.98 total; V1 $3.00 latest / $38.29 total NO-DELIVERY (<$50 lifetime); V2 $1.88 latest / $14.12 total NO-DELIVERY (<$50 lifetime). Auto-called per 4d/$50 doctrine.

Element tested: Complete scored B-roll production treatment versus plain authority presenter on the same saved render.

Status: Winner called (auto 2026-09-12) — automation blocked: creative source unavailable

Result vs decision logic: AUTOMATION BLOCKED — creative source unavailable after recovery ladder; retry automatically on the next run.

Ad set name: 20260906 - EXP - TBF - Exp 91 Scored Broll Mode - Champa Advertorial (Curses)

Live ad set name: 20260906 - P - EXP - TBF - Exp 91 Scored Broll Mode - Champa Advertorial (Curses)

Ad set ID: 120247733418800539

Live ad IDs:

• Control: 120247733419940539, 120247733420500539
• V1: 120247733421090539, 120247733421620539

Variant map:

• 20260906 - EXP - TBF - Exp 91 Control Scored Broll Mode - Champa Advertorial (Curses) — same-render plain presenter, voiceover only, corrected captions, 1.3x.
• 20260906 - EXP - TBF - Exp 91 V1 Scored Broll Mode - Champa Advertorial (Curses) — identical presenter and narration, seven realistic non-product cutaways (20.4%, first at 2.0 seconds), orchestral score at 0.22, identical corrected captions and 1.3x.

Delivery risk: Fresh Control + V1 pair in a new adset, never inserted into the incumbent. Any arm under $50 lifetime spend is a Meta rejection. If at least one arm delivers, the allocation winner is the verdict and rejected arms are never re-run. Only if all arms starve is this an early signal requiring one fresh-pair confirmation.

Date: 2026-09-06

Avatar: The Debt-Anxious (provisional)

Campaign ID: 120247651749480539

Campaign: 20260902 - TBF | Provisional Avatars - Testing

Destination shorthand: Champa Advertorial (Curses)

Destination URL: https://discover.karmaitems.com/tbf-1206-curses-advertorial-ev-001/

Family: champa-curse-qa-remake

Source observation: Follow-up production-mode comparison to TBF Exp 88; no winner is inferred from that active experiment. The new scored-mode sample is accepted for queue after caption repair.

Production authorization: Edward 2026-09-06: "can you fix the sample? and queue it"; "let's queue as an experiment vs original". The existing script and rendered sample are approved; no copy or paid-generation changes. Control interpretation: same-render plain presenter, stated in session, to isolate the treatment from avatar/voice take differences.

Hypothesis: The scored B-roll treatment earns stronger allocation than the same authority performance presented as voiceover-only talking head.

Novel variable: Scored B-roll mode as one bundled production treatment. This comparison does not separately attribute music versus B-roll.

Hold constant: Script sha256 b09135589757147350d5187a102b748e8eb8c6bb132cad15b660d05cf2cb1b1e; same saved Champa Dorje raw video, voice and narration; exact corrected caption files; 576x704, 24fps, 1.3x; same copy/headline/CTA/URL, campaign, live template 120247701239190539, targeting, pixel and page.

Decision logic: If V1 delivers and wins allocation, retain scored B-roll for this execution. If Control delivers and wins, retain plain authority. If the winner delivers and the rival starves, the delivered winner stands; do not re-run the rejected arm. If both deliver without a clear allocation winner, record inconclusive rather than invent a win. If all arms starve, allow one fresh-pair confirmation. Report spend, purchases and ROAS alongside allocation; no broader mechanism claim from this two-arm treatment test.

Spend authority: AdLauncher fixed CBO floor: 2000 cents.

Production route: Caption-only rebuild from saved media via canonical SoulGem caption, speed, music-mix and finalization functions. No provider regeneration. Run: output/soulgem-scored-broll-captionfix-20260906. AdLauncher package: output/meta-uploads/tbf-scored-broll-mode-20260906. Normal queued release cadence; no forced release.

AdLauncher request ID: produced-tbf-scored-broll-mode-20260906-4d67cd84

AdLauncher queue state: live (publisher proof; register auto-synced)

AdLauncher request SHA-256: 8a2f76a4535edadd6e97fbd8bb6ac371b7ae8ecafb3c027db0f7808c401c177f

Production QA: Full normalized caption sequence equals the approved script for both arms; identical ASS, transcript timing and narration before scoring; existing music/audio unchanged by caption repair; both files fully decode, faststart, 576x704 at 24fps and 132.083333 seconds. Source/staged/queue SHA-256 parity verified. Targeting follows the live Debt-Anxious template with AdLauncher canonical age-key removal for both arms.

Review media: Control https://tools.ambercollective.co/share/tbf-exp091-control-20260906.mp4 ; V1 https://tools.ambercollective.co/share/tbf-exp091-scored-broll-20260906.mp4

Queue evidence: output/meta-uploads/tbf-scored-broll-mode-20260906/verifier_output.json

Result: V1 — top spend $17.24 on 2026-09-11 (latest day). Total adset spend $94.72 over 4 days. Ranking: V1 $17.24 latest / $71.19 total; Control $2.01 latest / $23.53 total NO-DELIVERY (<$50 lifetime). Auto-called per 4d/$50 doctrine.

Element tested: Opening structure on the champa-curse-qa Authority UGC anchor — the original curse question/explanation versus a symptom stack that names common lived patterns, their quality-of-life cost, the jealousy diagnosis, and the copper solution before rejoining the exact original body.

Status: Complete

Result vs decision logic: Partial no-delivery fired: Control (question-first curse reveal) delivered while V1 (symptom-stack opening) stayed under the threshold, so Control is the verdict and V1 is a Meta rejection that is never re-run. Preserve the original question-first curse reveal for this anchor.

Learnings:

• [family] Preserve the question-first curse reveal for the Champa curse-question anchor — opening with the direct curse question and the ordinary-jealousy-not-black-magic reframe beat the symptom-stack opener, so keep that grammar when remaking this anchor.
• Transfer: Cheapest test: carry the question-first curse reveal to a second Champa anchor that currently opens with a direct promise, and test whether the question-first grammar beats the direct-promise open there too.

Family: authority-symptom-stack-hook

Parent: Exp 32

Move: Decompose a proven anchor — opening pattern

Production authorization: Edward approved both the exact V1 copy and production on 2026-09-05 (chat: "run both as seperate experiments against the original"); normal-cadence enqueue only, no forced release.

Date: 2026-09-05

Avatar: The Cursed

Campaign ID: 120237762972300539

Campaign: 20260222 - TBF | Curses - Testing

Destination shorthand: 17 Jetsetters (Cursed)

Destination URL: https://discover.karmaitems.com/tbf-1206-curses-advertorial-ev-001

Ad set name: 20260905 - EXP - TBF - Exp 86 Curse QA Symptom Stack Hook - 17 Jetsetters (Cursed)

Live ad set name: 20260905 - P - EXP - TBF - Exp 86 Curse QA Symptom Stack Hook - 17 Jetsetters (Cursed)

Ad set ID: 120247708399740539

Live ad IDs:

• Control: 120247708400860539
• V1: 120247708402090539

AdLauncher request ID: produced-tbf-exp086-curse-qa-symptom-stack-20260905-64995efe

AdLauncher queue state: live (publisher proof; register auto-synced)

AdLauncher request SHA-256: 892c38a51aaa81d77f52e202ee4f76db681aaca91b22ca90a2d924091dae5767

Source Trello card ID: chat-20260905-curse-qa-symptom-stack

Source observation: The approved SwipeSmith anchor champa-curse-qa (swipe sw_b04ddeafb0, sample ss_a7e1e4ce35, Meta ad 120238092100210539) is a 443-word Champa Dorje Authority UGC historical winner. Its opening asks whether the viewer has unknowingly been cursed, explains that the real curse is ordinary jealousy rather than Hollywood black magic, then rejoins at "When someone looks at you with jealousy..." The GetHookd reference first stacks familiar behaviours and fears, names how they reduce quality of life, then resolves with "And using hypnotherapy, we can change that." Edward selected that structure for this anchor.

Hypothesis: A symptom-stack opening will beat the original curse-question opening on allocation because it lets the viewer recognize several concrete patterns and their emotional cost before Champa names the jealousy mechanism and solution.

Novel variable: The contiguous opening from word 1 through "Here's how it works." Control keeps the source opening byte-for-byte. V1 replaces only that opening and rejoins byte-for-byte at "When someone looks at you with jealousy..." The 72-word increase and the equal 72-word delay to the product introduction are declared consequences of the tested opening, not hidden variables.

Hold constant: Every byte after the rejoin marker; Champa Dorje character; pinned Prayer Hall portrait and avatar prompt; voice AnUAVuhTZsESOSU2wluY; SoulGem authority_talking_head; 540p provider tier; neutral boxed captions; 1.3x playback; canonical Sheet headline, primary text, destination, and CTA; campaign, targeting, placements, and $20 CBO floor. Both arms are rendered fresh as a matched pair because the current SoulGem take architecture and dimensions differ from the historical anchor file.

Variant map:

• 20260905 - EXP - TBF - Exp 86 Control Curse QA Symptom Stack Hook - 17 Jetsetters (Cursed) — original 443-word anchor transcript byte-for-byte; script sha256 6e632ab84e42f693d49f86afa5329ad66166792b0193461a8e1dcf740e767ba6.
• 20260905 - EXP - TBF - Exp 86 V1 Curse QA Symptom Stack Hook - 17 Jetsetters (Cursed) — approved 515-word symptom-stack opening, exact original suffix; script sha256 dda0b2b7560554190aaa6c429dc584f040a845da361a5cab6eeab9f6d0f7b694.

Swap spec:

• Control opening ends: "Here's how it works."
• V1 sequence: four lived-pattern symptoms → peace/confidence/belief cost → jealousy-and-resentment diagnosis → "And using copper blessed with protection mantras, we can change that" → Champa identity and curse explanation.
• Exact rejoin: "When someone looks at you with jealousy, resentment, or ill will..."

Decision logic:

• Delivered: If both arms deliver, latest-day allocation leader after the canonical 4d/$50 floor is the verdict. V1 wins — adopt symptom-stack → cost → diagnosis → solution as the default opening pattern for Champa curse-question remakes. Control wins — preserve the original question-first curse reveal.
• Partial no-delivery: If only one arm delivers, the delivered arm is the verdict and the starved arm is a Meta rejection that is never re-run.
• Total no-delivery: If both arms starve, record an early signal only and require one fresh-pair confirmation.

Delivery risk: Fresh two-arm CBO adset; the challenger never enters the historical winner's adset. AdLauncher fixed CBO floor: 2000 cents. Destination fanout disabled so this remains exactly two ads.

Why it matters: This isolates whether the transferable asset in the GetHookd reference is its recognition-first opening architecture, rather than the hypnotherapy category. A V1 win gives Champa Authority UGC a reusable hook grammar; a Control win keeps the original high-curiosity question grammar.

Copy integrity: output/adchain/tbf-authority-symptom-stack-hooks-20260905/exp086-curse-qa/validation.json (ok=true); canonical Authority UGC verifier passes both arms.

Production route: AdChain experiment mode — fresh matched SoulGem pair — fresh Sheet/product/congruence gates — one two-media AdLauncher request — normal release cadence. No forced release.

Production QA: Both final files have exact approved captions and pass the matched-pair gate with pinned still sha256 21aecb423fa0b43fe7fdddbd244a8ddb8dfeca7c7b49e594b8ba5e71f63bf379, avatar-prompt sha256 a970e5dd0e0677a71ae3d91d92fb27883d67ebba7ce2353a382d64a8958f3bca, and take-motion-plan sha256 15ee056e73beb08fd695b6f497f09f633015f4c3435cdb1c9121a69859860dee shared byte-for-byte across arms. Control final sha256 f260723b9f10636bdfa6750ab733efd07e3710801d097cbd3c71a8bd070e74b0; V1 final sha256 d659e3985e4455d03e258ccf3da4e009ac558006b25a1e3ca5ba45efc9a2775c. An initial V1 render with a nonmatching generated avatar prompt was quarantined and never packaged; the corrected V1 is the only challenger in the launch request.

Result: Control — top spend $17.78 on 2026-09-09 (latest day). Total adset spend $106.89 over 4 days. Ranking: Control $17.78 latest / $68.82 total; V1 $9.35 latest / $38.07 total NO-DELIVERY (<$50 lifetime). Auto-called per 4d/$50 doctrine.

Element tested: Opening structure on the champa-wealth-curse Authority UGC anchor — the original direct wealth-curse promise and symptom setup versus a symptom stack that names money instability, its emotional cost, the envy diagnosis, and the copper solution before rejoining the exact original body.

Status: Complete

Result vs decision logic: Partial no-delivery fired: V1 (symptom-stack opening) delivered while Control (direct wealth-curse promise) stayed under the threshold, so V1 is the verdict and Control is a Meta rejection that is never re-run. Adopt the symptom-stack, cost, diagnosis, solution opening as the default for long Champa wealth-curse scripts.

Learnings:

• [family] For the long Champa wealth-curse script, open with the symptom stack — name the repeated money setbacks and their emotional cost, then the envy diagnosis and the copper solution — instead of the direct wealth-curse promise; the recognition-first opening earned the run.
• Transfer: Cheapest test: apply the symptom-stack opening to another long Champa script to test whether the recognition-first grammar is the default for long scripts, not just the wealth-curse anchor.

Family: authority-symptom-stack-hook

Parent: Exp 14

Move: Decompose a proven anchor — opening pattern

Production authorization: Edward approved both the exact V1 copy and production on 2026-09-05 (chat: "run both as seperate experiments against the original"); normal-cadence enqueue only, no forced release.

Date: 2026-09-05

Avatar: The Cursed

Campaign ID: 120237762972300539

Campaign: 20260222 - TBF | Curses - Testing

Destination shorthand: 17 Jetsetters (Cursed)

Destination URL: https://discover.karmaitems.com/tbf-1206-curses-advertorial-ev-001

Ad set name: 20260905 - EXP - TBF - Exp 87 Wealth Curse Symptom Stack Hook - 17 Jetsetters (Cursed)

Live ad set name: 20260905 - P - EXP - TBF - Exp 87 Wealth Curse Symptom Stack Hook - 17 Jetsetters (Cursed)

Ad set ID: 120247708685840539

Live ad IDs:

• Control: 120247708686310539
• V1: 120247708686540539

AdLauncher request ID: produced-tbf-exp087-wealth-curse-symptom-stack-20260905-f7aeacfb

AdLauncher queue state: live (publisher proof; register auto-synced)

AdLauncher request SHA-256: ced0258737136a5d4392e0e8ad490613fd707f6dfc6f1093cb031deb3da5c7fa

Source Trello card ID: chat-20260905-wealth-curse-symptom-stack

Source observation: The approved SwipeSmith anchor champa-wealth-curse (swipe sw_1d49e8f3b2, sample ss_7ecf0f2ec1, Meta ad 120242208504860539) is a 767-word Champa Dorje Authority UGC winner with $5,529.90 observed 90-day spend. Its opening promises that a wealth curse can move again in two weeks, identifies Champa, and describes money/peace instability before rejoining at "Sometimes it begins when someone looks at you with envy..." The GetHookd reference first stacks familiar behaviours and fears, names how they reduce quality of life, then resolves with "And using hypnotherapy, we can change that." Edward selected that structure for this anchor.

Hypothesis: A symptom-stack opening will beat the original direct wealth-curse promise on allocation because recognition of repeated money setbacks and lost security earns attention before the spiritual diagnosis and solution are introduced.

Novel variable: The contiguous opening from word 1 through "not every curse looks scary at first." Control keeps the source opening byte-for-byte. V1 replaces only that opening and rejoins byte-for-byte at "Sometimes it begins when someone looks at you with envy..." The 9-word increase and equal 9-word delay to the product introduction are declared consequences of the tested opening.

Hold constant: Every byte after the rejoin marker; Champa Dorje character; pinned Prayer Hall portrait and avatar prompt; voice AnUAVuhTZsESOSU2wluY; SoulGem authority_talking_head; 540p provider tier; neutral boxed captions; 1.3x playback; canonical Sheet headline, primary text, destination, and CTA; campaign, targeting, placements, and $20 CBO floor. Both arms are rendered fresh as a matched pair because the current SoulGem take architecture and dimensions differ from the historical anchor file.

Variant map:

• 20260905 - EXP - TBF - Exp 87 Control Wealth Curse Symptom Stack Hook - 17 Jetsetters (Cursed) — original 767-word anchor transcript byte-for-byte; script sha256 a7e1e90e2156151b3fec382ff0e23cbb31317171fc70e30880c7f7474a44da84.
• 20260905 - EXP - TBF - Exp 87 V1 Wealth Curse Symptom Stack Hook - 17 Jetsetters (Cursed) — approved 776-word symptom-stack opening, exact original suffix; script sha256 09d6c257f530168d23d9f81e7017b99c89b049ee182fc34a93fade05c279a827.

Swap spec:

• Control opening ends: "not every curse looks scary at first."
• V1 sequence: four money-instability symptoms → confidence/hope/future cost → envy/root-chakra diagnosis → "And using copper blessed with protection mantras, we can change that" → Champa identity and generational warning.
• Exact rejoin: "Sometimes it begins when someone looks at you with envy, with resentment..."

Decision logic:

• Delivered: If both arms deliver, latest-day allocation leader after the canonical 4d/$50 floor is the verdict. V1 wins — adopt symptom-stack → cost → diagnosis → solution as the default opening pattern for long Champa wealth-curse scripts. Control wins — preserve the direct promise and immediate wealth-curse framing.
• Partial no-delivery: If only one arm delivers, the delivered arm is the verdict and the starved arm is a Meta rejection that is never re-run.
• Total no-delivery: If both arms starve, record an early signal only and require one fresh-pair confirmation.

Delivery risk: Fresh two-arm CBO adset; the challenger never enters the historical winner's adset. AdLauncher fixed CBO floor: 2000 cents. Destination fanout disabled so this remains exactly two ads.

Why it matters: This is the higher-confidence replication of Exp 86 on the longer current-spend anchor. A V1 win across both experiments promotes the pattern to an Authority UGC family rule; a split result keeps it anchor-specific.

Copy integrity: output/adchain/tbf-authority-symptom-stack-hooks-20260905/exp087-wealth-curse/validation.json (ok=true). The byte-exact source and V1 both fail only the generic standalone-I heuristic because the original says "My name is Champa Dorje"; all other canonical checks pass, so the exact shared taxonomy mismatch is narrowly reconciled.

Production route: AdChain experiment mode — fresh matched SoulGem pair — fresh Sheet/product/congruence gates — one two-media AdLauncher request — normal release cadence. No forced release.

Production QA: Both final files have exact approved captions and pass the matched-pair gate with pinned still sha256 21aecb423fa0b43fe7fdddbd244a8ddb8dfeca7c7b49e594b8ba5e71f63bf379, avatar-prompt sha256 2c273425928c4cec8d198d24481e2346e630b01ab7e34d02deba354bc0a3df30, and take-motion-plan sha256 52067292477f6e6caa10eb7bf1f087d9d2bc2bc2ff03d49b172c49397e3cacaf shared byte-for-byte across arms. Control final sha256 c1740bd2b88150d92105fe80ff5bf551f94d56f94846f06d2e3ba4db7924d914; corrected V1 final sha256 849f0ab23511630c509b827817f41682a18d5364441694deb0efa1e70fc8ea17. The first V1 visual was quarantined because its generated avatar prompt differed from Control. A subsequent 63-chunk splitter fallback was stopped before video generation. The shipped V1 reuses its already-approved 12 narration-audio chunks, regenerates the visual against Control's exact prompt and motion plan, and restores the verified caption timing locally; neither quarantined attempt was packaged.

Result: V1 — top spend $13.51 on 2026-09-09 (latest day). Total adset spend $91.31 over 4 days. Ranking: V1 $13.51 latest / $73.82 total; Control $8.48 latest / $17.49 total NO-DELIVERY (<$50 lifetime). Auto-called per 4d/$50 doctrine.

Element tested: Hook type and narrator voice

Status: Complete

Family: monk-trial

Result: Mystical Knowledge hook with 1st person monk voice scaled hardest ($600 spend, 2.37x ROAS). Direct Pain Point hook with 1st person monk voice also converted well ($202 spend, 2.58x ROAS) but was manually turned off during the test period. Forbidden Truth hooks and 3rd person narration were Meta-rejected regardless of scriptwriter.

Source: Intra-adset analysis (2026-06-09) — Ad set: "20260519 - B - TBF Monk Trial Task - 17 Jetsetters (Curses)" (7 script ads, $870 total 90-day spend)

Ad set name: 20260519 - B - TBF Monk Trial Task - 17 Jetsetters (Curses)

Test setup: Same ad set, same Jetsetters (Curses) audience. All ads share the same "curse on root chakra → copper bracelet breaks it" thesis and Champa Dorje monk character. What varies: hook type (Mystical Knowledge vs Direct Pain Point vs Forbidden Truth), narrator voice (1st person monk vs 3rd person narrator), and scriptwriter (Mark, Richard, Jose Rafael, Nicole). Note: Richard's ads were manually turned off at one stage during the test, so his scaled ad's spend is artificially capped — it was performing when live.

Variant (scaled, best): Mark Hook 2 — Mystical Knowledge hook ("The strange reason women over 50 get unlucky"), 1st person monk voice, 410 words. $600 spend, 2.37x ROAS. Still running.

Variant (scaled, performing but turned off): Richard Hook 2 — Direct Pain Point hook ("If your money keeps disappearing no matter what you do"), 1st person monk voice, 767 words. $202 spend, 2.58x ROAS. Was converting well but manually turned off during the test — spend is artificially capped, not a Meta rejection.

Variant (Meta rejected): 5 ads ($6–$29 spend range). Includes two Forbidden Truth hooks (Jose Rafael, Nicole), one 3rd person Direct Pain Point (Richard without hook intro), one 1st person Direct Pain Point (Mark Hook 1), and one 1st person Forbidden Truth (Jose Rafael Hook 1 — got a lucky purchase on $9 spend, not a real signal).

Hypothesis: Which hook type best carries the curse thesis for Jetsetters? Mystical Knowledge (curiosity + specificity), Direct Pain Point (problem-first), or Forbidden Truth (secrecy + revelation) — and does 1st person monk narration outperform 3rd person narrator?

Learnings:

• [avatar: Cursed] Lead with a specific, curiosity-driven claim ("the strange reason women over 50 get unlucky") rather than a generic forbidden-secret opener ("most people think a curse looks like something dramatic") — Jetsetters respond to specificity and mystery, not vague secrecy
• [family] First person monk voice consistently outperformed 3rd person narration — the same body script (Richard vs Richard Hook 2) only scaled when it had a 1st person monk hook bolted on
• [family] Shorter scripts (410 words) can outperform longer ones (767 words) when the hook does the heavy lifting — Mark Hook 2 was the shortest script and the top performer
• [avatar: Cursed] Forbidden Truth hooks were rejected across all scriptwriters for this audience — the "most spiritual teachers won't say this" / "most people think a curse looks dramatic" framing doesn't create enough specific curiosity for Jetsetters

Element tested: Headline curiosity structure (static ad)

Status: Complete

Family: monk-trial

Parent: Exp 003

Ad set name: 20260613 - B - TBF Statics | Mystical Knowledge Headline Structure - 17 Jetsetters (Curses)

Result: V2 ("What most women over 50 don't realize about getting unlucky") won by far by meaningful spend. V2 received $399.89 90-day spend; all other variants were under $50 and are treated as noise / no meaningful delivery. The prior "Control runner-up" line from early logging is NOT confirmed by 90-day data — Control (V1) had only $7.92 spend. Ranking by 90-day spend: V2 >> V6 ($23.13) > V5 ($14.73) > V3 ($9.39) > V1 ($7.92) > V4 ($6.40).

Date: 2026-06-09 (planned) · result logged 2026-06-23 · rechecked 2026-07-07

Avatar: The Cursed (Jetsetters)

Creative type: Static

Source: Follow-up to exp 003 — isolating which component of "the strange reason women over 50 get unlucky" carries the win, starting with the curiosity structure. Supabase ads_data_master rechecked 2026-07-07 — Ad set: "20260613 - B - TBF Statics | Mystical Knowledge Headline Structure - 17 Jetsetters (Curses)" (6 ads, ~$461.46 total 90-day spend).

Hypothesis: Exp 003 winner Mark Hook 2 stacked three things in its hook: curiosity structure ("the strange reason"), demographic anchor ("women over 50"), and negative outcome verb ("get unlucky"). Holding demographic anchor and outcome verb constant, we vary the curiosity structure to determine whether "the strange reason" specifically is the asset, or whether other curiosity structures perform equally.

Novel variable: Headline curiosity structure (the syntactic frame of the headline)

Hold constant:

• Demographic anchor: "women over 50"
• Negative outcome: "get unlucky" (and natural inflections of it)
• Static image, layout, CTA, brand mark
• Jetsetters (Curses) audience
• Destination page

Variants:

• Control (V1) — The strange reason women over 50 get unlucky — ad 120244462944860539
• V2 — What most women over 50 don't realize about getting unlucky — ad 120244462943940539
• V3 — The hidden pattern behind women over 50 getting unlucky — ad 120244462942570539
• V4 — Why women over 50 keep getting unlucky — ad 120244462941830539
• V5 — The one thing women over 50 miss about getting unlucky — ad 120244462881230539
• V6 — How women over 50 quietly get unlucky — ad 120244462879300539

Decision logic:

• If Control wins clean → "the strange reason" structure is the asset; brief next batch around it
• If variants 2/3/5 (specificity-of-missed-knowledge frames) all beat Control → the asset is "the audience doesn't know this," not the specific phrase
• If variant 4 ("why X keeps") wins → ongoing-state framing matters more than novelty framing
• If variants perform within noise → structure is not the asset; isolate demographic anchor or outcome verb next (exps 006, 007)

Why it matters: Determines whether the next batch of Mystical Knowledge static headlines should lock the "strange reason" phrase or whether we can substitute in other curiosity frames without losing the win. Result: substitute — but lock the personal-ignorance frame ("don't realize"), not arbitrary curiosity structures.

Result vs decision logic: No pre-registered branch fired cleanly. V2 beat Control (rules out "Control wins clean"). But V3 and V5 — the other two missed-knowledge frames grouped with V2 in the second branch — were equal LAST in meaningful spend, not beating Control, so the family-level hypothesis ("2/3/5 all beat Control → generic missed-knowledge is the asset") is falsified. V4 was last (rules out "ongoing-state wins"). And V2 won "by far," so it is not within-noise. The 90-day recheck confirms all non-V2 variants are under $50 / no meaningful delivery; the earlier "Control runner-up" note is not supported by mature data. The win is specific to V2's framing, not to the missed-knowledge family.

Learnings:

• [family] The asset is NOT the curiosity structure generically, and NOT the "strange reason" phrase specifically — Control (strange reason) was beaten into runner-up only in early logging; 90-day data puts Control at $7.92, i.e. noise.
• [family] The asset is NOT "missed-knowledge" as a category — V3 ("hidden pattern") and V5 ("the one thing... miss"), same family, were equal last.
• [avatar] What is unique to V2: it attributes the knowledge gap to the READER's personal, active ignorance ("you don't realize") rather than describing a mysterious external object ("the strange reason" / "the hidden pattern") or a weaker passive ("miss"). Personal active-ignorance framing beats both object-centric mystery and generic missed-knowledge.
• [family] Implication for next batch: lock the "don't realize" personal-ignorance frame as the new control and test refinements of it (e.g. "don't realize" vs "have no idea" vs "don't know") rather than substituting alternative curiosity structures.

Per-variant numbers: Rechecked 2026-07-07. 90-day spend / revenue: V2 $399.89 / $290.86 (~0.73x ROAS); V6 $23.13 / $92.90 (~4.02x ROAS, under-$50 noise); V5 $14.73 / $0; V3 $9.39 / $0; V1 $7.92 / $95.40 (~12.05x ROAS, under-$50 noise); V4 $6.40 / $0. Total ~$461.46. V2 is the only variant with meaningful delivery.

Element tested: Copy compression / message length (extractive, complete-sentence deletion only)

Status: Complete

Family: monk-trial

Parent: Exp 003

Result vs decision logic: The V1-450w-wins-cleanly branch fired: the 450-word compressed version ($62.72) was the only arm that crossed $50; the 767-word Control ($26.74) and 250-word V2 ($30.03) were both NO-DELIVERY. Under the amended doctrine the starved arm is a Meta rejection, not an unmeasured rival — the allocation winner IS the frame verdict. 450 words is the confirmed message length for this script family — the 767-word depth is not load-bearing, and the 250-word cut was rejected by the auction, establishing a floor.

Learnings:

• [family] 450 words is the confirmed compression target for the Richard/Champa message family. The 767-word depth adds no delivery value; the 250-word cut loses it.
• [family] Two starved arms bracket the winner: too long (767w) rejected, too short (250w) rejected. The auction mapped the viable length band in one experiment.
• Transfer: Applies to message-compression decisions on other long scripts in this account. Cheapest test: compress the next over-long script to the 450w band and compare allocation against its own history.

Altitude: Rung 7 — copy syntax; polish-altitude test; unmapped angle space exists.

Ad set name: 20260710 - EXP - TBF - Exp 14 Richard Champa Compression - Priestess Evolver

Ad set ID: 120246462443100539

Account ID: act_4092631590966722

Campaign ID: 120237762972300539

Surface: facebook_feed_video

Live ad IDs:

• Control: 120246465326610539
• V1: 120246465334180539
• V2: 120246465342150539

Winner adset ID: 120246462443100539

Winner arm: V1

Winner ad ID: 120246465334180539

Winner spend: 62.82

Winner complete through: 2026-08-26

Winner compared arm IDs: Control, V1, V2

Winner creative SHA-256: 687748d198151cf8bc1e1860121e577d501ae5d1fe9f5e6c2040611269cf048c

Winner media SHA-256: c4a01824c0bc105f3a23583f3dbb21e4a16fd074fdabb38fda6829882ad98af4

Winner media path: output/junglelab-experiment-maker-backfill/tbf-exp-014/winner-v1.mp4

Winner copy SHA-256: 9086c8d388210ac555fa41bd22e7c36f8a203854dd38b51118c7d5ad1e3aa222

Winner copy path: output/junglelab-experiment-maker-backfill/tbf-exp-014/winner-v1.txt

Winner creative format: video

Winner production style: First Person Contemporary VSL

Winner live delivery SHA-256: e86d0a5ae5cd726949fdd6ea95f1c03c43791727397efeed676e37c21d129124

Live ad set name: 20260710 - EXP - TBF Exp 014 - Priestess Evolver

Variant map:

• 20260710 - EXP - TBF - Exp 14 Control Money Hook Trim - Priestess Evolver
• 20260710 - EXP - TBF - Exp 14 V1 450w Compression Money Hook Trim - Priestess Evolver
• 20260710 - EXP - TBF - Exp 14 V2 250w Compression Money Hook Trim - Priestess Evolver

Trello: https://trello.com/c/2CHwGAPj/355-20260710-tbf-exp-014-richard-champa-message-compression

Scripts: https://docs.google.com/document/d/149qALz-pHxd9R6T6Rc6WbdSCsVGRyJMSvRl4PdX_RwE/edit?usp=drivesdk

Live destination: https://discover.karmaitems.com/tbf-2005-crs-priestess-advertorial-evolver-001/

Campaign: 120237762972300539 20260222 - TBF | Curses - Testing

Template adset: 120242455309210539 CORE - B - Priestess Evolver

Live adset: 120246462443100539 20260710 - EXP - TBF - Exp 14 Richard Champa Compression - Priestess Evolver — ACTIVE

Ads (historical launch, now PAUSED after Money Hook Trim cutover):

• Control — 120246462445630539 — PAUSED
• V1 450w Compression — 120246462449750539 — PAUSED
• V2 250w Compression — 120246462456350539 — PAUSED

Budget contract: CBO adset daily min spend target 2000 cents ($20); no adset budget; parent campaign daily budget 58300 cents ($583). Repaired from erroneous publish.json floor of 3000 cents ($30).

Batch artifacts: /root/.hermes/output/meta-uploads/20260710-tbf-exp014-priestess-compression/

Repair provenance: /root/.hermes/output/meta-uploads/20260710-tbf-exp014-priestess-compression/repair-20/

Launch status: Live / ACTIVE — 2026-07-10; Money Hook Trim cutover completed 2026-07-10

Result: V1 — top spend $0.11 on 2026-07-18 (latest day). Total adset spend $119.49 over 6 days. Ranking: V1 $0.11 latest / $62.72 total; V2 $0.00 latest / $30.03 total NO-DELIVERY (<$50 lifetime); Control $0.00 latest / $26.74 total NO-DELIVERY (<$50 lifetime). Auto-called per 4d/$50 doctrine.

Operator destination override:

• Sheet destination: https://discover.karmaitems.com/tbf-1206-curses-advertorial-ev-001
• Override destination: https://discover.karmaitems.com/tbf-2005-crs-priestess-advertorial-evolver-001/
• Override naming suffix: Priestess Evolver
• Instructed by: Edward
• Reason: Edward explicitly directed TBF Exp 014 to the Priestess Evolver destination.
• Future AdLauncher must use --allow-non-sheet-url --override-reason or the canonical manifest equivalent.

Hypothesis: If extractive compression improves allocation/conversion while the literal winning message and production stay fixed, the 767-word depth is not load-bearing; if Control wins, depth/progression is load-bearing.

Prospect claim: The Cursed responds to the full Richard Hook 2 Direct Pain Point message; this variant works iff the same literal claims at 450 or 250 words perform comparably or better.

Placement: Problem-aware. The avatar already believes something is wrong with their luck/fortune; the test varies how much narrative depth is required to carry the curse mechanism to a click.

Novel variable: Extractive word-count compression (complete-sentence deletion only). Sentence order, wording, and claims are preserved; only whole sentences are removed.

Hold constant:

• Audience: The Cursed (source audience: 17 Jetsetters Curses)
• Destination URL: https://discover.karmaitems.com/tbf-2005-crs-priestess-advertorial-evolver-001/
• Campaign context: 20260222 - TBF | Curses - Testing
• Primary text, headline, CTA, offer, and Sheet contract (unchanged from source ad)
• SoulGem translation format, same Champa Dorje character/photo/voice/style/music/captions for all three renders
• Product: Tibetan Bracelet of Fortune
• Avatar: The Cursed
• Creative type: First Person Contemporary VSL
• Hook style: Direct Pain Point

Primary test variants:

• Control — source script, 767 words

> If your money keeps disappearing no matter what you do, it may not be your spending. It may be a wealth curse attached to your root chakra, and once that curse is broken, your fortune can begin moving again in as little as two weeks. My name is Champa Dorje. Money disappearing from your life faster than it should? You earn it, you save it, you try to be careful with it, but somehow it leaves almost as soon as it arrives, not all at once, slowly, almost quietly. Unexpected problems start appearing one after another. The peace inside the home disappears, and even when good things happen, they never seem to last very long. In Tibet, monks have warned about this for generations, because not every spiritual problem comes from black magic, not every curse comes from a spell, and not every curse looks scary at first. Sometimes it begins when someone looks at you with envy, with resentment, the kind of people who hide behind smiles, someone watching your life, your home, your family, your small successes, while quietly wishing they had what you have. Most people think those emotions disappear on their own, but in Tibetan spiritual teaching, envy leaves behind energy, and over time it can settle near your root chakra, the energy center connected to your safety, your stability, your money, and your place in this world. And once this energetic interference attaches to you, it becomes difficult for fortune to flow naturally through your life. This is why you may feel stuck despite doing everything right. You can work hard and still lose money. You can pray and still feel exhausted. You can finally begin recovering, only to watch another setback appear again. After that, your timing changes, your peace changes, your money changes, your confidence changes, and even the way you see your future changes. Monks refer to this as a wealth curse. It does not mean you are weak, it does not mean you are careless with money, and it does not mean you did something wrong. It means someone's bad energy may have attached to your foundation and started draining what was meant for you. It does not usually leave by itself. You cannot always think it away. You cannot always pray it away. And you cannot always manifest it away. The good news is, this condition does not have to last forever. It can be weakened. Not through luck, not through positive thinking alone, but through spiritual protection rituals passed down for generations. One story of 17 travelers helped bring this teaching beyond the Himalayas. These people already had money, but their fortune had started leaking. Their peace was gone. Their timing was off. Their opportunities were blocked. So they traveled deep into Tibet and met a monk named Yonge. They expected a long ritual, but Yonge showed them something simple on his wrist. A copper bracelet. He said, when copper is blessed with protection mantras and worn against your skin, it can help break the attachment. It can help seal the leak in your root chakra. And it can help protect you from future envy, resentment, and ill will. This is why Karma Items now offers the Tibetan Bracelet of Fortune. It is not ordinary jewelry. It is spiritual protection. Protection for you if money keeps disappearing. Protection for you if your luck changed after someone turned against you. Protection for you if you know, deep down, that envy has been blocking your path. And when the envy curse is stopped, your fortune can begin to move again. Some people say they start noticing small shifts in as little as 14 days. Money feels easier to hold. Unexpected blessings start appearing. And the heavy feeling around your life begins to lift. But there is one warning. This bracelet is not for greed. It is for people with good intentions. People who want peace, stability, and abundance so they can protect themselves and help the people they love. Over 12,000 people have already ordered one. Right now, Karma Items is offering a limited 50% discount and every bracelet is protected by a 90-day money-back guarantee. So if this speaks to you on any level, if your money keeps disappearing, if you want to break the wealth curse in two weeks, or if you want to start attracting unexpected money in 14 days, do not ignore that feeling. Your fortune may not be gone. It may simply be blocked by someone else's bad energy. Click below to see if the Tibetan bracelet of fortune is still available today.

• V1 — 450w Compression — 447 words

> If your money keeps disappearing no matter what you do, it may not be your spending. It may be a wealth curse attached to your root chakra, and once that curse is broken, your fortune can begin moving again in as little as two weeks. My name is Champa Dorje. Money disappearing from your life faster than it should? You earn it, you save it, you try to be careful with it, but somehow it leaves almost as soon as it arrives, not all at once, slowly, almost quietly. Sometimes it begins when someone looks at you with envy, with resentment, the kind of people who hide behind smiles, someone watching your life, your home, your family, your small successes, while quietly wishing they had what you have. Most people think those emotions disappear on their own, but in Tibetan spiritual teaching, envy leaves behind energy, and over time it can settle near your root chakra, the energy center connected to your safety, your stability, your money, and your place in this world. And once this energetic interference attaches to you, it becomes difficult for fortune to flow naturally through your life. Monks refer to this as a wealth curse. It means someone's bad energy may have attached to your foundation and started draining what was meant for you. The good news is, this condition does not have to last forever. It can be weakened. Not through luck, not through positive thinking alone, but through spiritual protection rituals passed down for generations. One story of 17 travelers helped bring this teaching beyond the Himalayas. These people already had money, but their fortune had started leaking. Their peace was gone. Their timing was off. Their opportunities were blocked. So they traveled deep into Tibet and met a monk named Yonge. They expected a long ritual, but Yonge showed them something simple on his wrist. A copper bracelet. He said, when copper is blessed with protection mantras and worn against your skin, it can help break the attachment. It can help seal the leak in your root chakra. This is why Karma Items now offers the Tibetan Bracelet of Fortune. Protection for you if money keeps disappearing. Protection for you if your luck changed after someone turned against you. Protection for you if you know, deep down, that envy has been blocking your path. Some people say they start noticing small shifts in as little as 14 days. Over 12,000 people have already ordered one. Right now, Karma Items is offering a limited 50% discount and every bracelet is protected by a 90-day money-back guarantee. Click below to see if the Tibetan bracelet of fortune is still available today.

• V2 — 250w Compression — 262 words

> If your money keeps disappearing no matter what you do, it may not be your spending. It may be a wealth curse attached to your root chakra, and once that curse is broken, your fortune can begin moving again in as little as two weeks. My name is Champa Dorje. Money disappearing from your life faster than it should? Sometimes it begins when someone looks at you with envy, with resentment, the kind of people who hide behind smiles, someone watching your life, your home, your family, your small successes, while quietly wishing they had what you have. Most people think those emotions disappear on their own, but in Tibetan spiritual teaching, envy leaves behind energy, and over time it can settle near your root chakra, the energy center connected to your safety, your stability, your money, and your place in this world. And once this energetic interference attaches to you, it becomes difficult for fortune to flow naturally through your life. It means someone's bad energy may have attached to your foundation and started draining what was meant for you. A copper bracelet. He said, when copper is blessed with protection mantras and worn against your skin, it can help break the attachment. It can help seal the leak in your root chakra. Some people say they start noticing small shifts in as little as 14 days. Right now, Karma Items is offering a limited 50% discount and every bracelet is protected by a 90-day money-back guarantee. Click below to see if the Tibetan bracelet of fortune is still available today.

Decision logic:

• If Control wins clearly → the full 767-word depth/progression is load-bearing; do not compress this message for this avatar/production set.
• If V1 450w Compression wins clearly → a moderate extractive compression retains the winning message while improving allocation/conversion; 767-word depth is not load-bearing.
• If V2 250w Compression wins clearly → the message can be delivered in a very short form; progression/proof layering is not load-bearing.
• If V1 and V2 both beat Control → compression itself is the asset; the shortest passing variant becomes the new production default.
• If challengers perform within noise of Control → message length is not a decisive variable for this avatar/production set; keep the Control as the default.
• If any challenger spend < $50 → NO-DELIVERY for that challenger; log as Meta rejection per doctrine, but any load-bearing conclusion about compression stays at Early-signal confidence until one fresh-pair re-run confirms.

Delivery risk: Starvation-prone removal test against an incumbent with conversion history. Launch as a fresh ad set containing the Control clone plus both challengers; do not drop challengers into the incumbent ad set.

Why it matters: Determines whether the proven Richard Hook 2 / Champa Dorje Direct Pain Point message can be compressed without losing conversion. If compression wins, future Cursed variants can be produced more cheaply and tested faster; if Control wins, depth/proof layering is load-bearing and must be protected in future scripts.

Money Hook Trim cutover: 2026-07-10 — replacements activated and originals paused under the same live adset.

• Cutover batch dir: /root/.hermes/output/meta-uploads/20260710-tbf-exp014-money-hook-trim/
• Trim starts: Control 19.32s, V1 18.34s, V2 18.46s
• Replacement ads (ACTIVE):
• Control — 120246465326610539 — video 1001114769401430 — trim 19.32s
• V1 450w Compression — 120246465334180539 — video 1027909056265732 — trim 18.34s
• V2 250w Compression — 120246465342150539 — video 1385560896852335 — trim 18.46s
• Original ads (PAUSED):
• Control — 120246462445630539 — video 1492343958881563
• V1 450w Compression — 120246462449750539 — video 1023041407009262
• V2 250w Compression — 120246462456350539 — video 2064442354154780
• Update draft IDs: replacement active 6a5116305aed26c2e2f1ecca / 6a5116315aed26c2e2f1ecd0 / 6a5116335aed26c2e2f1ecd2; original paused 6a51167d5aed26c2e2f1ed55 / 6a51167f5aed26c2e2f1ed64 / 6a5116815aed26c2e2f1ed76
• Analytical status remains Planned/Pending; historical launch IDs preserved.

Element tested: Extractive message length on the scaled Joan/Miranda Overlooked VSL — same complete-sentence deletion recipe as Exp 014, transferred onto the 1993 timeline-fix Control.

Status: Planned — scripts locked on the 1993-fix Control; do not enqueue until Exp 036 is called

Family: monk-trial

Parent: Exp 014

Altitude: Rung 7 — copy syntax / production-cost rule

Move: Translate

Ad set name: 20260815 - EXP - TBF - Exp 49 Miranda Compression - Expert Healer (Overlooked)

Destination shorthand: Expert Healer (Overlooked)

Variant map:

• 20260815 - EXP - TBF - Exp 49 Control Miranda Compression - Expert Healer (Overlooked) — 1993 timeline-fix Joan VSL verbatim (922w, sha256 a9937de519747c05ad572857c1e409e5d1933b1a3157cf97451efab1d1322b94); reuse the hyphen-only 1993-fix video sha256 3fc3e50701aa303a053c4dffe828a3ccef253185aaa794a7d07c0fe0e95c4ecf — NOT the Facebook reel and NOT the drain file named CONTROL VIDEO (that one still says 1986)
• 20260815 - EXP - TBF - Exp 49 V1 450w Miranda Compression - Expert Healer (Overlooked) — extractive 463w complete-sentence subsequence (sha256 763bc12117bd9d3164806a8d77ffc287ac0d56bc7d34c45c3885e80b2d257ddd)

Result: Pending

Date: 2026-08-15

Avatar: The Overlooked

Creative type: First Person Contemporary VSL

Campaign ID: 120237768514430539

Campaign: 20260222 - TBF | Overlooked - Testing

Live destination: https://discover.karmaitems.com/tbf-1206-healer-overlooked-advertorial-ev-001/

Source observation: Exp 014 mapped a 450-word band on the Richard/Champa Cursed message. The honest scaled over-long Overlooked script is the Joan/Miranda VSL after the 1993 timeline fix (first win 1993 / $5.4M matches the real Joan Ginther fact; bracelet 1992; Drepung; ages dropped). Control is that locked VO and its already-produced video, not the old 1986 Miranda cut. Exp 036 is already live on a close-structure question against an earlier Miranda cut; this experiment stays staged on that read and must not launch into the same Overlooked CBO while 036 is open.

Hypothesis: If the 450-word band is a TBF production rule and not Cursed/monk-trial-only, an extractive 450w cut of the 1993-fix Joan VSL will take allocation against that Control. If Control wins or V1 starves, Overlooked wants the long form and the Exp 014 band does not transfer.

Prospect claim: The Overlooked currently stays for the full Joan proof-and-close on the googleable 1993 timeline; V1 works iff the same literal wins, researcher proof, monastery mechanism, and close still convert when interior scaffolding is deleted.

Placement: Product-aware — the avatar already knows the Joan story; the test varies how much narrative depth is required to carry it.

Novel variable: Extractive word-count compression (complete-sentence deletion only). Sentence order, wording, and claims are preserved; only whole sentences are removed.

Declared coupled variable: Length drop 922 → 463 is the variable. The V5 “next seven minutes” line is deleted in V1 because it would be a false duration claim at the compressed length. The close block is held constant so this test does not re-ask Exp 036.

Hold constant:

• 1993 timeline-fix facts: jackpot in 1993, scratch tickets after, bracelet from Drepung in 1992, “more than 30 years” left on Control only, ages absent
• Protagonist, story order, and literal winning claims (four lottery wins, 17-trillion / 18-septillion odds, University of Illinois professor, Daniel Reyes 19/31, Tenzin chi/conductor, 50 bracelets a month)
• Close block: “The statisticians still argue…” through “Grab yours before this month’s blessing ends.”
• Audience: The Overlooked
• Destination URL: https://discover.karmaitems.com/tbf-1206-healer-overlooked-advertorial-ev-001/
• Destination shorthand: Expert Healer (Overlooked)
• Campaign: 20260222 - TBF | Overlooked - Testing (120237768514430539)
• Primary text, headline, CTA, offer, and Sheet contract
• Production: Control reuses the existing 1993-fix Control video; V1 matches that video’s voice, pacing, and framing; only spoken length changes
• Product: Tibetan Bracelet of Fortune
• Spelling: keep source Drepung (do not revert to Drapung)

Primary test variants:

• Control — 1993 timeline-fix source script, 922 words, sha256 a9937de519747c05ad572857c1e409e5d1933b1a3157cf97451efab1d1322b94
• V1 — 450w Compression — 463 words, sha256 763bc12117bd9d3164806a8d77ffc287ac0d56bc7d34c45c3885e80b2d257ddd — ordered complete-sentence subsequence; keep indices 0,1,14,17,19,20,24,26,35,36,38,39,40,41,42,43,45,47,50,51,56,58,63,66,70,71,76,79,84,85,86,87,88,89,90,92,93,94,95,96,97

Decision logic:

• V1 delivers and Control starves → 450w band transfers to Overlooked long VSLs; brief future long TBF VSLs to the 450w band. [product]
• Control delivers and V1 starves → 450w is Cursed/monk-trial only; do not compress Overlooked long-form. [family]
• Both deliver, V1 takes allocation → transfer confirmed; 450w becomes the TBF briefing default for long VSLs.
• Both deliver, Control takes allocation → depth is still load-bearing for Overlooked even when the short cut can get some delivery; keep Joan long.
• Either arm under $50 lifetime → that arm is NO-DELIVERY = Meta rejection; never re-run the rejected arm; the delivered winner is the verdict.
• Both under $50 → whole-adset rejection; no directional read.

Delivery risk: Fresh two-arm CBO adset. Never inject V1 into the live Joan/Miranda ITER adset. Starvation-prone near-duplicate of a scaled incumbent — pre-register the amended-doctrine NO-DELIVERY branch above. AdLauncher fixed CBO floor: 2000¢ per adset.

Sequencing / launch gate: Do not enqueue until Exp 036 is called. If Edward kills the transfer after 036, abandon this card rather than launch it.

Transferable rule: If this resolves, future TBF ads will be built differently: brief every long VSL to the 450-word band, or treat 450 words as Cursed/monk-only and leave Overlooked long. Scope: [product] if it transfers, [family] if it dies.

Why it matters: This is the cheapest test that upgrades or kills the Exp 014 production rule off the family that discovered it, on the current Joan Control rather than the retired 1986 timeline.

Spend contract: Fresh CBO adset in campaign 120237768514430539; no adset budget. AdLauncher fixed CBO floor: 2000¢ per adset, not per arm.

Production route: Manual Trello / Drive filing — Holding (VSLs). Control + V1 in one Google Doc. Control video is the existing 1993-fix Control — do not re-render it. V1 is a new spoken-length cut matching that video. Checkpoint 3: nothing goes live on Meta without Edward's go after Exp 036.

Scripts: adchain/runs/2026-08-15_tbf-exp-049-miranda-compression/

Control script source: https://docs.google.com/document/d/127AI_4B9ZWPNfUyZLnkHGf8Er_pSkaMKUSJUvFPOUrk/edit

Google Doc: https://docs.google.com/document/d/16LR_iwHqQ88ZjRC9wIZSze168usKFLuY3jQnSeQ7UWU/edit?usp=drivesdk

Drive folder: https://drive.google.com/drive/folders/1Tz1qSmdvO1VdgxBNUMeeNqlrqIw6lThm

Trello: https://trello.com/c/Jt39snWO/475-20260815-tbf-overlooked-avatar-miranda-experiment-message-compression

Source Trello card ID: 6a805d27ed0375deb22f1dac

Reference video: https://tools.ambercollective.co/media/junglelab/tbf-exp049/20260807-joan-v5-1993-timeline-fix-control.mp4

Control video (reuse): sha256 3fc3e50701aa303a053c4dffe828a3ccef253185aaa794a7d07c0fe0e95c4ecf

Do not use: https://www.facebook.com/reel/1290319659847016 — that reel still speaks 1986. The drain file named CONTROL VIDEO.mp4 (sha 29f62d80…) is the same 1986 V5 lead.

1993-fix Trello: https://trello.com/c/ONx1DQGe/450-20260807-tbf-overlooked-joan-copy-payoff-v5-corrected-script-1993-timeline-fix-expert-healer-overlooked

Element tested: Whether the anchor's observable-improvement sentences, dropped in the Exp 14 compression, earn allocation when restored verbatim to the delivered 397-word Champa Control — a close that states only a 14-day timing cue versus the same close carrying the anchor's three improvement sentences.

Status: Complete

Result vs decision logic: Branch 2 (Partial no-delivery) fired: V1 delivered ($67.63 lifetime) and Control starved ($15.32 lifetime), so the delivered arm is the verdict and the starved arm is a rejection never re-run. Branch 1 (Delivered, both arms) not fired; Branch 3 (Total no-delivery) not fired. Verdict: restoring the anchor's observable-improvement sentences to the close wins — V1 over the timing-cue-only Control.

Learnings:

• [family] Restore the observable-payoff sentences in the close — a payoff the viewer can picture (“money feels easier to hold”) earns the read, while a payoff stated only as a date does not.
• [family] When you compress a source VSL, preserve the observable-outcome sentences and take the length savings from elsewhere — the Exp 14 compression cut load-bearing copy.
• Transfer: Cheapest test: push the close to full concreteness — rewrite the two non-observable restored sentences (“unexpected blessings start appearing”, “the heavy feeling begins to lift”) as pictureable outcomes and test that fully-concrete close against the partly-concrete V1.

Family: monk-trial

Parent: Exp 014

Move: Diagnose and restore — concrete outcome

Production authorization: Edward approved production 2026-09-05 (chat "go"); normal-cadence enqueue; no forced release. Destination fanout disabled via AdLauncher's canonical override at Edward's direction.

Date: 2026-09-05

Avatar: The Cursed

Campaign ID: 120237762972300539

Campaign: 20260222 - TBF | Curses - Testing

Destination shorthand: 17 Jetsetters (Cursed)

Destination URL: https://discover.karmaitems.com/tbf-1206-curses-advertorial-ev-001

Ad set name: 20260905 - EXP - TBF - Exp 85 Concrete Outcome Restoration - 17 Jetsetters (Cursed)

Live ad set name: 20260905 - P - EXP - TBF - Exp 85 Concrete Outcome Restoration - 17 Jetsetters (Cursed)

Ad set ID: 120247732356740539

Live ad IDs:

• Control: 120247732358460539
• V1: 120247732359210539

AdLauncher request ID: produced-tbf-exp085-concrete-outcome-restoration-20260905-r2-36abe913

AdLauncher queue state: live (publisher proof; register auto-synced)

AdLauncher request SHA-256: c0603973766aa2714e4c4a2179179286b829903423f52d3c488d603b46b1dc72

Queue receipt: Exactly two queue-owned media in Control → V1 order (Control sha256 1154f91623a3e2e8d5b1cb24b137aad7de8d75e5a00b1c04af40ca801e12521f, V1 sha256 0a5b3de0346b61e3e3fb79d6a43f54c3a87914455dba844918b490df26e6708f). The ad↔page congruence gate was genuinely re-run for this batch (ad_congruence_gate.py gate) and returned pass on 3 judge samples per variant across 2 variants; its report is at the batch's congruence/gate_report.json. Canonical Sheet copy/URL, product identity, custody hash parity, targeting provenance, and the $20 CBO spend contract all passed. Destination fanout disabled via AdLauncher's canonical operator_approval override. Template adset 120247685019890539 (Exp 67, live). No release was forced.

Source observation: The champa-wealth-curse anchor (swipe sw_1d49e8f3b2, SwipeSmith ss_7ecf0f2ec1, Meta ad 120242208504860539, 767 words) closes its payoff with a 14-day timing cue followed by three observable-improvement sentences: "Money feels easier to hold. Unexpected blessings start appearing. And the heavy feeling around your life begins to lift." The Exp 14 extractive compression (767 to 447 words) kept the timing cue and dropped all three; the later money-hook trim removed only the pre-hook. The delivered 397-word Control (sha256 421d5ba271da23b81404f75869b65b1c3368f2be06d7123b99a677c96be3f1c9, trimmed from winner ad 120246465334180539) therefore says when a change might arrive and never says what changes. Separately, Exp 56 found that in the Cursed dense-caption family Meta allocated to the arm that kept concrete, observable payoff language over abstract phrasing (V1 $55.92 lifetime vs Control $19.99 NO-DELIVERY) — directional only, since neither arm purchased and the Control starved.

Hypothesis: Restoring the anchor's observable-improvement sentences to the delivered Control's close will win allocation, because a payoff the viewer can picture beats a payoff stated only as a date.

Novel variable: Presence of the anchor's three observable-improvement sentences immediately after the existing 14-day sentence. Pure extractive restoration — V1 adds no new words; every restored sentence appears verbatim in the anchor transcript in the same order. The 19-word length increase (397 to 416) is a declared consequence of the restoration, not a second variable.

Hold constant: Product tbf; The Cursed avatar; 17 Jetsetters (Cursed) destination; surface facebook_feed_video; production style First Person Contemporary VSL; SoulGem authority_talking_head with character Champa Dorje and use_product_photos: false (control renderer sha256 65df89eb9af6bd2463626c99095665e0a5f7a35072432b79c4eed33219fa8504); CTA learn_more; canonical Sheet headline and primary text; campaign 120237762972300539 and its targeting; the 14-day timing cue, which is already in the delivered Control and stays in both arms; every other byte of the delivered Control. Control and V1 are freshly rendered together from the locked scripts so no production confound separates them.

Variant map:

• 20260905 - EXP - TBF - Exp 85 Control Concrete Outcome Restoration - 17 Jetsetters (Cursed) — delivered Champa Control verbatim; 397 words; script sha256 421d5ba271da23b81404f75869b65b1c3368f2be06d7123b99a677c96be3f1c9.
• 20260905 - EXP - TBF - Exp 85 V1 Concrete Outcome Restoration - 17 Jetsetters (Cursed) — Control plus the anchor's three observable-improvement sentences restored after the 14-day sentence; 416 words; script sha256 c5c95a6990510aa0443f6717ed62ee182e8a515af20a751f7132997e4a3241af.

Swap spec:

• Remove: nothing.
• Insert after: "Some people say they start noticing small shifts in as little as 14 days."
• Insert: "Money feels easier to hold. Unexpected blessings start appearing. And the heavy feeling around your life begins to lift."
• Rejoin at: "Over 12,000 people have already ordered one."

Decision logic:

• Delivered: If both arms deliver, the allocation winner is the verdict. V1 wins — restore the anchor's observable payoff in every Champa-lineage script and treat the Exp 14 compression as having cut load-bearing copy. Control wins — the timing cue alone is sufficient, the compression was correct, and no further outcome detail is restored to this lineage.
• Partial no-delivery: If only one arm delivers, that delivered arm is the verdict and the starved arm is a Meta rejection that is never re-run.
• Total no-delivery: If both arms starve, record an early signal only and require one fresh-pair confirmation.

Delivery risk: Fresh two-arm CBO adset; the challenger is never dropped into the incumbent's adset. The arms are near-siblings differing by 19 words (TBF 013 starvation shape), so the partial-no-delivery branch above is the pre-registered amended-doctrine verdict. AdLauncher fixed CBO floor: 2000 cents. Destination fanout disabled — two ads, not four.

Transferable rule: If V1 wins, extractive compressions of a source VSL must preserve the observable-payoff sentences and take their length savings elsewhere. If Control wins, a stated timing cue substitutes for outcome detail in this avatar's close, and Exp 56's concreteness rule stays scoped to the dense-caption family.

Interpretation limit: The anchor's own outcome span is only partly concrete — "money feels easier to hold" is observable, "unexpected blessings start appearing" and "the heavy feeling begins to lift" are not. This test asks whether restoring the source's outcome detail earns allocation, which is a weaker question than Exp 56's concreteness variable. A V1 win does not prove concreteness specifically.

Production note: Both arms were rendered fresh on 2026-09-05 rather than reusing the delivered Control creative. references/decomposition-control-video-reuse.md would normally reuse a proven winner's live video, but SoulGem's authority_talking_head take architecture changed on 2026-09-04 (e54bf56919 — at most two continuous takes, block/punch-in machinery dropped) and the route baseline is now 576x704 rather than the live creative's 720x942, so reuse would have made the arms differ in take structure and dimensions as well as in the tested span. The first render pair was voided and its queue request produced-tbf-exp085-concrete-outcome-restoration-20260905-4c2c5124 was cancelled before release (reason render_confound_arms_have_different_backgrounds), because: pipeline.py picks the presenter still with a vision selector keyed on the script (selector.select(photos, script)), so the two arms landed in different settings — Control a meditation room, V1 a mountain terrace — a visual confound larger than the tested 19-word span. The shipped pair re-renders both arms with the still pinned via --identity-image to Champa_Dorje_-_Prayer_hall.png (sha256 21aecb423fa0b43fe7fdddbd244a8ddb8dfeca7c7b49e594b8ba5e71f63bf379), matching the live Exp 14 creative's setting, and with Control's avatar_prompt.txt (sha256 85c1ab123392ca3b644004bd877f403c927472294621d10745d4855bdb332391) copied into V1's workdir before launch so both arms share one scene description; the pin held on both. Matched pair: SoulGem authority_talking_head, character Champa Dorje, voice AnUAVuhTZsESOSU2wluY, 576x704, two takes per arm. Control 164.04s sha256 1154f91623a3e2e8d5b1cb24b137aad7de8d75e5a00b1c04af40ca801e12521f; V1 171.13s sha256 0a5b3de0346b61e3e3fb79d6a43f54c3a87914455dba844918b490df26e6708f. Burned captions on both arms were repaired post-render with no paid stage re-run — the monk's name transcribed as Yonga/Yongge, 17 as seventeen, sentence boundaries moved, and the brand name Karma Items and the product name Tibetan Bracelet of Fortune rendering lowercase — and both caption tracks are now equal to their approved scripts under a case- and punctuation-sensitive comparison. The first caption pass used a normalizer that lowercased and stripped punctuation, which hid every one of those defects. Residual variance the pipeline does not let us remove: TTS chunking and beat-break insertion are not deterministic across runs, so the arms differ slightly in pause structure on top of the tested span.

Production route: AdChain experiment mode — SoulGem authority_talking_head with the Control's recorded renderer inputs — fresh Sheet/product/congruence gates — one two-media AdLauncher request — normal release cadence. No forced release.

Scripts: output/adchain/tbf-cursed-concrete-outcome-restoration-20260905/scripts/ (control.txt = delivered Champa Control verbatim; v1-restored.txt)

Result: V1 — top spend $13.51 on 2026-09-11 (latest day). Total adset spend $82.95 over 4 days. Ranking: V1 $13.51 latest / $67.63 total; Control $0.90 latest / $15.32 total NO-DELIVERY (<$50 lifetime). Auto-called per 4d/$50 doctrine.

Element tested: Experiential concreteness in the approved Cursed dense-caption remake: abstract financial pain/payoff versus specific observable losses and improvements.

Status: Complete

Result vs decision logic: The pre-registered decision logic is prose, so branch verdicts are recorded here in prose. V1 (concrete bookends) is the delivered winner under the 4d/$50 highest-spend doctrine: $55.92 lifetime vs Control $19.99 NO-DELIVERY (<$50 lifetime) — the starved Control is an auction rejection, not an unmeasured rival. The 'both arms ≥ $50 with no purchases' rescue-failure branch did not fire because Control never reached $50. Neither arm recorded a purchase in the window, so V1 misses the 1.45 ROAS promotion bar: the concrete direction is confirmed by the frame verdict but is directional rather than promotion-grade, firing the pre-registered rule that future SwipeSmith translations must preserve concrete, observable pain and payoff even when the product mechanism is spiritual.

Learnings:

• [avatar] For The Cursed, preserve concrete, observable pain and payoff language even when the product mechanism is spiritual — the concrete-bookends version is the arm the auction chose to run, so keep the specificity instead of reverting to abstract luck/drain phrasing.
• [family] In the dense-caption family, SwipeSmith translations must keep the source's concrete, observable losses and improvements at the opening and outcome bookends — translating them into abstract belief language weakened the creative, and concreteness is a keeper, not a failed diagnosis.
• Transfer: Cheapest test: decompose the winning bookends — run the concrete opening alone against the concrete outcome span alone on the shared skeleton to learn which end carries the effect, since this result validates the pair only as a bundle.

Family: dense-caption-swipe-translation

Move: Diagnose and restore — concrete problem/outcome

Production authorization: Edward approved paid render and normal-cadence enqueue on 2026-08-22; no forced release.

Date: 2026-08-22

Avatar: The Cursed

Campaign ID: 120237762972300539

Campaign: 20260222 - TBF | Curses - Testing

Template adset: 120246890583210539 — ACTIVE readback verified 2026-08-22

Destination shorthand: Champa Advertorial (Curses)

Destination URL: https://discover.karmaitems.com/tbf-0707-crs-advertorial-ev-001/

Ad set name: 20260822 - EXP - TBF - Exp 56 Concrete Problem Outcome - Champa Advertorial (Curses)

Live ad set name: 20260822 - P - EXP - TBF - Exp 56 Concrete Problem Outcome - Champa Advertorial (Curses)

Ad set ID: 120247530745880539

Live ad IDs:

• Control: 120247530747530539
• V1: 120247530748090539

Source observation: The four-ad remake group spent $81.06 with zero purchases. Its stronger V2 Copy Twist control spent $48.55, produced one checkout, and produced zero purchases. Compared with the original billboard, the remake translated concrete pain and outcomes into abstract “luck bleeding out / drain eased” language that required more belief and was harder to picture.

Hypothesis: Replacing abstract luck/drain language with specific, observable financial losses and improvements will increase purchases.

Novel variable: Experiential concreteness at the opening and outcome bookends only.

Hold constant: Dense-caption format; Sarah voice; exact still sha256 521e6be6239032fd55393af332510f202158c2d9cf3003a21d77a24b68d0f219; exact music sha256 abf131308adce21d89ec35b1a7476118e64b3746a2f87bcbbe85aeee5b00b151; 4:5; caption crawl; TBF — The Cursed audience; curse mechanism; canonical Sheet headline and primary text; CTA learn_more; Champa destination; campaign and targeting. Control and V1 are freshly produced together from the approved scripts.

Variant map:

• 20260822 - EXP - TBF - Exp 56 Control Concrete Problem Outcome - Champa Advertorial (Curses) — approved V2 Copy Twist Control; 481 words; script sha256 821bc033ec23e5ac4be9da8a2106c2f39dc987eb5e563248aad11df2906692f2.
• 20260822 - EXP - TBF - Exp 56 V1 Concrete Problem Outcome - Champa Advertorial (Curses) — concrete opening + concrete outcome spans; 487 words; script sha256 64a301ba1a16a7198a95a31e4269811f03d0b159a54d018c11a6f39284702e39.

Decision logic: Primary outcome is purchase ROAS. V1 must beat its simultaneous Control and reach at least 1.45 ROAS to promote. Checkout, landing-page view, and link CTR are diagnostics only. An arm below $50 lifetime spend is NO-DELIVERY, not a loss. If both arms reach at least $50 with no purchases, the hypothesis failed to rescue the creative.

Delivery risk: Fresh two-arm CBO adset, separate from Exp 57. The exact same newly rendered Control media may be staged in both experiment requests, but each request creates fresh Control and V1 ad entities. AdLauncher fixed CBO floor: 2000¢ per adset, not per arm. Destination fanout is explicitly disabled under Edward's fixed-Champa approval so the two-arm design does not silently become four ads.

Transferable rule: If V1 wins, future SwipeSmith translations must preserve concrete, observable pain and payoff even when the product mechanism is spiritual. If Control wins, abstraction was not the primary commercial failure.

Production route: AdChain experiment mode → canonical SoulGem dense_caption_overlay with locked keyframe/music reuse → fresh Sheet/product/congruence gates → one two-media AdLauncher request → normal release cadence. No forced release.

Scripts: adchain/runs/tbf-cursed-dense-caption-diagnosis-restoration-20260822/

Render receipt: Fresh shared Control: 176.44s, 720×900 H.264/AAC, sha256 bf32717e3866e35030f53d61774e30d520611939f59f88f5a76992b0b21b2ac7. Fresh Exp 56 V1: 183.64s, 720×900 H.264/AAC, sha256 3e405d136756c5e78109d1ef77e060e595f1b9041b0ef5103e6e4e5063f13cdc. Both fully decoded and passed visual contact-sheet review. Each preserves exact still sha256 521e6be6239032fd55393af332510f202158c2d9cf3003a21d77a24b68d0f219, music sha256 abf131308adce21d89ec35b1a7476118e64b3746a2f87bcbbe85aeee5b00b151, Sarah voice, 4:5, and faststart.

AdLauncher request ID: produced-tbf-exp056-concrete-problem-outcome-20260822-8efe36f3

AdLauncher queue state: live (publisher proof; register auto-synced)

AdLauncher request SHA-256: fb6c7ecfbfac60d95cf5435fba0b01614eb1a88d480c3ce44eb355e3bbdc172d

Queue receipt: Exactly two queue-owned media in Control → V1 order. Strict request, Sheet, product, ad↔page congruence, custody, targeting, Champa-only destination, explicit fanout-disabled approval, and $20 CBO-floor checks passed. No release was forced.

Result: V1 — top spend $15.64 on 2026-08-28 (latest day). Total adset spend $75.91 over 4 days. Ranking: V1 $15.64 latest / $55.92 total; Control $5.89 latest / $19.99 total NO-DELIVERY (<$50 lifetime). Auto-called per 4d/$50 doctrine.

Element tested: Commercial credibility in the approved Cursed dense-caption remake: light proof and vague availability versus a substantial proof/value/offer/guarantee block.

Status: Complete

Result vs decision logic: V1 (the proof/value/offer/guarantee block) beat its simultaneous Control on allocation and was the only arm to drive purchases, while Control stalled below the delivery floor. The pre-registered promotion gate had two conditions: V1 clearing its Control (satisfied) and V1 clearing a 1.45 ROAS bar (not met; V1 landed near 1.03). The proof-stack direction is confirmed, but V1 is directional rather than promotion-grade.

Learnings:

• [avatar] Carry the source billboard's full commercial stack into The Cursed dense-caption translations — concrete wearer and review counts, a stated price and savings, and an explicit guarantee — instead of trimming it to light proof and a vague availability nudge.
• [family] In the dense-caption family the closing commercial block is the load-bearing sell: a proof/value/offer/guarantee close beat a lighter proof-and-availability close on the same creative, so do not soften the sell when localizing a source billboard.
• Transfer: Cheapest test: decompose the winning bundle — isolate which single commercial component (proof stack, price and savings, or guarantee) carries the lift, since the bundle's win validates the whole and not any single part.

Family: dense-caption-swipe-translation

Move: Diagnose and restore — commercial credibility

Production authorization: Edward approved paid render and normal-cadence enqueue on 2026-08-22; no forced release.

Date: 2026-08-22

Avatar: The Cursed

Campaign ID: 120237762972300539

Campaign: 20260222 - TBF | Curses - Testing

Template adset: 120246890583210539 — ACTIVE readback verified 2026-08-22

Destination shorthand: Champa Advertorial (Curses)

Destination URL: https://discover.karmaitems.com/tbf-0707-crs-advertorial-ev-001/

Ad set name: 20260822 - EXP - TBF - Exp 57 Commercial Credibility - Champa Advertorial (Curses)

Live ad set name: 20260822 - P - EXP - TBF - Exp 57 Commercial Credibility - Champa Advertorial (Curses)

Ad set ID: 120247530847340539

Live ad IDs:

• Control: 120247530849250539
• V1: 120247530850590539

Source observation: The remake retained the source billboard's dense-caption surface but weakened what it sold: 12,000 wearers, 1,400 reviews, and “17 people met Yongey,” then lunar scarcity and “check availability,” versus the original's large proof stack, explicit daily price/savings, benefits, and 90-day guarantee.

Hypothesis: Adding a substantial proof stack plus explicit price, savings, benefits, and guarantee will increase purchases.

Novel variable: Commercial substantiation density in the final block only. This intentionally bundles proof, value, offer, and guarantee; a win validates the bundle, not any single component.

Hold constant: Dense-caption format; Sarah voice; exact still sha256 521e6be6239032fd55393af332510f202158c2d9cf3003a21d77a24b68d0f219; exact music sha256 abf131308adce21d89ec35b1a7476118e64b3746a2f87bcbbe85aeee5b00b151; 4:5; caption crawl; TBF — The Cursed audience; curse mechanism; canonical Sheet headline and primary text; CTA learn_more; Champa destination; campaign and targeting. Control and V1 are freshly produced together from the approved scripts.

Variant map:

• 20260822 - EXP - TBF - Exp 57 Control Commercial Credibility - Champa Advertorial (Curses) — approved V2 Copy Twist Control; 481 words; script sha256 821bc033ec23e5ac4be9da8a2106c2f39dc987eb5e563248aad11df2906692f2.
• 20260822 - EXP - TBF - Exp 57 V1 Commercial Credibility - Champa Advertorial (Curses) — proof/value/offer/guarantee replacement block; 486 words; script sha256 062ef9faddf5eac0e332f3d04c972a9f3a1be95f3315219d778b9a5d5bb79058.

Decision logic: Primary outcome is purchase ROAS. V1 must beat its simultaneous Control and reach at least 1.45 ROAS to promote. Checkout, landing-page view, and link CTR are diagnostics only. An arm below $50 lifetime spend is NO-DELIVERY, not a loss. If both arms reach at least $50 with no purchases, the hypothesis failed to rescue the creative.

Delivery risk: Fresh two-arm CBO adset, separate from Exp 56. The exact same newly rendered Control media may be staged in both experiment requests, but each request creates fresh Control and V1 ad entities. AdLauncher fixed CBO floor: 2000¢ per adset, not per arm. Destination fanout is explicitly disabled under Edward's fixed-Champa approval so the two-arm design does not silently become four ads.

Transferable rule: If V1 wins, future SwipeSmith translations must carry a credible commercial proof/value/offer/guarantee block proportionate to the source. If Control wins, the lighter commercial stack was not the primary failure.

Production route: AdChain experiment mode → canonical SoulGem dense_caption_overlay with locked keyframe/music reuse → fresh Sheet/product/congruence gates → one two-media AdLauncher request → normal release cadence. No forced release.

Scripts: adchain/runs/tbf-cursed-dense-caption-diagnosis-restoration-20260822/

Render receipt: Fresh shared Control: 176.44s, 720×900 H.264/AAC, sha256 bf32717e3866e35030f53d61774e30d520611939f59f88f5a76992b0b21b2ac7. Fresh Exp 57 V1: 169.64s, 720×900 H.264/AAC, sha256 933521b300530fc277e3541f08b0e7470d20e4f31b967aa452df8b8c686f9d97. Both fully decoded and passed visual contact-sheet review. Each preserves exact still sha256 521e6be6239032fd55393af332510f202158c2d9cf3003a21d77a24b68d0f219, music sha256 abf131308adce21d89ec35b1a7476118e64b3746a2f87bcbbe85aeee5b00b151, Sarah voice, 4:5, and faststart.

AdLauncher request ID: produced-tbf-exp057-commercial-credibility-20260822-86f8d700

AdLauncher queue state: live (publisher proof; register auto-synced)

AdLauncher request SHA-256: 09f2f9c33dd1d43797926c54a0a06ac64061d782544db30c229350d3d0ead059

Queue receipt: Exactly two queue-owned media in Control → V1 order. Strict request, Sheet, product, ad↔page congruence, custody, targeting, Champa-only destination, explicit fanout-disabled approval, and $20 CBO-floor checks passed. No release was forced.

Element tested: Whether Exp 56's winning concrete opening/outcome bookends add on top of Exp 57's winning commercial-credibility block in the Cursed dense-caption remake — the two winning spans stacked on the shared skeleton versus the stronger single winner.

Status: Complete

Result vs decision logic: Partial no-delivery fired: V1 (combined restoration) delivered while Control (credibility block alone) stayed under the threshold, so V1 is the verdict and Control is a Meta rejection that is never re-run. The concrete opening/outcome bookends add on top of the commercial-credibility block, so the combined restoration becomes this family's control.

Learnings:

• [family] Keep the concrete opening/outcome bookends stacked on the commercial-credibility block when rebuilding the Cursed dense-caption script — the combined span stack beat the credibility block alone.
• [family] Winning spans from sibling isolations stack on a shared skeleton — promote the combined restoration as this family's new control and default to combining disjoint winners before testing narrower variations.
• Transfer: Cheapest test: pit the combined restoration (bookends + credibility block) head-to-head against the live production Cursed dense-caption incumbent to see whether the stacked winner displaces what is already running.

Family: dense-caption-swipe-translation

Parent: Exp 057

Move: Recombine winners — stack disjoint winning spans

Production authorization: Edward approved the experiment design 2026-09-02 (chat); normal-cadence enqueue; no forced release.

Date: 2026-09-02

Avatar: The Cursed

Campaign ID: 120237762972300539

Campaign: 20260222 - TBF | Curses - Testing

Destination shorthand: Champa Advertorial (Curses)

Destination URL: https://discover.karmaitems.com/tbf-0707-crs-advertorial-ev-001/

Ad set name: 20260902 - EXP - TBF - Exp 78 Combined Restoration - Champa Advertorial (Curses)

Live ad set name: 20260902 - P - EXP - TBF - Exp 78 Combined Restoration - Champa Advertorial (Curses)

Ad set ID: 120247690356290539

Live ad IDs:

• Control: 120247690358740539
• V1: 120247690360230539

AdLauncher request ID: produced-tbf-exp078-combined-restoration-20260902-r2-127c3602

AdLauncher queue state: live (publisher proof; register auto-synced)

AdLauncher request SHA-256: ecdf548aad94dd34da42ea09d09de0508848c32f8f13be1db3ba0bac3714975b

Source observation: Exps 56 and 57 each beat the shared abstract V2 Copy Twist Control on allocation and both Controls starved: Exp 56 V1 (concrete bookends) $84.67 lifetime vs Control $23.14 NO-DELIVERY; Exp 57 V1 (commercial-credibility block) $67.97 lifetime vs Control $40.53 NO-DELIVERY. Word-level diff proves the winning spans are disjoint on the 481-word skeleton: Exp 56 changed words 0-30 and 294-336; Exp 57 changed words 336-481; they meet at word 336 with zero overlap.

Hypothesis: Stacking both allocation-winning spans on the shared skeleton will beat the stronger single winner (Exp 57 V1) on allocation.

Novel variable: Presence of Exp 56's concrete opening/outcome bookends on the Exp 57 V1 winner. Assembly assertions verified V1 is byte-identical to Exp 57 V1 outside Exp 56's span positions and byte-identical to Exp 56 V1 outside Exp 57's span position.

Hold constant: Dense-caption format; Sarah voice; exact still sha256 521e6be6239032fd55393af332510f202158c2d9cf3003a21d77a24b68d0f219; exact music sha256 abf131308adce21d89ec35b1a7476118e64b3746a2f87bcbbe85aeee5b00b151; 4:5; caption crawl; TBF - The Cursed audience; curse mechanism; canonical Sheet headline and primary text; CTA learn_more; Champa destination; campaign and targeting; Exp 57's commercial block present verbatim in both arms. Control and V1 are freshly produced together from the locked scripts.

Variant map:

• 20260902 - EXP - TBF - Exp 78 Control Combined Restoration - Champa Advertorial (Curses) — Exp 57 V1 winner verbatim; 486 words; script sha256 062ef9faddf5eac0e332f3d04c972a9f3a1be95f3315219d778b9a5d5bb79058.
• 20260902 - EXP - TBF - Exp 78 V1 Combined Restoration - Champa Advertorial (Curses) — combined restoration (Exp 56 winning bookends + Exp 57 winning block); 492 words; script sha256 33c9748f7e9c354f9f30078147d1b70d32650a41c24ef0b331019e5161e68d59.

Decision logic:

• Delivered: If both arms deliver, the allocation winner is the verdict. V1 wins — the concrete bookends add on top of the credibility block; the combined restoration becomes this family's control. Control wins — the credibility block alone is the stronger asset; span wins do not stack.
• Partial no-delivery: If only one arm delivers, that delivered arm is the verdict and the starved arm is a Meta rejection that is never re-run.
• Total no-delivery: If both arms starve, record an early signal only and require one fresh-pair confirmation.

Delivery risk: Fresh two-arm CBO adset. The arms are near-siblings sharing everything but the bookend spans (TBF 013 starvation shape) — the partial-no-delivery branch above is the pre-registered amended-doctrine verdict. AdLauncher fixed CBO floor: 2000 cents.

Transferable rule: If V1 wins, winning spans from sibling isolations on a shared skeleton stack — combine restoration winners by default in this family. If Control wins, span wins are not additive — test stacks before promoting them.

Production route: AdChain experiment mode — canonical SoulGem dense_caption_overlay with locked keyframe/music reuse — fresh Sheet/product/congruence gates — one two-media AdLauncher request — normal release cadence. No forced release.

Scripts: output/adchain/tbf-cursed-combined-restoration-cp1-20260902/scripts/ (control.txt = Exp 57 V1 verbatim; v1-combined.txt; copy gate passed 2026-09-02)

Result: V1 — top spend $15.78 on 2026-09-09 (latest day). Total adset spend $126.70 over 5 days. Ranking: V1 $15.78 latest / $80.33 total; Control $11.19 latest / $46.37 total NO-DELIVERY (<$50 lifetime). Auto-called per 4d/$50 doctrine.

Element tested: Revision methodology as a bundled treatment — the same live winning script critiqued by each of the three rival copy chiefs (copychief / research-intake-improvement-chief / affiliate-chief), each critique implemented by an isolated fresh session, run head-to-head against the untouched Control.

Status: Pending delivery — corrected-control copy locked at replacement CP1 (Edward approval 2026-08-29, “continue”); clean-script editor handoff verified on Holding (Post Implementation Revisions); no Meta entities yet

Family: overlooked-mccathie-lightning

Date: 2026-08-29

Avatar: The Overlooked

Campaign ID: 120237768514430539

Campaign: 20260222 - TBF | Overlooked - Testing

Template adset: 120246829306190539 — re-verify ACTIVE readback at launch time (Exp 65 verified 2026-08-29)

Destination shorthand: Expert Healer (Overlooked)

Destination URL: https://discover.karmaitems.com/tbf-1206-healer-overlooked-advertorial-ev-001/

Base ad: control creative = Meta video 2468241400345996, canonical control ad 120246306939860539 ("20260706 - TBF - Expert Healer (Overlooked) - Peter Mccathie - New Leads - New Leads 2 - Expert Healer (Overlooked)", Overlooked Testing, $3.4k @ 1.88); the same video also runs as 20260806 production duplicates (UK $5.0k / CA $1.0k / AU $1.0k) — family proof ~$10.4k. Control arm reuses this exact video.

Workspace: products/tbf/manual-experiments/chief-bakeoff-mccathie-20260829/ (critiques, arms, integrity notes); plan share/chief-bakeoff-mccathie-experiment-plan-20260829.html; CP1 page share/chief-bakeoff-mccathie-cp1-20260829.html

Ad set name: 20260829 - EXP - TBF - Exp 66 Chief Critique Bakeoff McCathie - Expert Healer (Overlooked)

Variant map:

• 20260829 - EXP - TBF - Exp 66 Control Chief Critique Bakeoff McCathie - Expert Healer (Overlooked) — control; copy chief-bakeoff-mccathie-20260829/control.txt; 887 words; sha256 74c4b422294670d6edb730cf2dae586c062a8ddcb8f5c755c29458715e40dbaa; transcript frozen from actual Meta video 2468241400345996; final doc is plain script only.
• 20260829 - EXP - TBF - Exp 66 V1 Chief Critique Bakeoff McCathie - Expert Healer (Overlooked) — copychief arm; copy …/arms/v1.txt; 852 words; sha256 7ba9bd482d2dd02d321714c230e2081dccf0f1f9f2ab221f305d76138b30660b; final file is plain script only.
• 20260829 - EXP - TBF - Exp 66 V2 Chief Critique Bakeoff McCathie - Expert Healer (Overlooked) — research-intake-improvement-chief arm; copy …/arms/v2.txt; 913 words; sha256 c7176e37e576b45257d43a9902f5cc86a2e2228c97f6f37ffa4fee094cd6ef7f; final file is plain script only.
• 20260829 - EXP - TBF - Exp 66 V3 Chief Critique Bakeoff McCathie - Expert Healer (Overlooked) — affiliate-chief arm; copy …/arms/v3.txt; 625 words; sha256 bc3cda18d3c8a961e1fc8de02d620169c6bb4232d0c555ac7296b50dfcf25b7d; final file is plain script only.

Hypothesis: Applying a chief-critique revision policy to a proven winning script changes Meta allocation; arms isolate WHICH revision policy (preserve / research-expand / conversion-attack) earns more than leaving the winner alone.

Experiment classification: Multi-change bundle bakeoff, preregistered as such — the treatment is the methodology, not any single line. The result names a winning critique method; mining a winning bundle's individual changes is a follow-up isolation experiment, never a conclusion of this one.

Novel variable: Revision methodology (none / copychief / RIIC / affiliate-chief) applied to identical base footage and script.

Hold constant: Product, avatar, destination URL + shorthand, canonical Sheet primary text and headline (identical all arms), ad style (Third Person Contemporary VSL), format (video), base footage, campaign lane, AdLauncher fixed CBO floor: 2000¢.

Decision logic:

• Delivered: Fresh allocation resolves the preregistered question; winner = highest spend, last day.
• Partial no-delivery: A starved arm is rejected; the delivered allocation winner is the verdict and rejected arms are not rerun.
• Total no-delivery: If all arms starve, record an early signal and permit one fresh-pair confirmation.

Production route: Human video editor via Trello Holding (Post Implementation Revisions) — one card, verified actual control video as approved control, and one consolidated plain-script Google Doc containing the control and all three arms under headings. Implementation/change material is kept separate from the final script document. SoulGem not involved. AdLauncher remains the sole queue and publisher on delivery.

Final script contract: The consolidated Google Doc contains plain scripts only—no editing notes, change logs, B-roll directions, mechanism labels, or QA commentary. Production instructions remain on Trello, separate from the script document.

Google Doc: https://docs.google.com/document/d/1CP2FpvodpQiHjEUqJ54EYimzPK9Rwn-auykq4P1i_Ow/edit (Control + V1 Copychief + V2 RIIC + V3 Affiliate)

Drive folder: https://drive.google.com/drive/folders/1vEvcP0gJY6r9ghxow4igy6_FNwbDJzvG (one consolidated Google Doc + verified actual control video 2468241400345996; all superseded Docs moved to Drive trash)

Trello card: https://trello.com/c/xJoWLPYI — verified on Holding (Post Implementation Revisions)

Integrity notes: All three chiefs independently caught the control's 50-vs-150 cadence drift. Copychief's session read the workspace README after its independent catch (disclosed; note since moved out of workspace). RIIC critique took a bounded delta pass after the research intake was enriched (4,273 → 5,415 words, LP objection extraction; win/loss verdicts excluded from intake). Orchestrator wrote none of the copy.

Element tested: Named wound on the live Peter McCarthy news-doc — luck / impossible-odds versus unmatched potential, three declared spans only.

Status: Pending delivery — editor handoff filed; awaiting Frame.io Control + V1

Family: overlooked-mccathie-lightning

Parent: Exp 66

Date: 2026-08-30

Avatar: The Overlooked

Campaign ID: 120237768514430539

Campaign: 20260222 - TBF | Overlooked - Testing

Template adset: 120246829306190539 — re-verify ACTIVE readback at launch time

Destination shorthand: Expert Healer (Overlooked)

Destination URL: https://discover.karmaitems.com/tbf-1206-healer-overlooked-advertorial-ev-001/

Base ad: Meta video 2468241400345996, canonical control ad 120246306939860539

Ad set name: 20260830 - EXP - TBF - Exp 70 Peter Unmatched Potential - Expert Healer (Overlooked)

Variant map:

• 20260830 - EXP - TBF - Exp 70 Control Peter Unmatched Potential - Expert Healer (Overlooked) — live Peter luck-pain script; 885 words; sha256 74614cd624b6a4ecee2eed5c68a91e9728d372a91e5eb891672b0fdfe5e4835f
• 20260830 - EXP - TBF - Exp 70 V1 Peter Unmatched Potential - Expert Healer (Overlooked) — same Peter, three unmatched-potential spans; 913 words; sha256 11d84bff6a03a6f35deee700f3fcd68e38d32dd84a5f9f63dc5bd1c7d68972eb

Delivery risk: Launch Control + V1 as a fresh CBO adset pair; never inject into the incumbent adset. Any arm under $50 lifetime spend is a Meta rejection — a real verdict against that arm, not an unmeasured rival. If at least one arm delivers, the allocation winner IS the verdict (the winner may be the Control) and rejected arms are never re-run. Only if ALL arms starve is the result an early signal requiring one fresh-pair confirmation.

Hypothesis: Rewriting this Peter winner to name unmatched potential (the pain Meta is funding on Exp 52's intended dest) earns more allocation than keeping luck / impossible-odds. If V1 wins, future Peter-format ads name unmatched potential. If Control holds, keep luck-pain on this story.

Novel variable: Named wound only — luck versus unmatched potential on three declared spans. Byte identity outside those spans.

Hold constant: Product, avatar, destination URL + shorthand, canonical Sheet headline and primary text, Third Person Contemporary VSL, Dramatic Incident, campaign lane, human-editor production on the existing Peter timeline, AdLauncher fixed CBO floor: 2000¢.

Decision logic:

• V1 top spender with ≥$50 lifetime → rewrite this winner for unmatched potential.
• Control top spender with ≥$50 lifetime → keep luck-pain on this Peter story.
• One arm below $50 lifetime while the other delivers → classify the starved arm NO-DELIVERY and call the delivered allocation winner.
• Both arms below $50 lifetime → whole-adset rejection; no directional read.

Spend contract: Fresh CBO adset in campaign 120237768514430539; no adset budget. AdLauncher fixed CBO floor: 2000¢ per adset, not per arm.

Production route: Human video editor revision of live Peter cut via Trello Holding (Post Implementation Revisions). One Drive doc (Control + V1, yellow on the three changed spans). Control is a delivered file ending in Control, not a live-ad URL. VO/caption patch over non-face footage; no new on-camera lines. Frame.io review folder [NEEDS CREATING]. SoulGem/Pictory not involved.

Google Doc: https://docs.google.com/document/d/13M18ycXzFlXTLU7XXsWUeSBdvBdL9NRf1K9aHcWwWxc/edit

Drive folder: https://drive.google.com/drive/folders/1p-b-a2bByFgi_15zb7G1gIkzRGxe994j

Trello card: https://trello.com/c/CmSzxnS6 — verified on Holding (Post Implementation Revisions)

Approved control file: 20260830-EXP70-TBF-Overlooked Peter Unmatched Potential Control.mp4 (Drive 1WIaluk-TPXY0ENdEhKW-cbpbii4f4oDu; source Meta video 2468241400345996)

Result: Pending

Element tested: Script archetype — character framing, proof structure, promise type

Status: Complete — analytical comparison of 1 winner (Joan) vs 6 underperformers (V7–V12)

Family: ginther-vsl

Ad set name: N/A — analytical comparison across multiple ad sets, not a single-adset test

Date: 2026-06-07

Avatar: The Overlooked

Creative type: Authority UGC / Testimonial scripts (Tibetan Bracelet of Fortune)

Context: Joan Ginther testimonial dramatically outperformed 6 follow-up scripts (V7 Marcus, V8 Elena, V9 Denise, V10 Martin, V11 Gwen, V12 Paul). Denise was the only one of the six that performed "somewhat okay." Question: what made Joan work and the rest fail?

Three hypotheses tested against the actual scripts:

H1 — Joan is the only believable peer for Overlooked ✅ Confirmed (with refinement). Original framing ("respectable middle, not too lucky / too tragic") was directionally right but not sharp enough. The real distinction: never-been-recognized peer, not just middle-class.

• V7–V12 all open with prior celebrated achievement: 10th Mountain Division commendation (Marcus), 30 years captaining 747 freighters (Elena), Post-Gazette write-up (Denise), Air Force commendation (Martin), Carnegie Hall + NYT review (Gwen), state press award + Columbia Journalism Review (Paul).
• Overlooked avatar literally says *"I have nothing to show for it"* and *"I have no luck. Seems luck just pass me by."* They've never been somebody — they feel invisible, not faded.
• Joan was uncelebrated: PhD, modest, retired, widowed, pension not stretching. Ordinary. Frano Selak was just a music teacher in Croatia. Both are unrecognized peers.
• The avatar can't project onto a Carnegie Hall violinist whose career stalled. That's a foreign tragedy.

H2 — External, mathematical, third-party proof drives believability ❌ Invalidated as originally stated. Every V7–V12 script has credentialed third-party-finds-the-pattern: Greta (IRS auditor 32yr), Dennis (homicide detective 38yr), Kevin (CPA 20yr), Nadia (Johns Hopkins neuropsychologist), Elliot (sound engineer 12yr), Ruth (retired librarian/archivist). If credentialed third parties were the driver, several would have worked.

• Refined version that might survive: population-level/statistical proof vs single-case credentialed audit. Joan had the University of Illinois researcher (Daniel Reyes) survey 31 multi-lottery winners and find 19 wore copper bracelets — that's sample-based evidence. V7–V12 all use "credentialed friend found a pattern in *my* records" — sophisticated anecdote, still N=1. Refined version untested in this batch.

H3 — Promise must match the actual avatar dream (quiet late-life relief, not comeback) ✅ Strongly confirmed. The avatar doc explicitly says: *"Having enough — not extraordinary wealth, just enough to stop worrying."*

• V7 Marcus → $340K federal grant + $1.7M city contract (career restored at high level)
• V8 Elena → $8,400 back pay + documentary about her life (recognition arc)
• V9 Denise → hotel catering covering rent increase + below-market kitchen ("contract alone covered three months of the rent increase") — the only one with the right promise
• V10 Martin → $8,400 client + project referral (career resumption)
• V11 Gwen → $2,800 chamber gig + studio rediscovers her demo (comeback arc)
• V12 Paul → $1,500 feature + national wire pickup (professional vindication)
• Denise performing "somewhat okay" while the other five tanked is exactly what H3 predicts. The others sell comeback arcs — the dream of someone who *had* something. The Overlooked never did.

Final Learnings:

• Character framing: Use uncelebrated, working-class characters with no prior recognition arc. The Overlooked never had a peak — don't write someone falling from one. Joan worked because she was ordinary. Carnegie Hall, Air Force commendations, NYT reviews all alienate the avatar.
• Promise calibration: Post-bracelet outcomes should be small and dignified — debt going down, a quiet windfall covering what's been mounting, a child finally calling. Not Carnegie Hall comebacks, not multi-million-dollar contracts. The "enough to stop worrying" promise from the avatar doc is literal.
• Proof structure (open question): Population-level statistical proof (researcher surveys N winners, finds pattern) may outperform single-case credentialed audit (CPA friend finds pattern in your records). Worth testing in next batch.
• What V7–V12 got right that we should keep: The "credentialed friend noticed the pattern before I did" mechanic itself is fine — the problem is the *character* and *promise*, not the structural device.

Next test: Write 3–5 new Overlooked scripts with (a) uncelebrated working-class protagonists with no prior recognition, (b) population-level proof source (researcher + sample), (c) dignified, modest post-bracelet outcomes. Compare against Joan baseline.

Element tested: Scientific plausibility / researcher proof block

Status: Complete

Family: ginther-vsl

Parent: Exp 002

Ad set name: 20260623 - TBF - Overlooked Avatar - Ginther VSL Experiment (Scientific Plausibility) - Expert Healer (Overlooked)

Date: 2026-06-20 (planned) · result checked 2026-07-01 · rechecked 2026-07-07

Avatar: The Overlooked

Creative type: First Person Contemporary VSL / Customer testimonial VSL

Trello: https://trello.com/c/9KeKHR6H

Scripts: https://docs.google.com/document/d/1soRmSEC0xhmxgHVk-ermdnJXH3CH30xrXhfGs7MC8EU/edit

Frame.io: https://next.frame.io/project/b99f8398-a80b-4a3e-8020-3a6a0c101625/7fc49dda-de92-4137-85e7-c1f0306c6c0b

Test setup: Control retains the Daniel Reyes researcher scene with statistical proof (19 of 31 multi-lottery winners wore a copper bracelet; 1-in-18-septillion data). Var A removes that scene and replaces it with Joan's own quiet mystical attribution. Joan identity, voice, modesty, four wins, mechanism, offer, and CTA are held constant.

Variant labels:

• Control: Researcher proof block included
• Var A: Mystical attribution only; researcher proof removed

Hypothesis: Does external statistical proof drive performance, or is narrator credibility as a PhD mathematician doing the work?

Result: Control won. Control ad ID 120245465723020539: 90-day spend $226.08, revenue $103.50, 14-day spend $226.08, 7-day spend $121.45 (~0.46x ROAS). Var A ad ID 120245465726610539: 90-day spend $6.66, revenue $0, 14-day spend $6.66, 7-day spend $0.10. Var A's under-$50 spend is treated as Meta rejected / not delivered, not as bad ROAS.

Result vs decision logic: No pre-registered Decision logic block existed for this entry. The outcome answers the hypothesis in favor of Control: the researcher/statistical proof block is load-bearing; removing it stopped meaningful delivery.

Source: Trello card, Google Doc scripts, Meta/Supabase ads_data_master, and AdKit 90-day report (result checked 2026-07-01, rechecked 2026-07-07).

Learnings:

• [avatar] Removing the researcher/statistical proof block stopped delivery; the external proof asset appears necessary for the Overlooked avatar to accept the bracelet mechanism
• [family] An under-$50 spend on Var A means Meta did not find an audience for mystical-only attribution — the platform signal rejects removing proof
• [family] Wait for mature 90-day data before finalizing, but early signal favors keeping the Daniel Reyes / researcher proof scene
• [avatar] If proof is load-bearing, future scripts should keep a population-level or credentialed third-party finding rather than relying on the protagonist's private mystical explanation

Element tested: Astronomical odds / statistical impossibility framing

Status: Complete

Family: ginther-vsl

Parent: Exp 002

Ad set name: 20260623 - TBF - Overlooked Avatar - Ginther VSL Experiment (Septillion Odds) - Expert Healer (Overlooked)

Date: 2026-06-20 (planned) · result checked 2026-07-01 · rechecked 2026-07-07

Avatar: The Overlooked

Creative type: First Person Contemporary VSL / Customer testimonial VSL

Trello: https://trello.com/c/OlQrRHrI

Scripts: https://docs.google.com/document/d/1cdhc-HTUxKkCKiCE0a1b70rUegAfMo7uY_86jK4wQ_E/edit

Frame.io: https://next.frame.io/project/b99f8398-a80b-4a3e-8020-3a6a0c101625/3268e057-ff03-40a5-a672-253b9f926ed8

Test setup: Control keeps all statistical impossibility framing: 1-in-17-trillion, 1-in-18-septillion, the Daniel Reyes researcher block, and the University of Illinois professor quote. Var A removes all statistical impossibility framing while holding constant Joan's identity, voice, modesty, all four wins with dollar amounts, the mechanism explanation, and the offer/CTA word for word.

Variant labels:

• Control: Full statistical odds framing
• Var A: No statistical impossibility framing

Hypothesis: Do astronomical odds figures drive performance by creating scientific plausibility that makes the bracelet explanation believable?

Result: Control won. Control ad ID 120245389140830539: 90-day spend $258.29, revenue $370.30, 14-day spend $258.29, 7-day spend $144.25 (~1.43x ROAS). Var A had no Supabase/AdKit row found (no meaningful delivery / Meta rejected).

Result vs decision logic: No pre-registered Decision logic block existed for this entry. The outcome answers the hypothesis in favor of Control: the astronomical/statistical odds framing is load-bearing; removing it eliminated delivery.

Source: Trello card, Google Doc scripts, Meta/Supabase ads_data_master, and AdKit 90-day report (result checked 2026-07-01, rechecked 2026-07-07).

Learnings:

• [family] Control won early; the astronomical/statistical odds framing appears to be the active proof asset
• [family] Removing all statistical impossibility framing eliminated delivery entirely
• [avatar] The 1-in-17-trillion, 1-in-18-septillion, and professor quote are not decorative — they are load-bearing credibility devices for the Overlooked avatar
• [family] Wait for mature 90-day data, but do not cut the odds framing from future scripts until a clean test disproves it

Element tested: Character voice / framing — modest believable peer vs confident/direct rewrite

Status: Active — launched 2026-06-25 (adset 120245672513310539; register auto-synced 2026-08-02)

Family: ginther-vsl

Parent: Exp 002

Ad set name: 20260625 - EXP - TBF - Exp 8 Ginther Modest Believable Peer - Expert Healer (Overlooked)

Live ad set name: 20260625 - EXP - TBF - Exp 8 Ginther Modest Believable Peer - Expert Healer (Overlooked)

Ad set ID: 120245672513310539

Live ad IDs:

• Control: 120245672526930539

Date: 2026-06-20 (planned) · result checked 2026-07-01 · rechecked 2026-07-07

Avatar: The Overlooked

Creative type: First Person Contemporary VSL / Customer testimonial VSL

Trello: https://trello.com/c/dQIvjQCI

Scripts: https://docs.google.com/document/d/1CPkdHvQpMfD4wxkW9Pqd5Ji8R94wOXoW6BuKqGJViu8/edit

Frame.io: https://next.frame.io/project/b99f8398-a80b-4a3e-8020-3a6a0c101625/1aaaf317-94ba-4500-82aa-ec4bb022287c

Test setup: Var A changes only Joan's disposition: she becomes confident/direct while keeping the same story, stats, and researcher scene. Var B is a full rewrite of Joan with the same character facts to test whether the character concept replicates independent of exact prose. Held constant: Joan identity, four wins plus dollar amounts plus years, the Daniel Reyes researcher scene, mechanism, offer, and CTA.

Variant labels:

• Control: Modest, believable peer voice
• Var A: Confident/direct voice only
• Var B: Full rewrite with same character facts

Hypothesis: Does performance come from Joan being a modest, believable peer — someone the audience recognizes as like them despite extraordinary results?

Result: No meaningful delivery in available data. Supabase ads_data_master and the AdKit 90-day report both returned no spend rows for active ad set 120245672513310539 as of 2026-07-07. The originally planned 20260623 ad set also produced no rows; the live active ad set is the 20260625 ad set above. No winner can be called.

Source: Trello card, Google Doc scripts, Meta/Supabase ads_data_master, and AdKit 90-day report (result checked 2026-07-01, rechecked 2026-07-07).

Learnings:

• [ad] Cannot interpret a voice/framing test without delivery; the ad set is active but has not yet accumulated measurable spend
• [ad] Do not rewrite Joan's modest peer framing until delivery produces a signal
• [ad] If delivery remains absent, treat the variant package as a Meta rejection and return to the proven control voice
• [ad] Once spend arrives, the priority comparison is Control vs Var A (isolated voice change) before evaluating Var B (full rewrite)

Element tested: Video pacing / playback speed

Status: Complete

Family: ginther-vsl

Result vs decision logic: The 1.3x-wins-clean branch fired: V1 at 1.3x speed was the only arm Meta allocated to ($65.82 total, $14.49 latest day), Control at 1.0x was NO-DELIVERY ($24.81). Under the amended doctrine the starved arm is a Meta rejection, not an unmeasured rival — the allocation winner IS the frame verdict. 1.3x playback is the confirmed delivery speed for the Joan Ginther VSL. Conversion-quality watch: no purchase data yet; monitor first conversions but do not re-run the pair.

Learnings:

• [family] 1.3x playback speed is the confirmed delivery pacing for the Joan Ginther VSL family. Future Ginther renders ship at 1.3x by default.
• [universal] A starved losing arm is a rejection by the auction, not a rival that was never measured. Meta preferred the faster cut; that preference is the verdict.
• Transfer: Applies to other long-form monk/testimonial VSLs in this account. Cheapest test: apply 1.3x to the next proven control render and compare allocation against its own history, not against a fresh 1.0x pair.

Ad set name: 20260702 - EXP - TBF - Exp 9 Ginther Playback Speed - Expert Healer (Overlooked)

Date: 2026-07-02 (corrected redo launched)

Avatar: The Overlooked

Creative type: First Person Contemporary VSL / Customer testimonial VSL

Variant map:

• 20260702 - EXP - TBF - Exp 9 Control Ginther Playback Speed - Expert Healer (Overlooked): Control — original Joan Ginther New Leads V1 control at normal playback speed
• 20260702 - EXP - TBF - Exp 9 V1 1.3x Ginther Playback Speed - Expert Healer (Overlooked): 1.3x speed — same verified New Leads V1 control sped up to 1.3x playback speed

Result: V1 — top spend $14.49 on 2026-07-18 (latest day). Total adset spend $90.63 over 5 days. Ranking: V1 $14.49 latest / $65.82 total; Control $0.00 latest / $24.81 total NO-DELIVERY (<$50 lifetime). Auto-called per 4d/$50 doctrine.

Source: Planned experiment requested 2026-07-02 — isolate whether the proven Joan Ginther control performs better when delivered faster. Corrected 2026-07-02 after Edward flagged that the first launch used the wrong source family; the valid redo uses the winner from 20260618 - ITER - TBF Joan Ginther UGC | New Leads - Expert Healer (Overlooked). Supabase/AdKit rechecked 2026-07-07.

Test setup: Same ad set family, same Overlooked audience, same Joan Ginther New Leads V1 winning control creative, same script, same narrator/voice, same visuals, same offer, same CTA, same destination page. What varies: final playback speed only. The 1.3x variant was produced from the same final control render by speeding the full video/audio to 1.3x with pitch correction / natural voice preservation. Captions are hardcoded/burned into the video frames, so they were not rebuilt, re-captioned, or separately retimed; they move faster only because the whole rendered video moves faster. Do not rewrite, trim, reorder, re-record, change music style, change first frame, or change offer mechanics.

Control: Original verified Joan Ginther New Leads V1 control at 1.0x playback speed

Variant 2: Same verified Joan Ginther New Leads V1 control at 1.3x playback speed

Operational handles (corrected redo launched 2026-07-02):

• Correct source control ad set: 120244929239680539 — 20260618 - ITER - TBF Joan Ginther UGC | New Leads - Expert Healer (Overlooked)
• Correct source control ad: 120244929239730539 — 20260618 - TBF Joan Ginther UGC | New Leads V1 - Expert Healer (Overlooked)
• Forbidden/wrong source family was not used for redo: 20260618 - ITER - TBF - Joan Ginther Lead Variants - Expert Healer (Overlooked)
• Source Trello card: https://trello.com/c/XIVjQB6u
• Source Frame.io folder: 2b70f889-71fb-4b71-a116-cb3ce7095e9f
• Corrected local control MP4: /root/.hermes/output/ad-experiments/tbf-joan-ginther-1p3x-redo-new-leads-20260702/control_original.mp4
• Corrected local 1.3x MP4: /root/.hermes/output/ad-experiments/tbf-joan-ginther-1p3x-redo-new-leads-20260702/joan_ginther_1p3x.mp4
• Corrected upload batch: /root/.hermes/output/meta-uploads/tbf-joan-ginther-1p3x-redo-new-leads-20260702/
• Corrected AdKit review URL: https://app.adkit.so/manage/drafts?drafts=6a4631c8631fc700175df872,6a4631c9631fc700175df874,6a4631c9631fc700175df876
• Corrected live ad set: 120246177186050539 (active)
• Corrected live Control ad: 120246177191590539 (active; video ID 976670612061010; creative ID 27025430240472231)
• Corrected live 1.3x ad: 120246177263980539 (active; video ID 2099725920967403; creative ID 1592424499255383)
• Sheet contract passed: sheet_contract.json in corrected upload batch; destination URL is https://discover.karmaitems.com/tbf-1206-healer-overlooked-advertorial-ev-001/.
• Verification files: correct_source_resolution.json, source_control_metadata.json, transform_metadata.json, manifest.json, sheet_contract.json, verify.json, publish.json, and live_status_verify.json in the corrected output folders.

Invalidated previous launch (deleted by Edward):

• Previous first launch was invalidated because it used the wrong source family / deleted live experiment ad sets.
• Deleted previous live ad set: 120246173118210539.
• Deleted previous Control ad: 120246173124610539.
• Deleted previous 1.3x ad: 120246173160220539.
• Previous duplicate ad set 120246173114550539 and duplicate ads 120246173120380539, 120246173159750539 were part of the invalidated/deleted run.
• Previous batch retained for audit only: /root/.hermes/output/meta-uploads/tbf-joan-ginther-1p3x-20260702/.

Hypothesis: Joan Ginther's control has strong proof and story assets, but the original pacing may be slower than ideal for Meta. If the same message is delivered 30% faster without changing content, it may hold attention better and improve front-end engagement while preserving the proof structure that makes the ad believable.

Novel variable: Playback speed / delivery pace only

Hold constant:

• Joan Ginther identity, first-person modest voice, story arc, four-win sequence, dollar amounts, Daniel Reyes researcher scene, statistical odds framing, mechanism explanation, offer, CTA, and destination page
• Same final control render as the source asset for both variants
• Same first frame, visuals, captions text, narrator voice identity, music style, aspect ratio, thumbnail, audience, budget, placement, and landing page
• Captions are burned into the rendered video frames; no caption rebuild, re-captioning, or separate caption retiming was done.

Decision logic:

• If 1.3x wins clean → pacing/delivery speed was suppressing performance; use faster delivery as the new production default for long Joan-style testimonial VSLs while keeping the message intact
• If Control wins clean → the original slower pacing supports believability/emotional processing; do not speed up proof-heavy Joan-style testimonial VSLs by default
• If both perform within noise → playback speed is not the primary bottleneck; keep the cheaper/original production default and test a different isolated variable next
• If 1.3x gets spend but lower conversion quality → faster pacing may improve attention but weaken trust/comprehension; consider a smaller speed lift such as 1.15x instead of 1.3x

Why it matters: This tests a production-level optimization without touching the winning Joan Ginther message. If speed alone lifts performance, future long proof-heavy VSLs can be made more efficient without rewriting the core control.

Element tested: Protagonist character inside the proven Joan Ginther script skeleton — is the 002/006/007 formula a character factory or is Joan a unicorn?

Status: Complete

Family: ginther-vsl

Parent: Exp 002

Result vs decision logic: Branch 1 fired: V4 Marge Bingo (peer-proximity of the luck venue — diner waitress, seven bingo jackpots in 14 months) is the only arm the auction scaled ($146.53 lifetime, $32.20 latest day on 2026-07-23, accelerating over the last five days). V3 Walter ($90.51 lifetime, declining from $16 to under $2/day) was a transient early leader that stalled; V2 Rosa ($32.91), Control Joan ($13.08), V5 Hazel ($3.13), and V1 Earl ($2.84) are all NO-DELIVERY (auction rejections). The Joan script skeleton is confirmed as a character factory: it produced a winning protagonist independent of Joan, across a different luck domain. Branch 2 (Control wins with challengers) did not fire — Control was starved. Branch 3 (all five challengers NO-DELIVERY) did not fire — V4 and V3 both delivered. Confound note remains active: challenger-vs-challenger is the primary read; V4 beat V3-V5 which share the SoulGem production path.

Learnings:

• [ad] Marge (diner waitress with seven bingo jackpots in 14 months, peer-proximity of the luck venue) is a confirmed formula character — the one protagonist the auction scaled from this six-way batch.
• [family] The Joan Ginther script skeleton is a confirmed character factory: it produced a winning protagonist independent of Joan and across a different luck domain. The factory works by swapping the protagonist inside the fixed beat sequence, not by cloning one character type.
• [family] Four starved characters are four auction rejections, not untested rivals — Earl, Rosa, Hazel, and Control Joan herself were each tried and declined.
• Transfer: Applies to character-batch expansion across TBF avatars. Cheapest test: run the next 5-character batch from the same skeleton on a second avatar; Marge becomes the character to beat.

Launch status: Live adset 120246559151220539 in campaign 120237768514430539; minSpend verified 3000¢ (draft verify.json); campaign daily budget auto-raised $450→$481 for the min-spend floor

Live ads:

• Control 120246559153890539 (video 1043818415049371) — Joan New Leads V1 verbatim re-upload
• V1 Earl 120246559157330539 (video 1454449570044076)
• V2 Rosa 120246559165000539 (video 1049137024246968)
• V3 Walter 120246559169300539 (video 2080313459220555)
• V4 Marge 120246559170710539 (video 1730131101617874)
• V5 Hazel 120246559177240539 (video 1716856629346592) — 720p re-render (first render came out 540p from a resume that lost the resolution override; NOT uploaded)

Scripts: /root/.hermes/output/adchain/tbf-exp017-ginther-character-batch/scripts/ (validation.json ok=true; control 34af8ea5…, v1 3ac6844f…, v2 f1a88072…, v3 6034ed31…, v4 433ff596…, v5 2d65a58d…)

Batch artifacts: /root/.hermes/output/meta-uploads/tbf-exp017-ginther-character-batch-20260714/ (sheet_contract ok, product identity ok, congruence pass, live_status_verify + activation)

Production note: adkit CLI NOT_AUTHENTICATED failure on first launch attempt — the node CLI needs ADKIT_API_KEY exported in the shell env (memory: adkit-cli-auth-env); re-run with env resumed from checkpoints.

Move: Translate / Lane A exploration batch — impact-ranked #1 in the 2026-07-14 design session (decompose disqualified: depth cap reached by 006+007; 008's third isolation never delivered). Prescribed as "Next test" by exp 002 and sitting in the Open Questions ledger since.

Ad set name: 20260714 - EXP - TBF - Exp 17 Ginther Character Batch - Expert Healer (Overlooked)

Variant map:

• 20260714 - EXP - TBF - Exp 17 Control Ginther Character Batch - Expert Healer (Overlooked) — Joan Ginther New Leads V1 winner (source ad 120244929239730539) re-uploaded verbatim
• 20260714 - EXP - TBF - Exp 17 V1 Earl Mail Carrier Ginther Character Batch - Expert Healer (Overlooked) — retired mail carrier, Dayton OH; four scratch-off wins, same gas station; tests formula minus the PhD credibility layer
• 20260714 - EXP - TBF - Exp 17 V2 Rosa Small Jackpots Ginther Character Batch - Expert Healer (Overlooked) — retired school-cafeteria worker, San Antonio; five modest jackpots $20k–$180k; tests modest-scale wins vs $20M spectacle
• 20260714 - EXP - TBF - Exp 17 V3 Walter Protection Luck Ginther Character Batch - Expert Healer (Overlooked) — night-shift hospital janitor, Pittsburgh; survived three fatal-odds accidents then $1M scratch-off; tests luck-domain transfer (protection luck)
• 20260714 - EXP - TBF - Exp 17 V4 Marge Bingo Ginther Character Batch - Expert Healer (Overlooked) — diner waitress 40 yrs, Buffalo; seven bingo jackpots in 14 months then $500k scratch-off; tests peer-proximity of the luck venue
• 20260714 - EXP - TBF - Exp 17 V5 Hazel Reno Ginther Character Batch - Expert Healer (Overlooked) — retired hotel housekeeper, Reno; five slot hand-pays in two years, banned from two casinos; tests the sharpest Overlooked emotional hook (luck passing her by daily)

Delivery risk: Lane-A batch launches as ONE fresh CBO adset (all six ads, no incumbent history — fresh-pair doctrine generalized). Control is a re-uploaded creative with a new ad ID, so no arm starts with an allocation advantage. Any arm with lifetime spend < $50 is NO-DELIVERY per Meta-rejection doctrine. If ALL five challengers are NO-DELIVERY, the "Joan is a unicorn" conclusion stays at Early-signal confidence until one fresh re-run confirms.

Date: 2026-07-14

Avatar: The Overlooked

Creative type: First Person Contemporary VSL / Customer testimonial — SoulGem ugc_talking_head, 4:5 (≥120s aspect policy), 720p (matched toward Control's delivery resolution; validity over cost, exp 016 precedent)

Source observation: The Ginther franchise is fatiguing — flagship ad 120241268246290539 at $17.3k 90d spend has decayed to ~$112/14d — and the entire Overlooked avatar rides on one character. Exps 006/007 proved the researcher proof block and odds framing are load-bearing; 002 identified the character formula (uncelebrated peer + population-level proof + modest promise). V7–V12 failed while VIOLATING that formula (celebrated characters, comeback promises, N=1 proof); the corrected batch has never been run.

Hypothesis: The Joan formula — uncelebrated working-class protagonist + escalating improbable-luck sequence with astronomical-odds framing + population-level researcher proof + modest "enough to stop worrying" promise — generates deliverable Overlooked winners independent of the Joan Ginther character.

Novel variable: Protagonist character only (identity, occupation, luck domain, win sequence, low point). Hook syntax is locked to the proven shape — "In [year], I won $[amount] in [thing]. Yes, you heard that right. $[amount]. I was [age], a [ordinary job]…" — with only character facts swapped (V3 Walter opens on his late $1M win then rewinds, per Edward's hook concern at checkpoint 1a).

Hold constant: Script skeleton beat-for-beat (modest win → escalating repeats with odds figures → late-life low point → combined odds + expert quote → young-researcher population survey finding the bracelet pattern with a Drepung subset → unpublished-finding credibility beat → researcher's-mother small-windfall transfer proof → Tenzin/copper-chi mechanism → 50/month full-moon scarcity → "let fortune find you" closer + CTA); ~870-word length; same audience template (cloned from adset 120240341554580539), campaign 120237768514430539 (20260222 - TBF | Overlooked - Testing), and identical copy across all six ads from the canonical Sheet row "Expert Healer (Overlooked)": destination https://discover.karmaitems.com/tbf-1206-healer-overlooked-advertorial-ev-001/, headline "Start Attracting Unexpected Money in 2 Weeks", Sheet primary text, CTA learn_more. All five challengers share identical production (SoulGem ugc_talking_head, 720p, 4:5, same captions/music policy). Pre-registered deviation: Control's creative is the original 1:1 editor render (453s) while challengers are 4:5 SoulGem renders — ugc_talking_head does not offer 1:1; uniform across challengers and folded into the production confound already noted in Decision logic.

Decision logic:

• Any challenger is top spender on the latest day with ≥$50 lifetime → formula CONFIRMED; the character factory opens — future Overlooked scripts are written from the formula, not as Joan edits.
• Control top spender AND ≥2 challengers deliver ≥$50 → formula produces viable-but-weaker peers; Joan stays anchor; the challenger ranking picks which characters to iterate.
• All five challengers < $50 lifetime → NO-DELIVERY across the batch; "Joan is a unicorn" logged at Early-signal confidence only (see Delivery risk).
• Confound note (pre-registered): Control carries editor-grade production vs the challengers' SoulGem renders, so challenger-vs-challenger ranking is the PRIMARY read; challenger-vs-Control is secondary and production-confounded.

Transferable rule: If this resolves, future TBF Overlooked ads will be built differently: the 002/006/007 formula becomes a character factory — new protagonists generated per the formula, applied to every future Overlooked script from AdMaker and AdChain instead of iterating the Joan asset.

Why it matters: The flagship is decaying and every production-lever test on it (008, 009, 015-Ginther) is starving for delivery. Whether the Overlooked pipeline gets a bench of successor characters or doubles down on defending one fatiguing asset is the single biggest fork in the product's next quarter of creative spend.

Spend contract: Fresh CBO adset in campaign 20260222 - TBF | Overlooked - Testing, no adset budget, daily_min_spend_target $30/day (replicates the approved 14.5/16 contract).

Production route: AdChain experiment mode — SoulGem ugc_talking_head renders for V1–V5 (new registry characters), Control re-uploaded from the verified local render /root/.hermes/output/ad-experiments/tbf-joan-ginther-1p3x-redo-new-leads-20260702/control_original.mp4 (New Leads V1). Credit estimate ~12.6k at 720p; balance 30,719 verified 2026-07-14.

Result: V4 Marge — top spend $32.20 on 2026-07-23 (latest day). Total adset spend $289.00 over 9 days. Ranking: V4 $32.20 latest / $146.53 total; V3 $1.89 latest / $90.51 total; V2 $0.00 latest / $32.91 total NO-DELIVERY (<$50 lifetime); Control $0.00 latest / $13.08 total NO-DELIVERY (<$50 lifetime); V5 $0.00 latest / $3.13 total NO-DELIVERY (<$50 lifetime); V1 $0.00 latest / $2.84 total NO-DELIVERY (<$50 lifetime). Winner called (auto recall 2026-07-24) — learnings pending.

Element tested: Position of the protection-luck survival arc inside the Walter script skeleton — is earned-luck the hook (first seconds) or the pre-win body?

Status: Complete

Family: ginther-vsl

Parent: Exp 017

Result vs decision logic: Branch 1 fired cleanly. V1 (Survival First) was the top spender on the latest day and cleared the $50 lifetime threshold. Control was NO-DELIVERY at $29.45. The survival arc is hook-load-bearing: open on the near-death event, not the win. Exp 017 "lead with the win first" rule is now scoped to money-luck characters only.

Learnings:

• [ad] For protection-luck characters, open on the near-death event, not the win. Walter's scaffolding collapse is the scroll-stopper; the ticket is the payoff. Leading with the win first — the Exp 017 rule — is specific to money-luck characters and loses audience attention when the luck domain is survival.
• [ad] The survival-first structure earns dramatically more audience attention than win-first in a clean head-to-head with identical production and the same character identity. The audience stays because the near-death event is inherently more gripping than a lottery announcement.
• [ad] When the character factory produces a new protection-luck protagonist, hard-code the opening shape: near-death event first, ticket payoff second. Do not re-derive this per script — it is now a locked structural brief for the Overlooked avatar.
• Transfer: Cheapest test: apply the survival-first opening to the next protection-luck character the factory produces. Verify audience retention in one fresh-pair against a win-first Control of the same character before baking it into the formula brief.

Ad set name: 20260719 - EXP - TBF - Exp 21 Walter Opening Beat - Expert Healer (Overlooked)

Live ad set name: 20260719 - EXP - TBF - Exp 21 Walter Opening Beat - Expert Healer (Overlooked)

Ad set ID: 120246726299300539

Variant map:

• 20260719 - EXP - TBF - Exp 21 Control Walter Opening Beat - Expert Healer (Overlooked) — Walter (017 V3 winner) verbatim re-upload; opens "In 2022, I won $1 million… But to understand that ticket, you have to know about the three times I should have died."
• 20260719 - EXP - TBF - Exp 21 V1 Survival First Walter Opening Beat - Expert Healer (Overlooked) — cold open on the 1994 scaffolding collapse, survival chronology runs straight, 2022 ticket arrives as the payoff; from the ticket sentence onward word-for-word identical to Control

Delivery risk: Fresh-pair doctrine — Control-clone + V1 launch as a fresh CBO adset pair, never injected into the live 017 adset (120246559151220539) where Walter is incumbent. Control is a re-uploaded creative with a new ad ID, so neither arm starts with an allocation advantage. Any arm with lifetime spend < $50 is NO-DELIVERY per Meta-rejection doctrine; if V1 is NO-DELIVERY, "win-first is hook-load-bearing" stays at Early-signal confidence until one fresh-pair re-run confirms.

Result: V1 — top spend $6.32 on 2026-07-25 (latest day). Total adset spend $99.57 over 4 days. Ranking: V1 $6.32 latest / $70.12 total; Control $4.42 latest / $29.45 total NO-DELIVERY (<$50 lifetime). Auto-called per 4d/$50 doctrine.

Date: 2026-07-19

Avatar: The Overlooked

Creative type: First Person Contemporary VSL / Customer testimonial — SoulGem ugc_talking_head, 4:5 (≥120s aspect policy), 720p (matched to 017 production, validity over cost)

Source observation: Exp 017 made Walter (protection-luck) the new Overlooked benchmark — top spend $19.97 latest / $56.28 total while Joan Control and four other formula characters starved. 017 proved WHICH character; it cannot say WHERE in the skeleton the survival arc does its work. Walter's Control opens on the win then rewinds to the three near-death escapes; the 017 learning "lead with the win first then rewind" was written for money-luck characters and may not hold for protection-luck, where the near-death event is the natural scroll-stopper.

Hypothesis: Exp 017 proved protection-luck (earned through survival) is the strongest Overlooked luck domain. This test resolves whether the survival arc is hook-load-bearing (V1 wins: open on the near-death, ticket is the payoff) or body-load-bearing (Control wins: win-first-then-rewind stays the correct shape even for protection-luck).

Novel variable: Order of the opening beat only. Control opens on the 2022 $1M win with a one-line forward-reference to the three escapes; V1 opens on the 1994 scaffolding collapse and runs the survival chronology straight to the 2022 ticket as payoff. Same story, same facts, same words from the ticket sentence onward — only the position of the first ~150 words changes.

Hold constant: Walter identity and all facts (night-shift hospital janitor, Pittsburgh, 1994 scaffolding / 2008 Route 28 pile-up / 2019 house fire, 2022 $1M scratch-off); the full post-ticket skeleton word-for-word (1-in-30-quadrillion actuary + "luckiest man in Pennsylvania" quote → Daniel Reyes population survey of 31 impossible-luck cases, 19 copper bracelets, 9 Drepung → unpublished-finding credibility beat → researcher's-mother $2,200 windfall transfer proof → Tenzin copper-chi mechanism → 50/month full-moon scarcity → "let fortune find you" closer + CTA); ~870-word band; same audience template (cloned from adset 120240341554580539), campaign 120237768514430539 (20260222 - TBF | Overlooked - Testing), and identical copy from the canonical Sheet row "Expert Healer (Overlooked)": destination https://discover.karmaitems.com/tbf-1206-healer-overlooked-advertorial-ev-001/, headline "Start Attracting Unexpected Money in 2 Weeks", Sheet primary text, CTA learn_more. Identical production on both arms (SoulGem ugc_talking_head, 720p, 4:5, same captions/music policy, same Walter character render settings) — removes the 017 editor-vs-SoulGem production confound by rendering Control fresh in SoulGem too.

Decision logic:

• V1 top spender on the latest day with ≥$50 lifetime → survival arc is hook-load-bearing; future protection-luck scripts open on the near-death event, not the win. The 017 "lead with the win first" learning is scoped to money-luck characters only.
• Control top spender with ≥$50 lifetime → win-first-then-rewind holds for protection-luck too; the survival arc works as pre-win body (earned-luck build), not as the scroll-stopper. The 017 hook learning generalizes across luck domains.
• Both arms < $50 lifetime → NO-DELIVERY across the pair; the 017 Walter signal itself drops to Early-signal confidence and gets one fresh-pair re-run before any structural conclusion is logged.

Transferable rule: If this resolves, future TBF Overlooked VSLs will be built differently: the position of the earned-luck beat in the skeleton becomes a fixed structural brief — applied to every new protection-luck formula character the factory generates, instead of re-deriving the opening shape per script.

Why it matters: The character factory (017) will keep producing protection-luck protagonists — the single highest-leverage structural question for every one of those future scripts is whether the near-death event leads or follows the win. One two-arm test settles it before the next character batch is briefed.

Spend contract: Fresh CBO adset in campaign 120237768514430539 (20260222 - TBF | Overlooked - Testing), no adset budget. AdLauncher fixed CBO floor: 2000¢.

Production route: AdChain experiment mode — SoulGem ugc_talking_head 720p 4:5 renders for BOTH arms (Control re-rendered in SoulGem to kill the 017 production confound; V1 from the survival-first script). Checkpoint 2: locked scripts. Checkpoint 3: pre-launch — nothing goes live on Meta without Edward's go.

Element tested: Protagonist identity/story inside the Joan Ginther Character Factory skeleton translated to The Cursed avatar — does the factory (Exp 017 confirmed on The Overlooked) open across TBF avatars?

Status: Complete

Result vs decision logic: Branch 2 fired: Control (Joan Ginther lineage) led with $127.49 lifetime, above the $50 bar, while every new Cursed character (Nadine, Frank, Dolores, Ray, Connie) stayed under $50 and was rejected. Branch 4 (per-arm NO-DELIVERY) fired for the five challenger arms. Branch 1 (top challenger above $50 → cross-avatar factory confirmed) did not fire — no challenger cleared the bar. Branch 3 (all arms below $50) did not fire because Control cleared it. The character factory that minted a winner on The Overlooked did not generalize to The Cursed: Joan travels, new Cursed characters do not generalize.

Learnings:

• [avatar] Keep the Joan Ginther lineage character as the protagonist when porting the factory to The Cursed — the credentialed four-win figure carries, while freshly minted everyman winners (hairdresser, electrician, nurse’s aide, forklift operator, postal worker) never got traction.
• [family] Treat the character factory as character-specific, not skeleton-specific: the 14-beat skeleton is portable, but its power to mint NEW characters does not transfer across avatars, so reuse the avatar’s proven protagonist rather than regenerating one.
• Transfer: Cheapest test: pit Joan against one new credentialed lineage protagonist (not an everyman) on The Cursed — if the new credentialed figure delivers, the factory’s lever is character type; if it starves, The Cursed rejects new protagonists outright and Joan is the fixed asset.

Family: ginther-vsl

Parent: Exp 017

Result: Control — top spend $0.03 on 2026-08-27 (latest day). Total adset spend $156.73 over 21 days. Ranking: Control $0.03 latest / $127.49 total; V2 $0.00 latest / $23.66 total NO-DELIVERY (<$50 lifetime); V4 $0.00 latest / $4.10 total NO-DELIVERY (<$50 lifetime); V1 $0.00 latest / $0.96 total NO-DELIVERY (<$50 lifetime); V3 $0.00 latest / $0.45 total NO-DELIVERY (<$50 lifetime); V5 $0.00 latest / $0.07 total NO-DELIVERY (<$50 lifetime). Auto-called per 4d/$50 doctrine.

Date: 2026-07-24

Avatar: The Cursed

Creative type: First Person Contemporary VSL / Customer testimonial

Hook Style: Customer Testimonial

Move: Translate / Lane A

Production: SoulGem ugc_talking_head, 720p, 4:5 — all six arms independent renders

Ad set name: 20260724 - EXP - TBF - Exp 30 Cursed Character Factory - 17 Jetsetters (Cursed)

Live ad set name: 20260724 - EXP - TBF - Exp 30 Cursed Character Factory - 17 Jetsetters (Cursed)

Ad set ID: 120246961343750539

Live ad IDs:

• Control: 120246961348270539, 120246961348780539
• V1: 120246961349720539, 120246961350620539
• V2: 120246961354400539, 120246961358850539
• V3: 120246961359550539, 120246961361060539
• V4: 120246961362030539, 120246961362690539
• V5: 120246961363400539, 120246961364450539

AdLauncher request ID: produced-tbf-exp030-cursed-character-factory-20260724-9c54b6ed

AdLauncher queue state: live (publisher proof; register auto-synced)

AdLauncher request SHA-256: c3a8edb09690704f101feb0e9b0d448b0c3e8cbc52906aff608c0db3b068011f

Campaign ID: 120237762972300539

Campaign: 20260222 - TBF | Curses - Testing

Canonical adset/ad variant map (junglelab_names.py validated):

| Arm | File | Protagonist | Word Count | Ad name | |---|---|---|---|---| | Control | scripts/control.txt | Joan Ginther lineage (PhD, Stanford, Texas) — four wins, $20.4M | 945 | 20260724 - EXP - TBF - Exp 30 Control Joan Ginther Cursed Character Factory - 17 Jetsetters (Cursed) | | V1 | scripts/v1-nadine-envy-target.txt | Nadine, retired Cleveland hairdresser — 3 same-store jackpots | 899 | 20260724 - EXP - TBF - Exp 30 V1 Nadine Envy Target Cursed Character Factory - 17 Jetsetters (Cursed) | | V2 | scripts/v2-frank-workplace-curse.txt | Frank, Detroit union electrician — $2.1M Powerball | 913 | 20260724 - EXP - TBF - Exp 30 V2 Frank Workplace Curse Cursed Character Factory - 17 Jetsetters (Cursed) | | V3 | scripts/v3-dolores-generational-drain.txt | Dolores, retired rural-Georgia nurse's aide — five $10k scratch-offs/month | 940 | 20260724 - EXP - TBF - Exp 30 V3 Dolores Generational Drain Cursed Character Factory - 17 Jetsetters (Cursed) | | V4 | scripts/v4-ray-draining-friend.txt | Ray, Fresno forklift operator — $47k Tachi Palace | 947 | 20260724 - EXP - TBF - Exp 30 V4 Ray Draining Friend Cursed Character Factory - 17 Jetsetters (Cursed) | | V5 | scripts/v5-connie-near-miss.txt | Connie, retired Tucson postal worker — $180k Megabucks gift ticket | 944 | 20260724 - EXP - TBF - Exp 30 V5 Connie Near Miss Cursed Character Factory - 17 Jetsetters (Cursed) |

Hypothesis: Joan skeleton generates deliverable characters across TBF avatars. Exp 017 confirmed the character factory on The Overlooked (V4 Marge winner). This test translates the factory to The Cursed avatar using the same 14-beat skeleton with only protagonist identity/story as the variable.

Novel variable: Protagonist identity/story only (character concept). Each arm has a different character, location, win type, and envy-source diagnosis.

Hold constant:

• All 14 Character Factory skeleton beats (hook → escalating improbable sequence → "It wasn't that I couldn't appreciate…" → big win with odds → Daniel Reyes researcher → bracelet noticed → Drepung origin → 31/19/9 subset → unpublished finding → mother proof $2,200 → Tenzin mechanism → copper conductor → scarcity 50/month full moon → close "let fortune find you")
• Cursed mechanism (hidden envy → Root Chakra wound bleeds fortune → copper seals the wound)
• Canonical Sheet copy/destination (17 Jetsetters (Cursed))
• Audience/campaign (TBF Curses Testing campaign 120237762972300539)
• Production settings (SoulGem ugc_talking_head 720p 4:5 for all six)
• AdLauncher CBO floor (2000 cents fixed)

Decision logic:

• Latest-day top challenger above $50 → Cross-avatar factory CONFIRMED; character factory opens across avatars.
• Control leader above $50 → Joan travels but new Cursed characters do not yet generalize.
• All arms below $50 → Early/no-delivery; preserve per-arm scoped NO-DELIVERY.
• Per-arm rule: Each arm with lifetime spend < $50 is NO-DELIVERY per Meta-rejection doctrine; no re-run of starved arm.

Policy blocker (RETIRED): The stale policy_review / launch_blocked_policy_rewrite gate introduced before commit 638191aa9 is retired for active ad workflows. Historical evidence only: output/adchain/tbf-character-factory-cursed-exp030/historical/policy_review_pre_638191aa9.json (non-authoritative). Active package no longer requires policy review, policy rewrite, or reapproval. Edward approved the exact six scripts ("approve. produce, queue") and later said "fix this and adchain to queue," which authorizes paid missing-character create + six SoulGem ugc_talking_head 720p 4:5 renders + canonical AdLauncher enqueue after parent verify_preproduction.py exits 0. No media and no enqueue yet in the preproduction package.

Character prerequisites: Control reuses registered Joan Ginther. Create-at-execution (missing): Nadine - The Retired Cleveland Hairdresser, Frank Hargrove - The Detroit Union Electrician, Dolores - The Rural Georgia Nurses Aide, Ray - The Fresno Forklift Operator, Connie - The Retired Tucson Postal Worker. Specs under character-specs/ (dry-run only until execution).

Execution: python3 /home/runner/.hermes/output/adchain/tbf-character-factory-cursed-exp030/execute_exp030.py (after parent verifiers). Batch: tbf-exp030-cursed-character-factory-20260724. CBO floor 2000¢. No Meta release/publish bypass.

Preproduction package: /home/runner/.hermes/output/adchain/tbf-character-factory-cursed-exp030/

Element tested: Spoken delivery imperfections (two natural disfluencies) in the winning Marge Bingo ugc_talking_head testimonial — whether harmless ASR/TTS-style stumbles increase authenticity.

Status: Complete

Result vs decision logic: Branch 2 (Control disfluent wins) fired cleanly with Control crossing the $50 threshold at $79.04 while V1 was starved. Branch 3 (starved-arm NO-DELIVERY) also fired for V1, but this does not weaken the primary finding — the starved arm confirms the disfluent winner rather than creating ambiguity. The hypothesis is confirmed: natural spoken imperfections are load-bearing authenticity signals.

Learnings:

• [style: Customer UGC] Do not clean up harmless spoken disfluencies in AI talking-head testimonials. When Marge said 'six six' instead of 'sixty-six' and stumbled on 'Saint Casimir's,' those imperfections were load-bearing — the platform directed nearly all delivery to the disfluent original while ignoring the clean-pronunciation twin. The imperfection IS the authenticity.
• [style: Customer UGC] When re-rendering or iterating on an AI testimonial that won, preserve any natural-seeming stumbles or pronunciation quirks. A perfectly clean delivery reads as synthetic. The winning Marge had two minor disfluencies; removing them made her uncanny enough that the platform ignored the variant outright.
• [style: Customer UGC] For future AI testimonials, intentionally introduce one or two controlled, harmless disfluencies (a slight stumble, a self-correction, a natural pause-filler) rather than scripting a flawless delivery. The imperfection signals real person to the platform.
• Transfer: Take an existing AI testimonial with clean delivery and introduce one controlled disfluency; run a two-arm head-to-head to confirm transferability beyond the Marge/Overlooked pair. Cheapest test: apply to the next winning AI testimonial in any avatar, same campaign, fresh ad set.

Family: ginther-vsl

Parent: Exp 017

Launch status: Queued 2026-07-26 — request produced-tbf-exp034-marge-spoken-disfluency-20260726-69c15fa3 state=queued request_hash e3a1a396e029db4394f6834a1e78c442731fc3b992e663ab3e3ebc458c737898

Queue directory: /home/runner/.hermes/output/adlauncher-queue/produced-tbf-exp034-marge-spoken-disfluency-20260726-69c15fa3

Batch artifacts: /root/.hermes/output/meta-uploads/tbf-exp034-marge-spoken-disfluency-20260726/

Hedra generation (V1 seg0): 85e9bfd6-d7c9-4705-abfb-7a4a059fc82d

Altitude: Rung 6 — delivery / verisimilitude micro-lever on a proven winner

Ad set name: 20260726 - EXP - TBF - Exp 34 Marge Spoken Disfluency - Expert Healer (Overlooked)

Live ad set name: 20260726 - EXP - TBF - Exp 34 Marge Spoken Disfluency - Expert Healer (Overlooked)

Ad set ID: 120246880731950539

AdLauncher request ID: produced-tbf-exp034-marge-spoken-disfluency-20260726-69c15fa3

AdLauncher queue state: live (publisher proof; register auto-synced)

AdLauncher request SHA-256: e3a1a396e029db4394f6834a1e78c442731fc3b992e663ab3e3ebc458c737898

Campaign ID: 120237768514430539

Campaign: 20260222 - TBF | Overlooked - Testing

Destination URL: https://discover.karmaitems.com/tbf-1206-healer-overlooked-advertorial-ev-001/

Destination shorthand: Expert Healer (Overlooked)

Avatar: The Overlooked

Variant map:

• 20260726 - EXP - TBF - Exp 34 Control Marge Spoken Disfluency - Expert Healer (Overlooked) — exact live Exp 17 V4 Marge Bingo winner captioned bytes reused verbatim (no re-render): /root/.hermes/output/adchain/tbf-exp017-ginther-character-batch/renders/v4-marge/output_captioned.mp4 sha256 cd718c9261b9ed5e652c5a5ff762e235acc1991da3b61bf71b53b195c5a57141
• 20260726 - EXP - TBF - Exp 34 V1 Marge Spoken Disfluency - Expert Healer (Overlooked) — same Marge portrait/voice/ugc_talking_head 720p 4:5 lineage; only opening segment regenerated with clean pronunciation (sixty-six; Saint Casimir's Parrish Hall); Control segments 1–3 preserved byte-for-byte; captions retain Control wording including I was 66 years old and Casimir's Parrish Hall

Result: Control — top spend $20.39 on 2026-07-31 (latest day). Total adset spend $86.95 over 4 days. Ranking: Control $20.39 latest / $79.04 total; V1 $0.00 latest / $7.91 total NO-DELIVERY (<$50 lifetime). Auto-called per 4d/$50 doctrine.

Date: 2026-07-26

Avatar: The Overlooked

Creative type: First Person Contemporary VSL / Customer testimonial — SoulGem ugc_talking_head, 4:5, 720p

Source observation: Exact ad 20260714 - EXP - TBF - Exp 17 V4 Marge Bingo Ginther Character Batch - Expert Healer (Overlooked) won Exp 17 (allocation + lifetime). Hypothesis from operator review: Marge may win partly because two spoken imperfections increase verisimilitude.

Hypothesis: Harmless natural disfluencies in AI talking-head testimonials increase authenticity enough to improve Meta allocation vs a clean-pronunciation twin that holds every other load-bearing element constant.

Novel variable: Spoken delivery cleanliness on two opening phrases only (I was six six / sixty-six years old; sa-Saint / Saint Casimir's Parrish Hall). Captions held to Control visible wording (66, Parrish).

Hold constant: Marge identity, portrait lineage, voice, SoulGem ugc_talking_head, 720p, 4:5; script body beyond the two pronunciation directives; segments 1–3 of the Control render byte-for-byte on V1; canonical Sheet copy/destination for Expert Healer (Overlooked); fresh CBO adset floor 2000¢; CTA learn_more; age-free advantage targeting cloned from Overlooked template lineage.

Decision logic:

• V1 (clean) beats Control with ≥$50 lifetime → clean pronunciation wins; strip harmless disfluencies on future AI testimonials.
• Control (disfluent winner bytes) beats V1 with ≥$50 lifetime → natural imperfections are load-bearing authenticity; keep/intentionally preserve mild disfluencies.
• Either arm < $50 lifetime → NO-DELIVERY per Meta-rejection doctrine; no causal claim on the starved arm.
• Evidence threshold $50 cumulative is decision logic only — not a daily budget.

Transferable rule: Harmless natural disfluencies can/cannot increase authenticity in AI talking-head testimonials (resolved by this two-arm head-to-head).

Delivery risk: ONE fresh CBO adset, both arms, no incumbent history. Control re-enters under a fresh ad ID so neither arm inherits allocation advantage. $50 cumulative threshold is evidence-only.

Spend contract: Fresh CBO adset in campaign 20260222 - TBF | Overlooked - Testing (120237768514430539), canonical AdLauncher floor expected_daily_min_spend_target_cents=2000 per adset (not per arm). Evidence threshold $50 cumulative only.

Production route: AdChain experiment mode — Control exact reuse (no re-render). V1 segment-0-only SoulGem CLI provider work (Hedra generation 85e9bfd6-d7c9-4705-abfb-7a4a059fc82d) + local restitch/caption rebuild; segments 1–3 byte-identical to Control. Queue via AdLauncher producer_type=adchain, finalizer_type=manual_receipt, no Trello. Run dir: /root/.hermes/output/adchain/tbf-exp034-marge-spoken-disfluency-20260726/. Batch: /root/.hermes/output/meta-uploads/tbf-exp034-marge-spoken-disfluency-20260726/.

Scripts: /root/.hermes/output/adchain/tbf-exp034-marge-spoken-disfluency-20260726/scripts/ (control sha 433ff596… exact Exp17 v4-marge; v1 sha 88882817… = two pronunciation directives only)

Element tested: Position of the protection-luck survival arc inside a NEW factory character's skeleton — confirms the Exp 021 survival-first rule on a second protection-luck protagonist (Dolores) before it is baked into the character-factory formula brief.

Status: Complete

Result vs decision logic: Branch 2 fired cleanly: Control (win-first opening) was top spender. This falsifies the generalization hypothesis — the Exp 021 survival-first result for Walter does NOT apply to Dolores. Both arms delivered, so this was a real head-to-head. The opening beat is a per-character decision, not a character-class law.

Learnings:

• [family] The survival-first opening is NOT a universal protection-luck rule. It worked for Walter McCathie (Exp 021) but lost to the win-first opening on Dolores. Do not hard-code a single opening shape into the factory brief — every new protection-luck protagonist needs their own opening-beat pair.
• [ad] For Dolores: lead with the ticket win and the forward-reference to "three times I should have died," then rewind through the escapes. The win-first hook carried both arms and won the head-to-head.
• [avatar] Dolores is a viable second protection-luck character for the Overlooked roster. The character identity (67-year-old cafeteria worker, Dayton, church-van crash) connects — add Dolores to the character rotation.
• [family] When creating new protection-luck characters, always test opening-beat order with a fresh pair. Never assume the prior character winner transfers — the Exp 021 Walter result stands but is Walter-scoped.
• Transfer: Cheapest test: the next protection-luck character opening-beat pair, one fresh CBO adset. The transfer is Overlooked-internal.

Family: ginther-vsl

Parent: Exp 021

Ad set name: 20260730 - EXP - TBF - Exp 38 Dolores Opening Beat Confirm - Expert Healer (Overlooked)

Live ad set name: 20260730 - EXP - TBF - Exp 38 Dolores Opening Beat Confirm - Expert Healer (Overlooked)

Ad set ID: 120246961561630539

AdLauncher request ID: produced-tbf-exp038-dolores-opening-beat-20260730-aa60388c

AdLauncher queue state: live (publisher proof; register auto-synced)

AdLauncher request SHA-256: 2f94936506f33a03ecc05960db23cd66e923ebec70f60d48cbd8f2c946611135

Variant map:

• 20260730 - EXP - TBF - Exp 38 Control Dolores Opening Beat Confirm - Expert Healer (Overlooked) — win-first: opens "In 2023, I won a million dollars on a scratch-off ticket. But to understand that ticket, you have to know about the three times I should have died." then rewinds through the three escapes; 947 words. Script: output/junglelab/2026-07-29-overlooked-protection-luck-2/control.txt sha256 eda9708a0f57fec4125e433095b67e75b54b000b12b7193d992190b0c4223226
• 20260730 - EXP - TBF - Exp 38 V1 Survival First Dolores Opening Beat Confirm - Expert Healer (Overlooked) — survival-first: cold open on the 1987 black-ice crash, chronology runs straight, ticket arrives as payoff; 884 words. Script: output/junglelab/2026-07-29-overlooked-protection-luck-2/v1-survival-first.txt sha256 f08473e54317c6edb11ee523e90dd30f49cdd6552fe41d9937ac4adee3c40adf
• Byte-identity: the only arm-unique text is Control's 63-word opening paragraph; from "Seventeen years after that crash," to EOF the arms are byte-identical (624 words), verified programmatically.

Delivery risk: Fresh-pair doctrine — Control + V1 launch as a fresh CBO adset pair in the Overlooked Testing campaign, never injected into a live adset. Both arms are brand-new creatives on a brand-new character, so neither starts with creative-history advantage. Any arm with lifetime spend < $50 is NO-DELIVERY per Meta-rejection doctrine; if BOTH arms starve, the Dolores character itself is the failure signal (factory formula miss), not the opening order.

Result: Control — top spend $20.54 on 2026-08-04 (latest day). Total adset spend $116.22 over 5 days. Ranking: Control $20.54 latest / $59.80 total; V1 $4.89 latest / $56.42 total. Auto-called per 4d/$50 doctrine.

Date: 2026-07-30

Avatar: The Overlooked

Campaign ID: 120237768514430539

Destination shorthand: Expert Healer (Overlooked)

Creative type: First Person Contemporary VSL / Customer testimonial — SoulGem ugc_talking_head, 4:5 (≥120s aspect policy), 720p (matched to 017/021 production)

Source observation: Exp 021 called V1 Survival First the winner (V1 $70.12 lifetime vs Control $29.45 NO-DELIVERY, Branch 1 per 4d/$50 doctrine) and its Transfer line pre-registered exactly this confirm: apply the survival-first opening to the next protection-luck character the factory produces, one fresh pair against a win-first Control of the same character, before hard-coding the opening shape into the Overlooked factory brief. Because 021's Control starved, the verdict rests on Meta's allocation, not a both-arms-delivered head-to-head — this pair separates "survival-first is a character-class law" from "Walter-specific quirk."

Hypothesis: Exp 021 proved the survival arc is hook-load-bearing for Walter. This test resolves whether that is a protection-luck character-class rule (V1 wins again on Dolores → lock the factory brief) or Walter-specific (Control wins → the 021 result does not generalize; opening shape stays per-character).

Novel variable: Order of the opening beat only. Control opens on the 2023 $1M win with the "three times I should have died" forward-reference then rewinds; V1 cold-opens on the 1987 black-ice crash and runs the survival chronology straight to the ticket as payoff. Identical words otherwise — the arms share every sentence except Control's 63-word opener; block order differs only as the swap requires.

Hold constant: Dolores identity and all facts (67, retired school-cafeteria worker, Dayton OH; 1986 Ruth/Drepung bracelet provenance worn daily since — acquired BEFORE the first escape per the Walter causal formula; 1987 church-van black-ice crash, 2004 carbon-monoxide leak/Biscuit, 2018 tornado; 2023 $1M scratch-off on Wayne Avenue); the full post-ticket skeleton word-for-word (1-in-30-quadrillion actuary + "luckiest woman in Ohio" → Daniel Reyes survey 31 cases / 19 copper / 9 Drepung with Dolores inside the pattern → unpublished-finding beat → researcher's-mother $2,200 windfall → Tenzin copper-chi conductor mechanism with after-60 channel closing → 50/month full-moon scarcity → "let fortune find you" closer + CTA); ~880–950 word band; spoken disfluency held constant in BOTH arms (operator decision 2026-07-30, Exp 034 assumed winner: "sixty— sixty-seven" in the intro and "Dre— Drepung" at first mention, both in shared text; captions display clean "sixty-seven"/"Drepung" per the 034 convention); identical production on both arms (SoulGem ugc_talking_head, 720p, 4:5, same captions/music policy, same Dolores character render settings); same audience template lineage as 021 (cloned from adset 120240341554580539), campaign 120237768514430539 (20260222 - TBF | Overlooked - Testing); canonical Sheet row "Expert Healer (Overlooked)": destination https://discover.karmaitems.com/tbf-1206-healer-overlooked-advertorial-ev-001/, headline "Start Attracting Unexpected Money in 2 Weeks", Sheet primary text, CTA learn_more.

Decision logic:

• V1 top spender with ≥$50 lifetime → survival-first confirmed as the protection-luck character-class rule; hard-code "near-death event first, ticket payoff second" into the Overlooked factory brief for every future protection-luck protagonist.
• Control top spender with ≥$50 lifetime → the 021 result does not generalize; survival-first stays Walter-scoped and the factory brief keeps opening shape as a per-character decision requiring one pair per new protection-luck character.
• Both arms < $50 lifetime → NO-DELIVERY across the pair; the failure signal is the Dolores character (factory formula miss on this identity), not the opening order — the survival-first rule stays at Exp 021 confidence and the next protection-luck character re-runs this confirm.

Transferable rule: If this resolves, future TBF Overlooked VSLs will be built differently: the character factory brief gains a locked structural rule — every protection-luck protagonist opens on the near-death event with the ticket as payoff — applied to all future factory output without per-script re-derivation.

Why it matters: The factory keeps producing protection-luck protagonists and the avatar currently rides on Walter the way it used to ride on Joan. This pair both hardens the highest-leverage structural rule for all of those future scripts AND debuts the roster's second protection-luck character as fatigue insurance.

Spend contract: Fresh CBO adset in campaign 120237768514430539 (20260222 - TBF | Overlooked - Testing), no adset budget. AdLauncher fixed CBO floor: 2000¢.

Production route: AdChain experiment mode — new Dolores SoulGem character create, then ugc_talking_head 720p 4:5 renders for BOTH arms from the SHA-locked scripts (disfluent spoken takes per the two directives, captions clean). Checkpoint 2: locked scripts (cleared at log time — SHAs above). Checkpoint 3: pre-launch — nothing goes live on Meta without Edward's go.

Element tested: Which game of chance carries the Overlooked luck-streak story — the repeated game inside the proven Marge skeleton, isolated on one character.

Status: Complete

Result vs decision logic: Branch 5 fired: V3 (slots) was the only arm to clear the $50 lifetime threshold and it led the latest day, so the pre-registered 'gambling venue/spectacle carries it' branch is the verdict. Control (bingo), V1 (lottery), V2 (free drawings), V4 (raffle), V5 (pull-tabs), and V6 (OTB) all starved under $50 lifetime, so branches 1-4 and 6 did not fire. Branch 7 (Meta policy note) did not fire: V3 delivered and won rather than being policy-rejected, so the gambling-adjacent win is a delivery signal, not a policy artefact. Margin caveat: V3 cleared the gate narrowly ($52.96 lifetime, runner-up V2 $36.79) and the latest-day lead was modest, so treat this as a directional allocation read.

Learnings:

• [avatar] Write The Overlooked's seven-win streak as a solitary, private machine win (nickel-and-quarter slots) rather than a communal or charitable venue - the machine framing is what carried, and the audience's idea of 'lucky' here reads personal and mechanical, not social.
• [family] Treat the game of chance as a load-bearing copy decision in the ginther-vsl family, not window dressing - swapping bingo for slots flipped the result, so choose the game deliberately and let it carry part of the story's meaning instead of defaulting to a safe communal setting.
• [avatar] Revisit the dignity constraint from Exp 002 - the modest, everyday machine-gambling story out-carried the respectable communal one, so don't assume a 'dignified' community win is the right frame for this avatar; a humble machine ritual may fit the audience's self-image better.
• Transfer: Cheapest test: re-express the solitary machine-win (slots) game in a second TBF avatar (Cursed) as a one-variant static - the cheapest way to learn whether the machine-luck rule widens beyond The Overlooked before promoting it product-wide.

Family: ginther-vsl

Parent: Exp 017

Ad set name: 20260730 - EXP - TBF - Exp 39 Game Of Chance - Expert Healer (Overlooked)

Live ad set name: 20260730 - P - EXP - TBF - Exp 39 Game Of Chance - Expert Healer (Overlooked)

Ad set ID: 120247326421030539

Live ad IDs:

• Control: 120247326425840539, 120247326428220539
• V1: 120247326430420539, 120247326431930539
• V2: 120247326433580539, 120247326435090539
• V3: 120247326437560539, 120247326439970539
• V4: 120247326443770539, 120247326445820539
• V5: 120247326448450539, 120247326449830539
• V6: 120247326452120539, 120247326456180539

AdLauncher request ID: produced-tbf-exp039-marge-game-of-chance-recovery-20260811-cb4edfca

AdLauncher queue state: live (publisher proof; register auto-synced)

AdLauncher request SHA-256: 5bdbb606045b79b8cd45fced027405eb7f8fbf68ac98e1b3639f7e55cc6bf123

Avatar: The Overlooked

Campaign ID: 120237768514430539

Destination shorthand: Expert Healer (Overlooked)

Creative type: First Person Contemporary VSL / Customer testimonial — SoulGem ugc_talking_head, 4:5, 720p (matched to 017 production; Control is the 017 V4 Marge live creative re-uploaded)

Variant map:

• 20260730 - EXP - TBF - Exp 39 Control Game Of Chance - Expert Healer (Overlooked) — bingo at St. Casimir's parish hall (017 V4 Marge winner verbatim). control.txt sha256 433ff596fa3169739af582506ee9043ef237abcfac376128af08595607fc6d97 — byte-identical to output/adchain/tbf-exp017-ginther-character-batch/scripts/v4-marge.txt; live video re-uploaded, no re-render
• 20260730 - EXP - TBF - Exp 39 V1 Lottery Draw Game Of Chance - Expert Healer (Overlooked) — state lottery numbers draw, same four numbers weekly (family birthdays). v1-lottery-draw.txt sha256 1f48a6b617321406…
• 20260730 - EXP - TBF - Exp 39 V2 Free Drawings Game Of Chance - Expert Healer (Overlooked) — free-entry store drawings / radio call-in / mail-in, no stake. v2-free-drawings.txt sha256 d8dacc689ae76201…
• 20260730 - EXP - TBF - Exp 39 V3 Slot Hand Pays Game Of Chance - Expert Healer (Overlooked) — monthly riverboat outing with her sister, nickel/quarter machines. v3-slot-handpays.txt sha256 ad569a22d8a3460e…
• 20260730 - EXP - TBF - Exp 39 V4 Church Raffle Game Of Chance - Expert Healer (Overlooked) — parish festival raffle books; parish context HELD, mechanic swapped. v4-church-raffle.txt sha256 c441d0a5f2135bfc…
• 20260730 - EXP - TBF - Exp 39 V5 Legion Pull Tabs Game Of Chance - Expert Healer (Overlooked) — dollar pull-tabs at the Legion bar. v5-legion-pulltabs.txt sha256 3e8e19e1c9cdfa4f…
• 20260730 - EXP - TBF - Exp 39 V6 OTB Horse Racing Game Of Chance - Expert Healer (Overlooked) — two-dollar OTB bets, long shots picked by name. v6-otb-horse-racing.txt sha256 2757183b8288a2f7…
• Scripts and builder: output/junglelab/2026-07-30-marge-game-of-chance/ (build_variants.py constructs each arm from Control via declared edits; verify.py asserts byte-identity outside them — 121 checks pass)

Delivery risk: Seven arms in one fresh CBO adset, by design. Starvation is the measurement, not a failure mode: the allocation winner is the verdict and a starved arm is a Meta rejection (Exp 017 doctrine). The CBO floor is per-adset, so seven arms cost the same as two and buy full coverage of the game space in one auction. Pre-registered per arm: lifetime spend < $50 = NO-DELIVERY = Meta rejection of that game. Expected outcome is 1–2 delivering arms; a full ranking among rejected arms is NOT expected and is not required for the verdict. Control is a re-upload under a new ad ID so it carries no ad-level history advantage (same treatment as Exp 021's Control, which starved).

Source observation: Exp 017 confounded character with game: Earl played scratch-offs, Rosa modest jackpots, Hazel slot hand-pays, Marge bingo, Joan the lottery — each game arrived attached to a different character, so Marge's win cannot be attributed to bingo rather than to Marge. Four of the six arms were money-luck variants, and four starved. The game of chance has therefore never been isolated on a single character.

Hypothesis: Marge won Exp 017 because of who she is, not what she plays — or because bingo's communal, small-stake, peer-proximate venue is itself load-bearing. Holding Marge and the entire skeleton constant and swapping only the repeated game resolves which.

Novel variable: The repeated game of chance that produces the seven-win streak. Nothing else moves. The arms span the stake/venue axis: no stake (V2), small communal commercial (Control bingo, V5 pull-tabs), communal charitable (V4 raffle), solitary purchased with personal ritual (V1), solitary machine gambling (V3), and communal with a skill-suspicion overlay (V6).

Hold constant: Marge's identity and every biographical fact (66, Buffalo diner waitress, forty years on the morning shift, her sister, the math teacher who eats at the diner, "I've balanced a tray for forty years. I believe in what I can carry."); the streak SHAPE (seven wins in fourteen months, first win $4,000, twenty years of playing without winning beforehand, 2021–2022); the 2023 $500,000 scratch-off payoff verbatim in every arm; the odds beat ("one in 6 trillion", Buffalo State professor) held constant on purpose so proof magnitude cannot vary; the entire back half byte-identical (Daniel Reyes interview → notices the copper bracelet → Stan the bus driver, Drepung, 2020 → 31/19/9 survey → unpublished finding → mother's $2,200 windfall → Tenzin copper-chi conductor → fifty bracelets a month on the full moon → closer + CTA); production (SoulGem ugc_talking_head, 720p, 4:5, same Marge portrait/voice/caption policy); audience template lineage, campaign 120237768514430539, and canonical Sheet copy for Expert Healer (Overlooked) — destination https://discover.karmaitems.com/tbf-1206-healer-overlooked-advertorial-ev-001/, headline "Start Attracting Unexpected Money in 2 Weeks", Sheet primary text, CTA learn_more.

Declared-edit discipline: Each challenger is built from Control by an explicit ordered list of sentence-level swaps (9–13 per arm, ≤5% word delta; all arms 906–925 words vs Control 893). verify.py asserts: byte-exact rebuild from Control, zero undeclared new sentences, zero silently dropped Control sentences, payoff/odds/back-half byte-identity, and an explicit literal allowlist for the two permitted back-half diction swaps. Two documented exceptions: (1) "arthritis" is preserved verbatim in every arm despite the copy gate's disease-term ban — it sits in held-constant text and byte-identity outranks gate-tidiness; verify.py asserts its presence so it cannot be silently tidied. (2) "jackpot" names the mechanic and therefore belongs to the tested variable, so it is swapped to game-neutral wording in V2/V4/V6 only (games that do not pay jackpots) and retained in Control/V1/V3/V5; Control is untouched, preserving the free re-upload of the proven creative.

Decision logic:

• Control (bingo) top spender with ≥$50 lifetime → the communal small-stake parish venue is load-bearing; keep building bingo-like characters and treat venue as part of the formula.
• V1 (lottery draw) wins → ritual and universality beat venue; every future factory character can be a lottery winner, the cheapest story to write and the audience's default mental model.
• V2 (free drawings) wins → the appeal is chance itself without gambling; strip gambling framing from the whole roster and lower fleet-wide policy risk.
• V4 (church raffle) wins over Control → the parish CONTEXT rather than the bingo MECHANIC is what works (V4 holds St. Casimir's and the parish ladies while swapping only the mechanic) — the cleanest single comparison in the batch.
• V3 (slots) or V5 (pull-tabs) or V6 (OTB) wins → gambling venue/spectacle carries it; revisit the dignity constraint inherited from Exp 002.
• All seven < $50 lifetime → whole-adset rejection; the failure is the ad set (creative or audience), not the game variable, and the read is void rather than negative.
• Meta policy note: if V3 or V6 is rejected on gambling-adjacent policy rather than on delivery, that NO-DELIVERY is a policy artefact and must not be logged as evidence about the game's appeal.

Transferable rule: If this resolves, future TBF Overlooked ads will be built differently: the win-type of every character the factory produces becomes a fixed brief item — applied to all future factory output — instead of being chosen ad hoc per script.

Why it matters: The factory (017) can generate characters indefinitely, but every one needs a game of chance and we have never known which game does the work. One ad set settles the win-type for the entire future roster, and V4-vs-Control additionally separates parish community from the bingo mechanic.

Depth cap note: This is the SECOND isolation on the Marge winning unit (Exp 034 spoken disfluency was the first), which is the doctrine limit. Marge is mapped after this; further decomposition requires a new scaled winner or an explicit operator override.

Sequencing dependency: Exp 034 is Active on Marge in the same CBO campaign. Launch AFTER 034 calls (~2026-07-31) so two near-identical Marge ad sets do not compete for one campaign budget, and so the Control's delivery style (disfluent vs clean) matches the freshest signal.

Spend contract: Fresh CBO adset in campaign 120237768514430539 (20260222 - TBF | Overlooked - Testing), no adset budget. AdLauncher fixed CBO floor: 2000¢ — per adset, not per arm; seven arms do not multiply the floor.

Production route: AdChain experiment mode — SIX renders only (V1–V6, SoulGem ugc_talking_head 720p 4:5, reusing Marge's locked portrait, voice and avatar_prompt.txt from the 017 batch). Control is the 017 V4 Marge live creative re-uploaded verbatim under a new ad ID — no re-render, per the decomposition-control-reuse doctrine. Checkpoint 2: locked scripts (cleared at log time — SHAs above). Checkpoint 3: pre-launch — nothing goes live on Meta without Edward's go.

Production status: Complete 2026-07-30 — six renders + reused Control, all_ready: true, zero hold-constant warnings. All gates pass per arm (ffprobe video+audio+valid moov/duration, caption_sensecheck.ok, caption-fidelity assertions); avatar_prompt.txt byte-identical across all six arms and the 017 original.

Batch artifacts: output/adchain/tbf-exp039-marge-game-of-chance-20260730/ (production_manifest.json, per-arm renders/, repair-backups/, CAPTION_DISFLUENCY_LESSONS.md)

Media (final sha256 / duration):

• Control (reused, no re-render) cd718c9261b9ed5e… / 362.1s — output/adchain/tbf-exp017-ginther-character-batch/renders/v4-marge/output_captioned.mp4
• V1 Lottery Draw 3c559370191fe3a6… / 380.4s
• V2 Free Drawings e4d5fcc33d3699ca… / 368.6s
• V3 Slot Hand Pays 4300744038f0a49f… / 379.7s
• V4 Church Raffle 5b62e4ae68052eed… / 378.6s
• V5 Legion Pull Tabs 7e0198fabe89dfe9… / 382.1s
• V6 OTB Horse Racing 7922a812ae1d6be9… / 365.2s

Spoken-delivery provenance (matters for interpreting this experiment): Delivery is EMERGENT from the locked script orthography on every arm — each arm's TTS input file is byte-identical to its locked script, so approved copy and spoken input are the same bytes. An earlier production pass hand-authored false starts on the age line ("66") and the venue proper nouns to imitate Control's natural disfluencies; that collided with the raw numeral/abbreviation and produced doubled utterances ("six six sixty-six years old") that Control does not have. Caught by Edward on review 2026-07-30 and fully reverted: the opening span of every arm was re-rendered from locked text with zero insertions, and the age line now matches Control's single compact token (0.48–0.58s vs Control 0.46s). Rule recorded in CAPTION_DISFLUENCY_LESSONS.md §0 — replicate a reference render's script orthography, never hand-author a stumble; the collision applies to every raw numeral and abbreviation, and a partial fix just relocates the defect.

Element tested: Visual-channel authenticity imperfection — whether synthetic capture/playback imperfections (handheld motion, sensor artifacts, a buffering freeze) in an AI talking-head testimonial increase authenticity/allocation the way spoken disfluencies do (Exp 034 confirmed the audio channel).

Status: Complete

Result vs decision logic: Branch 1 fired cleanly — V2 Propped Phone is the winning dose. The visual imperfection principle extends the Exp 034 audio finding: moderate propped-phone realism earns allocation over clean renders. The subliminal wobble (V1) was too subtle to trigger authenticity — Meta starved it. The overt treatments (V3 active handheld, V4 buffer freeze) were rejected harder than Control — even clean Control was starved. The sweet spot is moderate, consciously visible phone-style imperfection. Sharp dose-response curve with a narrow winning band. The two-channel authenticity finding (audio disfluencies + visual imperfection) is now closed.

Learnings:

• [family] Apply the propped-phone treatment (6% overscan, ±12px two-axis sway, ±0.5° rotation, mild sensor grain, ±1.5% exposure drift) as the default render finish for AI talking-head testimonials in this family. Moderate visual imperfection at this dose outperforms clean renders.
• [family] Do not use subliminal-only imperfection — the barely-conscious wobble was Meta-starved. The treatment must be consciously visible to work as an authenticity signal; sub-perceptual wobble earns no credit.
• [family] Do not push past propped-phone into active-handheld or glitch territory — both were rejected harder than the clean Control. Visual imperfection has a sharp dose-response curve; the winning band is narrow.
• [family] The visual channel now joins the audio channel as a proven authenticity lever — Exp 034 confirmed spoken disfluencies; Exp 043 confirms visual imperfection. Every future AI testimonial render should carry both disfluencies AND the propped-phone treatment. The two-channel finding is now closed.
• Transfer: Cheapest test: apply the V2 propped-phone plus spoken-disfluency treatment to one TBAB Dimmed UGC testimonial render and run a two-arm test against the current clean render to confirm the two-channel rule transfers across products.

Family: ginther-vsl

Parent: Exp 034

AdLauncher request ID: produced-tbf-exp043-marge-visual-imperfection-20260803-b1b22fd9

AdLauncher queue state: live (publisher proof; register auto-synced)

AdLauncher request SHA-256: 2ce089aa2f8438ae4612253f561ab984b9d2b1061cd7de6422b8ccf5fa9ec35e

Queue directory: output/adlauncher-queue/produced-tbf-exp043-marge-visual-imperfection-20260803-b1b22fd9

Media (final sha256 / duration, verify.py all_pass + congruence gate pass 2026-08-03): Control cd718c9261b9ed5e… / 362.12s (verbatim reuse) · V1 0c3bf5f10e2850b7… · V2 a19d56b5042f7e9f… · V3 120fb1be6bcd548d… · V4 9c05590e0030861f… (all 362.12s, audio md5 identical across arms)

Move: Adjacent territory (sibling authenticity channel) / Lane B

Altitude: Rung 8 — production/render doctrine for the whole AI-testimonial fleet

Ad set name: 20260803 - EXP - TBF - Exp 43 Marge Visual Imperfection - Expert Healer (Overlooked)

Live ad set name: 20260803 - P - EXP - TBF - Exp 43 Marge Visual Imperfection - Expert Healer (Overlooked)

Ad set ID: 120247053475120539

Campaign ID: 120237768514430539

Campaign: 20260222 - TBF | Overlooked - Testing

Destination URL: https://discover.karmaitems.com/tbf-1206-healer-overlooked-advertorial-ev-001/

Destination shorthand: Expert Healer (Overlooked)

Variant map:

• 20260803 - EXP - TBF - Exp 43 Control Marge Visual Imperfection - Expert Healer (Overlooked) — exact live Exp 17 V4 Marge winner captioned bytes reused verbatim (no re-render): output/adchain/tbf-exp017-ginther-character-batch/renders/v4-marge/output_captioned.mp4 sha256 cd718c9261b9ed5e652c5a5ff762e235acc1991da3b61bf71b53b195c5a57141
• 20260803 - EXP - TBF - Exp 43 V1 Subliminal Wobble Marge Visual Imperfection - Expert Healer (Overlooked) — 4% overscan, ±6px two-axis sway + micro-drift, ±0.23° rotation; barely conscious
• 20260803 - EXP - TBF - Exp 43 V2 Propped Phone Marge Visual Imperfection - Expert Healer (Overlooked) — 6% overscan, ±12px sway, ±0.5° rotation, mild sensor grain (noise alls=3), ±1.5% exposure drift
• 20260803 - EXP - TBF - Exp 43 V3 Active Handheld Marge Visual Imperfection - Expert Healer (Overlooked) — 12% overscan, ±30px sway, ~1° rotation, heavier grain (alls=5), ±3% exposure drift
• 20260803 - EXP - TBF - Exp 43 V4 Buffer Freeze Marge Visual Imperfection - Expert Healer (Overlooked) — footage freezes t=6.0→7.2s while audio continues, then catch-up jump (1.2s of footage skipped, lip-sync restored); otherwise untreated. Presented as "V5" in preview; renumbered sequential per Step-5 rule.
• Approved preview clips (25s, Checkpoint 1 record): tools.ambercollective.co/share/tbf-visimp-preview-20260803-{control,v1-minimal,v2-fuller,v3-strong,v5-buffer-freeze}.mp4 (preview v1-minimal→V1, v2-fuller→V2, v3-strong→V3, v5-buffer-freeze→V4)

Delivery risk: Five arms in one fresh CBO adset. Near-identical creatives (identical audio/script) starve by design — per Exp 039 doctrine, starvation is the measurement: the allocation winner is the verdict, a starved arm is a Meta rejection, expected outcome is 1–2 delivering arms. Pre-registered per arm: lifetime spend < $50 = NO-DELIVERY = Meta rejection of that treatment. Control is a re-upload under a new ad ID so it carries no ad-level history advantage.

Result: V2 — top spend $29.64 on 2026-08-08 (latest day). Total adset spend $166.91 over 5 days. Ranking: V2 $29.64 latest / $125.61 total; V1 $4.06 latest / $14.83 total NO-DELIVERY (<$50 lifetime); V3 $0.76 latest / $8.83 total NO-DELIVERY (<$50 lifetime); Control $0.27 latest / $10.74 total NO-DELIVERY (<$50 lifetime); V4 $0.00 latest / $6.90 total NO-DELIVERY (<$50 lifetime). Auto-called per 4d/$50 doctrine.

Date: 2026-08-03

Avatar: The Overlooked

Creative type: First Person Contemporary VSL / Customer testimonial — SoulGem ugc_talking_head lineage, 4:5, 832x1088 720p-class, 362s (matched to 017/034 production)

Source observation: Exp 034 confirmed that harmless spoken disfluencies in the Marge testimonial are load-bearing authenticity signals (disfluent Control $79.04 vs clean twin NO-DELIVERY). The visual channel of the same authenticity thesis has never been tested: every SoulGem render ships tripod-stable, clean-exposure footage.

Hypothesis: The imperfection-equals-authenticity principle extends beyond spoken delivery to the visual channel — synthetic capture imperfections (handheld motion, sensor artifacts) and/or transmission artifacts (a buffering freeze) will earn allocation over the clean-render Control.

Novel variable: Presence, type, and dose of visual imperfection only, spanning the axis: none (Control) → subliminal wobble (V1) → propped-phone realism (V2) → active handheld (V3) → playback glitch (V4).

Hold constant: Script, voice, and audio stream byte-identical across all five arms (treated arms stream-copy the Control audio; md5 2d26e02fa7522c60a7532122cf6394bc); Marge identity/portrait/render lineage (all arms derive from the same Exp 17 V4 master frames — zero new SoulGem/Hedra generation); caption text, timing, and rendering (same captions.ass, same libass burn, captions burned AFTER treatment so they stay screen-stable and audio-synced in every arm — a wobbling burned caption would be a fake-tell and a hidden second variable); duration 362.1s on all arms (V4's freeze is compensated by the catch-up skip); canonical Sheet copy/destination for Expert Healer (Overlooked); campaign 120237768514430539 (20260222 - TBF | Overlooked - Testing); audience template lineage 120240341554580539; CTA learn_more; AdLauncher fixed CBO floor 2000¢.

Decision logic:

• Any treated arm top spender with ≥$50 lifetime → visual imperfection extends the Exp 034 audio finding; the winning dose becomes the fleet render doctrine for AI talking-head testimonials (V1 → subliminal-only treatment; V2/V3 → overt handheld at that dose; V4 → transmission-glitch artifacts, extend doctrine beyond capture to playback).
• Control top spender with ≥$50 lifetime → the authenticity signal is audio-only; keep renders visually clean; imperfection doctrine stays confined to spoken delivery.
• Per-arm: lifetime spend < $50 = NO-DELIVERY per Meta-rejection doctrine; no re-run of a starved arm.
• All five arms < $50 lifetime → whole-adset rejection; read void, not negative.

Transferable rule: If this resolves, future AI talking-head testimonials fleet-wide will be rendered differently: a standard post-render imperfection treatment (or an explicit clean-render rule) becomes part of the production brief for every avatar and product — applied at zero render-credit cost via deterministic ffmpeg post-processing.

Why it matters: Exp 034 turned one accidental imperfection into doctrine; the fleet is already preserving/introducing spoken disfluencies on every new render (Exp 038 convention). Whether the same lever exists in the visual channel decides the render finish of every future testimonial — and the treatment costs nothing to apply.

Depth-cap note — operator override: This is the THIRD isolation on the Marge winning unit (034 disfluency, 039 game-of-chance, 043 visual channel). Edward was told the cap twice and explicitly ordered "Approve. Queue in adlauncher" on the Marge previews 2026-08-03 — recorded here as the doctrine-required explicit operator override. Marge is mapped after this; no further Marge decomposition without a new override.

Sequencing note: Exp 039 (seven Marge arms, same campaign) is Planned/unlaunched with production complete; its own sequencing rule was "launch after 034 calls" (now satisfied). 039 and 043 both re-upload the identical Marge Control into the same CBO campaign — whichever launches second should wait for the first to call so two Marge ad sets do not compete for one campaign budget.

Spend contract: Fresh CBO adset in campaign 120237768514430539 (20260222 - TBF | Overlooked - Testing), no adset budget. AdLauncher fixed CBO floor: 2000¢ — per adset, not per arm.

Production route: Local deterministic post-process (ffmpeg) — NO SoulGem/Hedra generation, zero render credits. Source: Exp 17 V4 Marge uncaptioned master output.mp4; treatment chain per arm as in the Variant map; captions re-burned from captions.ass after treatment; audio stream-copied. Control reused verbatim. Run dir: output/adchain/tbf-exp043-marge-visual-imperfection-20260803/ (render_arms.sh = the exact treatment definitions, verify.py, render.log). Batch: output/meta-uploads/tbf-exp043-marge-visual-imperfection-20260803/. Queue via AdLauncher enqueue_from_manifest (manual/no-Trello route, finalizer_type=manual_receipt), release on AdLauncher's own window.

Element tested: Post-production editing treatment on the proven Exp 017 V4 Marge Bingo testimonial — exact winner Control versus Amoo's edited cut, in one fresh two-arm ad set.

Status: Complete — whole-adset rejection; no directional read

Family: ginther-vsl

Parent: Exp 017

Ad set name: 20260806 - EXP - TBF - Exp 45 Marge Amoo Edit - Expert Healer (Overlooked)

Live ad set name: 20260806 - P - EXP - TBF - Exp 45 Marge Amoo Edit - Expert Healer (Overlooked)

Ad set ID: 120247150764470539

Avatar: The Overlooked

Campaign ID: 120237768514430539

Campaign: 20260222 - TBF | Overlooked - Testing

Template adset: 120246829306190539 — ACTIVE readback verified 2026-08-06

Destination shorthand: Expert Healer (Overlooked)

Destination URL: https://discover.karmaitems.com/tbf-1206-healer-overlooked-advertorial-ev-001/

Frame.io source: https://next.frame.io/project/b99f8398-a80b-4a3e-8020-3a6a0c101625/b81079e0-bd08-4164-baea-4b78863aa1f0

Variant map:

• 20260806 - EXP - TBF - Exp 45 Control Marge Amoo Edit - Expert Healer (Overlooked) — Marge Bingo Ginther — Expert Healer (Overlooked) Control.mp4; exact Exp 017 V4 live-winner bytes; sha256 cd718c9261b9ed5e652c5a5ff762e235acc1991da3b61bf71b53b195c5a57141; 362.124s; 832×1088.
• 20260806 - EXP - TBF - Exp 45 V1 Marge Amoo Edit - Expert Healer (Overlooked) — Marge Bingo Ginther — Expert Healer (Overlooked) — Amoo Edit.mp4; sha256 adf3d3d22bb8254361c454f31773ac5cf38cc09bd74db50d1b668bb3ec5a4b67; 360.000s; 1920×1920.

Hypothesis: Amoo's post-production treatment will earn more Meta allocation than the exact proven Marge Control. If it does, the editor treatment is a transferable finishing improvement for the Marge/Ginther testimonial lineage.

Novel variable: Post-production editing treatment. The spoken script is held constant: local faster-whisper parity found 878 matching words and a 0.968 sequence ratio; differences were numeric formatting and ASR spellings, not a substantive copy change.

Hold constant: Marge identity and story, spoken script, canonical Sheet headline Start Attracting Unexpected Money in 2 Weeks, canonical primary text, destination, CTA learn_more, campaign, audience template lineage, and one fresh CBO ad set. The assets intentionally preserve their delivered production resolution/frame-rate differences because those are part of the Amoo editing treatment.

Decision logic:

• V1 top spender with ≥$50 lifetime → adopt the Amoo edit treatment as the preferred Marge finishing route and test transfer on the next eligible testimonial.
• Control top spender with ≥$50 lifetime → retain the proven original finish; the Amoo edit does not improve allocation.
• Either arm below $50 lifetime → classify that arm NO-DELIVERY per the existing Meta-rejection doctrine.
• Both below $50 lifetime → whole-adset rejection; no directional read.

Result: Whole-adset rejection — Control $41.24 lifetime; V1 Amoo edit $39.24 lifetime; $80.48 total through 2026-08-13. Both arms remained below the registered $50-per-arm evidence bar.

Result vs decision logic: The pre-registered both-below-$50 branch fired. No Control-versus-Amoo winner is called and no finishing-treatment direction is inferred from this ad set.

Spend contract: Fresh CBO adset in campaign 120237768514430539; no adset budget. AdLauncher fixed CBO floor: 2000¢ per adset, not per arm.

Production route: Manual/no-Trello Frame.io intake → exact original downloads staged under output/meta-uploads/tbf-exp045-marge-amoo-edit-20260806/assets/ → live Sheet/product/congruence gates → one AdLauncher request with Control + V1. Intrinsic destination fan-out remains enabled from the canonical Sheet secondary destination.

Queue receipt: Request produced-tbf-exp045-marge-amoo-edit-20260806-519e204f; state queued; request sha256 48d106cfa67c06be31410bcc337f6dcf4edb1267248e99078874853ba4d1c283; verified exactly one matching request, two queue-owned media files with byte-identical hashes, exact names/copy/destination/spend floor, and no pre-existing platform IDs. Batch custody: output/meta-uploads/tbf-exp045-marge-amoo-edit-20260806/.

Element tested: Whether the named-dead substitute in Dolores's first near-death (Carol dies in the front seat Dolores always sat in) is load-bearing, or whether the same church-van wreck works as solo survival.

Status: Complete

Result vs decision logic: Branch 0 fired (Control top spender with lifetime above the floor): the named-dead substitute is load-bearing. Branch 2 fired (Control delivers, V1 starves): V1 is a Meta rejection, never re-run. Branches 1, 3, and 4 did not fire.

Learnings:

• [family] Keep a named person who did not make it in the first near-death of protection-luck scripts — the survival only reads as real when someone else died in their place.
• [family] Stop writing the solo-survival version of the church-van wreck; the named-dead substitute is the load-bearing proof beat, not decorative color.
• Transfer: Cheapest test: write the named-dead substitute into the next Overlooked protection-luck factory character and see whether the beat transfers beyond Dolores.

Family: ginther-vsl

Parent: Exp 038

Altitude: Rung 6 — structure/proof

Result: Control — top spend $4.97 on 2026-08-22 (latest day). Total adset spend $93.09 over 5 days. Ranking: Control $4.97 latest / $53.54 total; V1 $2.85 latest / $39.55 total NO-DELIVERY (<$50 lifetime). Auto-called per 4d/$50 doctrine.

Ad set name: 20260815 - EXP - TBF - Exp 50 Dolores Carol Dead Substitute - Expert Healer (Overlooked)

Live ad set name: 20260815 - P - EXP - TBF - Exp 50 Dolores Carol Dead Substitute - Expert Healer (Overlooked)

Ad set ID: 120247418357870539

Live ad IDs:

• Control: 120247418365630539, 120247418366390539
• V1: 120247418368510539, 120247418369400539

Variant map:

• 20260815 - EXP - TBF - Exp 50 Control Dolores Carol Dead Substitute - Expert Healer (Overlooked) — Exp 38 win-first Dolores Control verbatim, Carol dies in the front seat; 947 words. Script sha256 eda9708a0f57fec4125e433095b67e75b54b000b12b7193d992190b0c4223226. Control video reused from Exp 38 Control (output_captioned.mp4 sha256 5351c4bc208272f2841e5a3207e0dba60e10767d23903ded15b1a00e248d73d5), no re-render.
• 20260815 - EXP - TBF - Exp 50 V1 Solo Survival Dolores Carol Dead Substitute - Expert Healer (Overlooked) — same van, same ice, same Springfield run; nobody dies; she walks away. 912 words. Script sha256 9463d6f4fc68973b83b1698da3f61ae5dd4ff58a0d63bdb9cf9e4c4a0854d038.

Delivery risk: Fresh-pair doctrine — Control + V1 launch as a fresh CBO adset in the Overlooked Testing campaign, never injected into a live adset. Control re-enters under a new ad ID so it carries no creative-history advantage. Any arm under $50 lifetime spend is a Meta rejection — a real verdict against that arm, not an unmeasured rival. If at least one arm delivers, the allocation winner IS the verdict and rejected arms are never re-run. Only if ALL arms starve is the result an early signal requiring one fresh-pair confirmation.

Date: 2026-08-15

Avatar: The Overlooked

Campaign ID: 120237768514430539

Destination shorthand: Expert Healer (Overlooked)

Creative type: First Person Contemporary VSL / Customer testimonial — SoulGem ugc_talking_head, 4:5 (≥120s aspect policy), 720p (matched to 017/021/038 production)

Source observation: Exp 038 proved Dolores as a viable second protection-luck character and locked win-first as her opening. Operator vetoed the ledger's next-test (a third protection-luck character opening-beat pair) and required the follow-up to stay on the Dolores Control. The only beat a viewer actually remembers that is not factory template is Carol dying in the seat Dolores always sat in.

Hypothesis: Exp 038 proved Dolores + win-first delivers. This test resolves whether the named-dead substitute is load-bearing protection-luck proof (Control wins: future protection-luck characters need a person who did not make it) or shock color (V1 wins: the solo wreck is enough; stop writing a corpse into the first escape).

Prospect claim: The Overlooked currently believes luck never lands on people like them; this variant works iff the survival only feels real when someone else died in their place.

Placement: Problem-aware — they have seen lottery testimonials; the church-van near-death is the claim-form under test.

Novel variable: Presence of a named-dead substitute in the first near-death. Control: Carol asks for the front seat, dies, two feet of seat cushion. V1: same wreck, nobody dies, she walks away. Forced consumed sentence: "Carol wasn't lucky." Everything else is byte-identical.

Hold constant: Dolores identity and all other facts (67, retired school-cafeteria worker, Dayton OH; 1986 Ruth/Drepung bracelet worn daily since; 2004 carbon-monoxide / Biscuit; 2018 tornado; 2023 $1M scratch-off on Wayne Avenue); win-first opening with the "three times I should have died" forward-reference; the full post-ticket skeleton word-for-word (actuary + "luckiest woman in Ohio" → Daniel Reyes 31/19/9 → unpublished-finding → mother's $2,200 → Tenzin copper-chi → 50/month full-moon scarcity → "let fortune find you" closer + CTA); ~880–950 word band (Control 947 / V1 912, −3.7%); spoken disfluency in both arms ("sixty— sixty-seven", "Dre— Drepung"; captions clean); identical production (SoulGem ugc_talking_head, 720p, 4:5, same locked Dolores portrait + avatar_prompt + voice EjSQi64BmcbeF3GLTOEC); Control video is the Exp 38 winner bytes; same audience template lineage (cloned from adset 120240341554580539), campaign 120237768514430539; canonical Sheet row "Expert Healer (Overlooked)": destination https://discover.karmaitems.com/tbf-1206-healer-overlooked-advertorial-ev-001/, headline "Start Attracting Unexpected Money in 2 Weeks", Sheet primary text, CTA learn_more.

Decision logic:

• Control top spender with ≥$50 lifetime → the named-dead substitute is load-bearing; future protection-luck factory characters need a person who did not make it in the first escape.
• V1 top spender with ≥$50 lifetime → the solo wreck is enough; stop writing a corpse into the first escape. Carol is shock color, not the engine.
• Control delivers and V1 starves → same as Control win; V1 is a Meta rejection, never re-run.
• V1 delivers and Control starves → same as V1 win; Control is a Meta rejection, never re-run.
• Both arms < $50 lifetime → whole-adset rejection; no directional read. One fresh-pair confirmation only.

Transferable rule: If this resolves, future TBF Overlooked ads will be built differently: protection-luck characters either must or must not carry a named person who died in the protagonist's place — applied to every future factory protection-luck script, not just Dolores.

Why it matters: Dolores just entered the Overlooked rotation. Whether the factory keeps writing a dead friend into the first escape changes every protection-luck script after this one.

Spend contract: Fresh CBO adset in campaign 120237768514430539 (20260222 - TBF | Overlooked - Testing), no adset budget. AdLauncher fixed CBO floor: 2000¢.

Production route: AdChain experiment mode — Control reuses Exp 38 Control output_captioned.mp4 (script sha-identical). V1 is a new matched SoulGem ugc_talking_head 720p 4:5 render from the locked Dolores portrait/voice/avatar_prompt. Manual/no-Trello AdLauncher enqueue. Checkpoint 2: locked scripts (SHAs above). Operator authorized lock/render/enqueue 2026-08-15 ("keep going until its queued").

Scripts: output/adchain/tbf-exp050-dolores-carol-dead-substitute-20260815/scripts/ + output/junglelab/2026-08-15-dolores-carol-dead-substitute/

Builder: output/adchain/tbf-exp050-dolores-carol-dead-substitute-20260815/build_variants.py — Control + two declared spans; V1 rebuild hash-locked.

AdLauncher request ID: produced-tbf-exp050-dolores-carol-dead-substitute-20260815-ae7ec6a4

AdLauncher queue state: live (publisher proof; register auto-synced)

AdLauncher request SHA-256: 4c09d689f4999a76727e344417f120cbb3331ce4b22b4264e2c4cf23c08ad74e

Queue receipt: Request produced-tbf-exp050-dolores-carol-dead-substitute-20260815-ae7ec6a4; state queued; request sha256 4c09d689f4999a76727e344417f120cbb3331ce4b22b4264e2c4cf23c08ad74e. Queue custody contains exactly two media in Control → V1 order: Control sha256 5351c4bc208272f2841e5a3207e0dba60e10767d23903ded15b1a00e248d73d5 (Exp 38 winner bytes reused); V1 sha256 9d18d45e6584378d5a9abfa9e1a8467bf956c7f9fb8d33acdd4f85af6e55a9fa. No platform IDs created; no release forced.

Element tested: Combined production treatment on the proven Exp 017 V4 Marge Bingo testimonial — exact winner Control versus a new Amoo-edited, NEW VOICE ITERATION-labeled cut in one fresh two-arm ad set.

Status: Complete

Result vs decision logic: Branch 1 fired: V1 was top spender on the latest day and cleared the $50 lifetime bar, so the revised combined treatment bundle beats the original Marge finish. Branch 3 fired: Control stayed under $50 lifetime and is classified NO-DELIVERY while V1 delivered. Branch 2 (Control wins) and Branch 4 (both arms starved) did not fire. Because the edit and the voice were changed together, this validates the bundled treatment only — editing and voice remain unseparated, so the next test must isolate them.

Learnings:

• [family] Don't treat the exact original winner bytes as untouchable — a refreshed finish (re-edit plus a new voice read) on the proven Marge testimonial out-earned the untouched original, so refresh the finish of a proven testimonial rather than locking its original cut as sacred.
• [family] This win bundled the edit and the voice together, so it proves a finish refresh works but not which lever carried it — split editing and voice into separate variants next time before promoting either component as a rule.
• Transfer: Cheapest test: isolate the treatment bundle — run the same Marge script with one arm changing only the edit (original voice kept) and one changing only the voice (original edit kept) to learn which component carries the lift before promoting either to doctrine.

Family: ginther-vsl

Parent: Exp 045

Lineage note: Replication after Exp 045's whole-adset rejection; source winner is Exp 017 V4.

Date: 2026-08-18

Avatar: The Overlooked

Campaign ID: 120237768514430539

Campaign: 20260222 - TBF | Overlooked - Testing

Template adset: 120246829306190539 — ACTIVE readback verified 2026-08-18

Destination shorthand: Expert Healer (Overlooked)

Destination URL: https://discover.karmaitems.com/tbf-1206-healer-overlooked-advertorial-ev-001/

Frame.io source: https://next.frame.io/project/b99f8398-a80b-4a3e-8020-3a6a0c101625/7839508a-23ef-4d89-9337-1ea453a1d5df

Ad set name: 20260818 - EXP - TBF - Exp 53 Marge Amoo Edit New Voice - Expert Healer (Overlooked)

Live ad set name: 20260818 - P - EXP - TBF - Exp 53 Marge Amoo Edit New Voice - Expert Healer (Overlooked)

Ad set ID: 120247475712540539

Live ad IDs:

• Control: 120247475715140539, 120247475716390539
• V1: 120247475717900539, 120247475718770539

Variant map:

• 20260818 - EXP - TBF - Exp 53 Control Marge Amoo Edit New Voice - Expert Healer (Overlooked) — NEW VOICE ITERATION Marge Bingo Ginther — Expert Healer (Overlooked) - Control.mp4; exact Exp 017 V4 live-winner bytes; sha256 cd718c9261b9ed5e652c5a5ff762e235acc1991da3b61bf71b53b195c5a57141; 362.124s; 832×1088.
• 20260818 - EXP - TBF - Exp 53 V1 Marge Amoo Edit New Voice - Expert Healer (Overlooked) — NEW VOICE ITERATION Marge Bingo Ginther — Expert Healer (Overlooked) - EDIT.mp4; new Frame.io challenger; sha256 ae9c9d53a82ba2dedc28285422837de3d714eede75f883a942a5b65261a5bbdd; 360.067s; 1920×1920.

Hypothesis: The new combined Amoo-edit/new-voice treatment will earn more Meta allocation than the exact proven Marge Control. Exp 045 could not resolve the original Amoo cut because both arms remained below $50; this fresh challenger tests whether the revised treatment bundle can break that dead heat.

Experiment classification: Replication with a revised challenger, not a novel editing-only isolation. No existing Planned/Queued/Live/Complete experiment contains this challenger asset hash, but Exp 045 tested the same treatment family with a different Amoo cut.

Novel variable: Combined production-treatment bundle: Amoo post-production editing plus the source folder's NEW VOICE ITERATION treatment. These deltas are deliberately bundled in the delivered asset and are not separately identifiable.

Causal limit: A V1 win would validate the delivered treatment bundle only. It cannot be attributed to editing or voice separately. A follow-up isolation would be required before promoting either component as doctrine.

Hold constant: Marge identity and story; spoken script (local faster-whisper parity: 875 matching words, 0.966 sequence ratio; differences are numeric formatting/minor ASR spellings rather than a substantive copy rewrite); exact Exp 017 winner bytes as Control; canonical Sheet headline, primary text, destination, and CTA learn_more; campaign; audience template lineage; one fresh CBO ad set. The assets preserve their delivered resolution/frame-rate/audio differences because those are part of the combined treatment.

Decision logic:

• V1 top spender with ≥$50 lifetime → the revised combined treatment bundle beats the original finish; test editing versus voice separately before adopting either component.
• Control top spender with ≥$50 lifetime → retain the proven original Marge finish; do not promote the combined treatment.
• One arm below $50 lifetime while the other delivers → classify the starved arm NO-DELIVERY and call the delivered allocation winner.
• Both arms below $50 lifetime → whole-adset rejection; no directional read and no third Marge Amoo replication without a new explicit override.

Spend contract: Fresh CBO adset in campaign 120237768514430539; no adset budget. AdLauncher fixed CBO floor: 2000¢ per adset, not per arm.

Production route: Manual/no-Trello Frame.io intake → exact original downloads staged under output/meta-uploads/tbf-exp053-marge-amoo-edit-new-voice-20260818/assets/ → live Sheet/product/congruence gates → one AdLauncher request with Control + V1. Normal AdLauncher release cadence; no forced release.

AdLauncher request ID: produced-tbf-exp053-marge-amoo-edit-new-voice-20260818-9eb3923c

AdLauncher queue state: live (publisher proof; register auto-synced)

AdLauncher request SHA-256: 091243ca25942f509e4dc547d0bd6923e2a0dd1c552bba2034c2cdb21de7ea4f

Queue receipt: Request produced-tbf-exp053-marge-amoo-edit-new-voice-20260818-9eb3923c; state queued; request sha256 091243ca25942f509e4dc547d0bd6923e2a0dd1c552bba2034c2cdb21de7ea4f. Queue custody contains exactly two media in Control → V1 order: Control sha256 cd718c9261b9ed5e652c5a5ff762e235acc1991da3b61bf71b53b195c5a57141; V1 sha256 ae9c9d53a82ba2dedc28285422837de3d714eede75f883a942a5b65261a5bbdd. Strict post-enqueue readback passed every check; no platform IDs were created and no release was forced.

Result: V1 — top spend $16.27 on 2026-08-25 (latest day). Total adset spend $107.99 over 5 days. Ranking: V1 $16.27 latest / $70.85 total; Control $5.63 latest / $37.14 total NO-DELIVERY (<$50 lifetime). Auto-called per 4d/$50 doctrine.

Element tested: Whether the named-dead substitute that proved load-bearing for Dolores transfers to Walter's survival-first scaffolding opening.

Status: Complete

Result vs decision logic: Branch 1 (V1 top spender ≥ $50 → named-dead transfers, promote family-wide): not fired — Control held top allocation. Branch 2 (Control top spender ≥ $50 → named-dead is Dolores-specific, do not promote): fired — Control held top allocation across the window. Branch 3 (one arm below $50 while the other delivers → delivered winner is the verdict): not fired — both arms crossed $50 lifetime, so no arm was a Meta rejection. Branch 4 (both arms below $50 → whole-adset rejection / early signal): not fired — both arms delivered.

Learnings:

• [ad] For Walter, keep the survival-first opening — his solo scaffolding escape already carries the proof; the named-dead substitute does not earn its place in his story.
• [family] The named-dead beat did not transfer from Dolores to Walter — hold it as a Dolores-specific proof beat rather than promoting it to the protection-luck family brief.
• [avatar] The Overlooked's protection-luck scaffolding converts on the survival story itself — the death attribution is not the load-bearing element here.
• Transfer: Cheapest test: run the named-dead beat on a protection-framed opening for a third Overlooked character to separate the proof frame from Dolores's identity — the surviving hypothesis is that the beat is protection-frame-specific, not character-specific.

Ad set name: 20260826 - EXP - TBF - Exp 58 Walter Named Dead Transfer - Expert Healer (Overlooked)

Live ad set name: 20260826 - P - EXP - TBF - Exp 58 Walter Named Dead Transfer - Expert Healer (Overlooked)

Ad set ID: 120247575551280539

Live ad IDs:

• Control: 120247575552150539
• V1: 120247575552690539

Variant map:

• 20260826 - EXP - TBF - Exp 58 Control Walter Named Dead Transfer - Expert Healer (Overlooked) — exact live Exp 21 V1 Walter survival-first script.
• 20260826 - EXP - TBF - Exp 58 V1 Named Dead Walter Named Dead Transfer - Expert Healer (Overlooked) — Eddie Walsh occupies Walter's exact spot and dies in the 1994 collapse.

Delivery risk: Fresh two-arm CBO adset. Both arms receive freshly produced, matched opening performances in one session; neither uses the old Exp 21 final video as its complete media. The held-constant body is shared identically. Any arm under $50 lifetime spend is a Meta rejection. If at least one arm delivers, the allocation winner is the verdict and rejected arms are never re-run. Only if both arms starve is the result an early signal requiring one fresh-pair confirmation.

Family: ginther-vsl

Parent: Exp 050

Altitude: Rung 6 — structure/proof

Result: Control — top spend $37.80 on 2026-09-03 (latest day). Total adset spend $237.91 over 7 days. Ranking: Control $37.80 latest / $132.60 total; V1 $7.19 latest / $105.31 total. Auto-called per 4d/$50 doctrine.

Date: 2026-08-26

Avatar: The Overlooked

Campaign ID: 120237768514430539

Campaign: 20260222 - TBF | Overlooked - Testing

Template lineage: Exp 21 adset 120246726299300539; fresh ACTIVE targeting readback required at packaging.

Destination shorthand: Expert Healer (Overlooked)

Destination URL: https://discover.karmaitems.com/tbf-1206-healer-overlooked-advertorial-ev-001/

Creative type: First Person Contemporary VSL; Dramatic Incident; SoulGem ugc_talking_head; Walter; 720p; 4:5.

Source observation: Exp 50 proved that Carol dying in Dolores's usual seat was load-bearing proof: the solo-survival challenger starved. Exp 21 supplies the cleanest transfer character because Walter's survival-first opening already won, and Walter has only one prior isolation.

Hypothesis: The named-dead substitute is a transferable protection-luck proof beat, not a Dolores-specific story detail. If V1 wins, future protection-luck characters need a named person who died in the protagonist's exact place. If Control wins, the beat stays scoped to Dolores's church-van story.

Prospect claim: The Overlooked prospect believes survival can be dismissed as coincidence; the named-dead beat works iff a concrete person occupying Walter's exact place makes the counterfactual feel real.

Placement: Problem-aware, inside a mature protection-luck VSL family.

Novel variable: Presence of one named-dead substitute in the first near-death. Control: three men are hospitalized and Walter walks away. V1: coworker Eddie Walsh takes Walter's exact spot and dies. The coupled seam changes from “I didn't make anything of it” to “I didn't think that made me special” solely to keep Walter emotionally coherent after a coworker's death.

Hold constant: Walter identity; 1994 demolition-site side job; ninety-second water-jug escape; 2008 Route 28 crash; 2019 fire; 2022 million-dollar ticket; all proof, mechanism, offer, and CTA copy; 897-word source skeleton; campaign, audience, destination, headline, primary text, CTA, aspect ratio, Walter identity, voice, framing, captions, music, and playback treatment. Every byte from “For 14 years, nothing else extraordinary happened” onward is identical.

Primary test variants:

• Control — exact approved live-source script at control.txt; 897 words; sha256 2b59ecb81ac6ee92ce71f34150c5e79c41059c66f9ff75cfc75f861b4190fa22.
• V1 — Named Dead — approved challenger at v1-named-dead.txt; 918 words; sha256 6f4c469152b798e4fca2cf24a24605332e5faf64215569c1a71831c02b1553f7.

Decision logic:

• V1 top spender with at least $50 lifetime → the named-dead substitute transfers beyond Dolores; promote it to the protection-luck family brief.
• Control top spender with at least $50 lifetime → the named-dead beat is Dolores-specific; retain Walter's solo scaffolding escape and do not promote the rule family-wide.
• One arm below $50 while the other delivers → the delivered allocation winner is the verdict; the starved arm is a Meta rejection and is never re-run.
• Both arms below $50 → whole-adset rejection; treat as an early signal and run one fresh-pair confirmation before changing doctrine.

Spend contract: Fresh CBO adset in campaign 120237768514430539; no adset budget. AdLauncher fixed CBO floor: 2000¢ per adset, not per arm.

Production route: AdChain experiment mode. Generate one clean new opening performance for each arm in the same session with identical Walter/voice/shot settings; do not sub-clip or Frankenstein the tested opening. Reuse one identical held-constant body assembly in both arms. Re-transcribe both openings, every edit seam, and the tail before render acceptance. Script and render production are complete; the separate Meta-live approval was received and consumed by the AdLauncher enqueue.

Render receipt: Produced 2026-08-26 with Walter voice EWBLTPf2WxpVUmgO1MDK, the locked Exp 17 portrait/prompt, 720p Hedra requests, 832×1088 normalized output, TikTok captions, 1.0× playback, and no added imperfection treatment. Control: 298.201s, sha256 affc6fc74e0489ae3410c9f03e2c5dc64bcaf1baa43428cfe031c34baf77a24b, Hedra job 102dc5c0-f303-4707-89f2-89d8eecd4fb7. V1: 305.080s, sha256 7cb5f733cee3a27fbc009771a6f4b4ca3a459044587dfb5f5a677e3355405b7b, Hedra job ea2c93b9-474e-49de-bfa0-5505e825ef21. Both use shared body artifact sha256 a2e2095f0a30adf11b86c3daba06180e25a50cfb1cc3f0cae75702b7b2cbc00f; caption sense-check, faststart, stream, opening, seam, anchor, and tail validation passed.

AdLauncher request ID: produced-tbf-exp058-walter-named-dead-transfer-20260826-2519ba95

AdLauncher queue state: live (publisher proof; register auto-synced)

AdLauncher request SHA-256: 77e8ed5af53b3561de7e4d0a73aa288369f29bda7f6d912be625bb14b3ab825d

Queue receipt: Exactly two queue-owned videos in Control → V1 order. Fresh Sheet row A7:Z7, product identity, ad↔page congruence, live Exp 21 template targeting/page/pixel, primary-only destination approval, strict request, source→staged→queue custody, and 2000¢ CBO-floor checks passed. No release was forced and no Meta entity exists yet.

Why it matters: This is the cheapest direct transfer test of Exp 50. It determines whether the named-dead substitute becomes a reusable construction rule for every future Overlooked protection-luck character or remains local color in Dolores's story.

Element tested: Speaker-footage realism on the Joan Ginther Overlooked UGC — the current CONTROL cut versus the same video with more-realistic / less-AI presenter footage (NEW AVATAR), in one fresh two-arm ad set.

Status: Complete

Result vs decision logic: Branch 1 fired: V1 (less-AI presenter footage) led delivery and crossed the lifetime threshold, so the realism pass improves performance and should be applied to the next eligible talking-head VSL before promotion to doctrine. Branch 2 did not fire: Control did not lead. Branch 3 fired: V1 delivered while Control stayed under the lifetime threshold, so Control is NO-DELIVERY and V1 is the delivered winner. Branch 4 did not fire: at least one arm crossed the threshold, so this is not a whole-adset rejection.

Learnings:

• [family] Apply a less-AI realism pass to the presenter's footage before shipping the next cut in this VSL family — a more natural-looking speaker holds the audience where an obviously-AI presenter loses it.
• Transfer: Cheapest test: apply the same less-AI realism pass to the next talking-head VSL in a different family in one new render, to learn whether the finishing step transfers beyond ginther-vsl.

Family: ginther-vsl

Lineage note: Control bytes are the live ITER V5 render from 20260812 - TBF - 17 Jetsetters (Overlooked) - Joan Ginther UGC - New Control Lead V5 - New Visuals (Amoo) (queue produced-daily_drain_20260812_6a684029c6b73677478a2ada-6ea7f8d7, completed). Re-uploaded under a fresh ad ID in this experiment's own ad set. That ITER is provenance only, not a completion or launch gate.

Altitude: Rung 8 — production polish

Date: 2026-08-30

Avatar: The Overlooked

Campaign ID: 120237768514430539

Campaign: 20260222 - TBF | Overlooked - Testing

Template adset: 120246829306190539 — re-verify ACTIVE readback at launch time (Exp 65 verified 2026-08-29)

Destination shorthand: Expert Healer (Overlooked)

Destination URL: https://discover.karmaitems.com/tbf-1206-healer-overlooked-advertorial-ev-001/

Frame.io source: https://next.frame.io/project/b99f8398-a80b-4a3e-8020-3a6a0c101625/view/0a6621e7-9f14-4989-8967-0d51c10d2747 (folder 0dd777e7-4548-429a-b813-f6fdda937215)

Ad set name: 20260830 - EXP - TBF - Exp 69 Ginther Speaker Realism - Expert Healer (Overlooked)

Live ad set name: 20260830 - P - EXP - TBF - Exp 69 Ginther Speaker Realism - Expert Healer (Overlooked)

Ad set ID: 120247684867290539

Live ad IDs:

• Control: 120247684869100539, 120247684869780539
• V1: 120247684870360539, 120247684871080539

Variant map:

• 20260830 - EXP - TBF - Exp 69 Control Ginther Speaker Realism - Expert Healer (Overlooked) — TBF - 17 Jetsetters (Overlooked) - Joan Ginther UGC - New Control Lead V5-New Visuals V4 CONTROL.mp4; sha256 b79dbb35ec2ab5c048122350a434d2d45c56cafe3051e5abde79150fb0d558ae; 394.733s; 1920×1920 30fps.
• 20260830 - EXP - TBF - Exp 69 V1 Ginther Speaker Realism - Expert Healer (Overlooked) — TBF - 17 Jetsetters (Overlooked) - Joan Ginther UGC - New Control Lead V5-New Visuals V4 NEW AVATAR.mp4; sha256 92aa7461d672a8188a98deaa2554bab37a397df890557cc887f20f9db50e2cce; 394.733s; 1920×1920 30fps.

Hypothesis: Improving the realism of the speaker's footage so it looks less AI earns more Meta allocation than the current CONTROL speaker footage. If V1 wins, less-AI presenter footage is a transferable finishing step for this VSL family.

Experiment classification: Single-variable visual isolation within the ginther-vsl family. Only the picture of the speaker changes; B-roll, captions, script, and audio stay the same.

Novel variable: Speaker-footage realism only — CONTROL presenter picture versus the NEW AVATAR less-AI restyle. Proof of isolation: ffmpeg stream-level MD5s show both arms' audio streams are bit-identical (1d4e54c6dc13f59eaad1b6079b009137) while the video streams differ (qa/audio_parity.json).

Causal limit: A result attributes to the picture treatment that makes the speaker look less AI. It cannot separate face identity from texture or grade if those changed together in the NEW AVATAR export.

Hold constant: Audio (voice, script, music — bit-identical stream), character and story, duration (394.733s), resolution/frame rate (1920×1920, 30fps), canonical Sheet headline Start Attracting Unexpected Money in 2 Weeks, canonical Sheet primary text (sha256 0e07253aab1c4572a82ff700b25f1bdbb29166ef36b895c5d563346826911b9b, length 1339), destination, CTA learn_more, campaign, audience template lineage, one fresh CBO ad set.

Decision logic:

• V1 top spender with ≥$50 lifetime → less-AI speaker footage improves allocation; apply the realism pass to the next eligible talking-head VSL before promoting it to doctrine.
• Control top spender with ≥$50 lifetime → the current speaker footage stands; the realism pass does not improve allocation.
• One arm below $50 lifetime while the other delivers → classify the starved arm NO-DELIVERY and call the delivered allocation winner.
• Both arms below $50 lifetime → whole-adset rejection; no directional read.

Spend contract: Fresh CBO adset in campaign 120237768514430539; no adset budget. AdLauncher fixed CBO floor: 2000¢ per adset, not per arm.

Production route: Manual/no-Trello Frame.io intake → exact original downloads staged under output/meta-uploads/tbf-exp069-ginther-speaker-realism-20260830/assets/ → live Sheet/product/congruence gates → one AdLauncher request with Control + V1. Normal AdLauncher release cadence; no forced release.

AdLauncher request ID: produced-tbf-exp069-ginther-speaker-realism-20260830-bd5b6968

AdLauncher queue state: live (publisher proof; register auto-synced)

AdLauncher request SHA-256: 45ab24a86f0a7fa95f003cfd4171008cc77c69b08b367972cadb08b038f4638c

Queue receipt: Request produced-tbf-exp069-ginther-speaker-realism-20260830-bd5b6968; state queued; request sha256 45ab24a86f0a7fa95f003cfd4171008cc77c69b08b367972cadb08b038f4638c. Queue custody contains exactly two media in Control → V1 order. No platform IDs were created and no release was forced.

Result: V1 — top spend $17.84 on 2026-09-08 (latest day). Total adset spend $108.07 over 4 days. Ranking: V1 $17.84 latest / $66.07 total; Control $10.49 latest / $42.00 total NO-DELIVERY (<$50 lifetime). Auto-called per 4d/$50 doctrine.

Element tested: Editing style of the video for The Overlooked avatar — original Control cut vs the same footage re-edited (1.3x speed pass, tighter assembly/pacing), both arms in one new EXP adset, ranked by Meta allocation.

Status: Complete

Result vs decision logic: Branch 'Control outspends Re-Edit' fired cleanly. Control dominated allocation vs Re-Edit by 75:1. Original pacing is confirmed as the right editing standard for The Overlooked — the speed-up actively degraded allocation, and Meta starved the re-edit arm completely. The editing lever is real but the direction is toward the original, more deliberate cut.

Learnings:

• [avatar] Keep the original pacing for The Overlooked testimonial footage — the 1.3x speed-up re-edit failed to deliver. Longer, more deliberate timing carries authenticity for this avatar. The tight-edit instinct (faster pacing equals better retention) is wrong here.
• [family] The editing lever is real — a decisive delivery split on identical script and character — but the winning direction is original pacing, not tighter cuts. For this testimonial family, a clean tripod-stable video at natural speed is the delivery leader.
• Transfer: Cheapest test: apply the same 1.3x re-edit pass to one TBAB Dimmed testimonial and run a two-arm edit-style test to see if pacing sensitivity is avatar-specific or universal.

Family: overlooked-twist-editing-style

AdLauncher request ID: produced-tbf-exp042-overlooked-twist-editing-style-20260803-71f55cc5

AdLauncher queue state: live (publisher proof; register auto-synced)

AdLauncher request SHA-256: dcf38f9d481c5e589018a71a5fdb6bfd9531f6398dd09e1cdae29008308d82be

AdLauncher submission SHA-256: e170697a37e4d42187f34e1849343063a3cd4c9e5151c7afc044785d4bfc148d

Queue cardinality: 1 adset / 2 video media / 2 ad names, exact Control → V1 Re-Edit order; strict readback verifier 16/16 green.

Ad set name: 20260803 - EXP - TBF - Exp 42 Contemporary Twist Editing Style - Expert Healer (Overlooked)

Live ad set name: 20260803 - P - EXP - TBF - Exp 42 Contemporary Twist Editing Style - Expert Healer (Overlooked)

Ad set ID: 120247053075230539

Avatar: The Overlooked

Campaign ID: 120237768514430539

Campaign: 20260222 - TBF | Overlooked - Testing

Template adset: 120246829306190539 — fresh ACTIVE readback verified 2026-08-03 before dry run

Destination shorthand: Expert Healer (Overlooked)

Live destination: https://discover.karmaitems.com/tbf-1206-healer-overlooked-advertorial-ev-001/

Media: Both arms 1080x1080, 30fps, H264+AAC. Control 347.50s, 464860282 bytes, sha256 c1d697b0d079cd27b056de2cc18e8339ca1fd2382fadb96cc46e0c0e017d89cc. V1 Re-Edit 302.17s, 405036714 bytes, sha256 bea0f0d91d01de81d49c719d9d04f6b816af6f69e92e904781b1450ab2075c22.

Frame.io source: Folder (RE-EDITED) 20260629 - TBF - Overlooked Avatar - First Person Contemporary Twist Variants; project https://next.frame.io/project/b99f8398-a80b-4a3e-8020-3a6a0c101625/d91e5dff-204d-410c-aa71-d75e5e6e71eb

Destination fan-out: Intrinsic destination fan-out stays at its DEFAULT (ENABLED). Secondary destination: https://discover.karmaitems.com/tbf-2607-ovl-advertorial-ev-001/ — shorthand Champa Advertorial (Overlooked). 2 arms x 2 destinations = 4 Meta ads at release.

Batch dir: /home/runner/.hermes/output/meta-uploads/tbf-exp042-overlooked-twist-editing-style-20260803

Variant map:

• 20260803 - EXP - TBF - Exp 42 Control Contemporary Twist Editing Style - Expert Healer (Overlooked) — original cut, 347.50s, sha256 c1d697b0d079cd27b056de2cc18e8339ca1fd2382fadb96cc46e0c0e017d89cc
• 20260803 - EXP - TBF - Exp 42 V1 Re-Edit Contemporary Twist Editing Style - Expert Healer (Overlooked) — same footage re-edited, 1.3x speed pass, tighter assembly/pacing, 302.17s, sha256 bea0f0d91d01de81d49c719d9d04f6b816af6f69e92e904781b1450ab2075c22

Hypothesis: Tighter editing and pacing (1.3x re-edit pass on identical footage and script) changes Meta allocation versus the original cut; if it does, editing style is a real performance lever for this avatar and the re-edit pass should become default finishing.

Novel variable: Editing style of the video only — pacing, assembly, and the 1.3x speed pass.

Hold constant: Spoken script, speaker/character, offer, Sheet-canonical headline Start Attracting Unexpected Money in 2 Weeks, Sheet-canonical primary text (sha256 0e07253aab1c4572a82ff700b25f1bdbb29166ef36b895c5d563346826911b9b, length 1339), destination URL, CTA learn_more, audience/targeting (US, home+recent, advantage audience on), campaign, adset, one adset for both arms, intrinsic destination fan-out at default (ENABLED).

Decision logic:

• Re-Edit outspends Control → tighter editing/pacing is the lever; adopt the 1.3x re-edit pass as default finishing for this avatar.
• Control outspends Re-Edit → original pacing is already right and the speed-up degrades it; keep the original cut as the finishing standard.
• Both deliver comparably → editing style is not a meaningful allocation lever for this footage; direct future effort elsewhere.
• Honest caveat: Control carries no prior creative-history advantage here — both arms are new ad entities in a fresh adset, so allocation differences reflect the edit itself.

Spend contract: Fresh CBO adset in TBF Overlooked Testing, no adset budget. AdLauncher fixed CBO floor: 2000¢ — per adset, not per arm.

Element tested: Visual editing treatment on the ER Nurse First Person Contemporary VSL — the original Control cut versus the same video re-edited to Amoo's editing-chief feedback, in one fresh two-arm ad set.

Status: Complete

Result vs decision logic: Control was the delivered allocation winner. Branch 2 fired: Control was top spender with at least $50 lifetime, so the original cut stands and the editing-chief feedback pass does not improve the finish. Branch 3 also fired: V1 stayed under $50 lifetime and is NO-DELIVERY, with Control as the delivered winner. Branch 1 (V1 top spender) and Branch 4 (both arms under $50) did not fire. The hypothesis — that Amoo acting as editing chief produces a visual edit that earns more allocation than the original Control cut — is falsified.

Learnings:

• [family] Keep the original ER Nurse pacing cut as the finished asset for this footage — a picture re-cut directed by an editing chief did not improve it, so leave the picture alone and stop chasing visual polish.
• [family] With the script and voice held byte-identical and only the picture changing, the re-cut changed the outcome not at all — the next lever on this VSL family is a message or voice change, not a visual edit.
• Transfer: Cheapest test: hold the winning original Control cut and vary one script element (a hook or narrator beat) on the same footage to learn whether the message layer, not the picture, moves the result.

Family: overlooked-twist-editing-style

Parent: Exp 042

Lineage note: Second editing-treatment test on this footage (Exp 042: original pacing beat a 1.3x speed re-edit 75:1). Also third test in the Amoo lineage (Exp 045 → 053), and the first where Amoo acts as editing chief — reviewing and directing the editor's revision — rather than editing directly, and the first where only the picture differs between arms.

Date: 2026-08-29

Avatar: The Overlooked

Campaign ID: 120237768514430539

Campaign: 20260222 - TBF | Overlooked - Testing

Template adset: 120246829306190539 — ACTIVE readback verified 2026-08-29

Destination shorthand: Expert Healer (Overlooked)

Destination URL: https://discover.karmaitems.com/tbf-1206-healer-overlooked-advertorial-ev-001/

Frame.io source: https://next.frame.io/project/b99f8398-a80b-4a3e-8020-3a6a0c101625/4b8173be-a225-4a5f-83ef-88949eea4735 (folder ER Nurse Control Video Feedback)

Ad set name: 20260829 - EXP - TBF - Exp 65 ER Nurse Amoo Editing Chief - Expert Healer (Overlooked)

Live ad set name: 20260829 - P - EXP - TBF - Exp 65 ER Nurse Amoo Editing Chief - Expert Healer (Overlooked)

Ad set ID: 120247683622720539

Live ad IDs:

• Control: 120247683624470539, 120247683625830539
• V1: 120247683626900539, 120247683627660539

Variant map:

• 20260829 - EXP - TBF - Exp 65 Control ER Nurse Amoo Editing Chief - Expert Healer (Overlooked) — 10082026 - TBF - Overlooked Avatar - ER Nurse - Control Video.mp4; sha256 69f9394001fdbc3477ed861e1135b86c29b66752eabcf6b41bf22b637c50331f; 347.500s; 1080×1080 30fps.
• 20260829 - EXP - TBF - Exp 65 V1 ER Nurse Amoo Editing Chief - Expert Healer (Overlooked) — 10082026 - TBF - Overlooked Avatar - ER Nurse - Feedback Video V1.mp4; sha256 fe8490fa4b67135e71aec3b02775972f6a1d5d6f89c50881a0d2364defdaaae4; 347.500s; 1080×1080 30fps.

Hypothesis: Amoo acting as video editing chief — reviewing the cut and directing the editor's revision — produces a visual edit that earns more Meta allocation than the original Control cut. If V1 wins, the editing-chief feedback loop is a transferable finishing step for this VSL family.

Experiment classification: Single-variable visual-edit isolation within the Amoo lineage. Exp 045/053 bundled Amoo's own edit with delivery differences (053 also a new voice); here only the picture differs between arms.

Novel variable: Visual editing treatment only — the picture re-cut to Amoo's editing-chief feedback. Proof of isolation: ffmpeg stream-level MD5s show both arms' audio streams are bit-identical (da6403c76de58f36a92d95ecb15e4778) while the video streams differ (qa/audio_parity.json).

Causal limit: A result attributes to the visual re-edit made under Amoo's direction. It cannot separate Amoo's feedback quality from the editor's execution, and there is no third arm with a non-Amoo re-edit for comparison.

Provenance note: Both arms are fresh Frame.io exports. The Control's bytes match no previously launched custody asset (checked against the 2026-08-19 Paramedic V1–V5 hashes and Exp 042's Control hash), so both arms enter Meta as new ad entities with no creative-history advantage. Duration exactly matches the Exp 042 Control cut (347.500s); byte-level ancestry is unverifiable because Exp 042's queue custody assets were cleaned after completion.

Hold constant: Audio (voice, script, music — bit-identical stream), character and story, duration (347.500s), resolution/frame rate (1080×1080, 30fps), canonical Sheet headline Start Attracting Unexpected Money in 2 Weeks, canonical Sheet primary text (sha256 0e07253aab1c4572a82ff700b25f1bdbb29166ef36b895c5d563346826911b9b, length 1339), destination, CTA learn_more, campaign, audience template lineage, one fresh CBO ad set.

Decision logic:

• V1 top spender with ≥$50 lifetime → Amoo's editing-chief feedback pass improves the finish; apply the feedback loop to the next eligible VSL before promoting it to doctrine.
• Control top spender with ≥$50 lifetime → the original cut stands; the feedback pass does not improve allocation.
• One arm below $50 lifetime while the other delivers → classify the starved arm NO-DELIVERY and call the delivered allocation winner.
• Both arms below $50 lifetime → whole-adset rejection; no directional read.

Spend contract: Fresh CBO adset in campaign 120237768514430539; no adset budget. AdLauncher fixed CBO floor: 2000¢ per adset, not per arm.

Production route: Manual/no-Trello Frame.io intake → exact original downloads staged under output/meta-uploads/tbf-exp065-er-nurse-amoo-editing-chief-20260829/assets/ → live Sheet/product/congruence gates → one AdLauncher request with Control + V1. Normal AdLauncher release cadence; no forced release.

AdLauncher request ID: produced-tbf-exp065-er-nurse-amoo-editing-chief-20260829-a95546ce

AdLauncher queue state: live (publisher proof; register auto-synced)

AdLauncher request SHA-256: 30fb19dcaac172434223e422aaa52e10f4075a73a4a204af815bcfddd25c1fac

Queue receipt: Request produced-tbf-exp065-er-nurse-amoo-editing-chief-20260829-a95546ce; state queued; request sha256 30fb19dcaac172434223e422aaa52e10f4075a73a4a204af815bcfddd25c1fac. Queue custody contains exactly two media in Control → V1 order: Control sha256 69f9394001fdbc3477ed861e1135b86c29b66752eabcf6b41bf22b637c50331f; V1 sha256 fe8490fa4b67135e71aec3b02775972f6a1d5d6f89c50881a0d2364defdaaae4. Strict post-enqueue readback passed all 22 checks; no platform IDs were created and no release was forced.

Result: Control — top spend $40.07 on 2026-09-07 (latest day). Total adset spend $130.25 over 4 days. Ranking: Control $40.07 latest / $126.92 total; V1 $0.87 latest / $3.33 total NO-DELIVERY (<$50 lifetime). Auto-called per 4d/$50 doctrine.

Element tested: Hook framing for the Spiritually Curious Skeptic avatar (skeptical-scientist control vs pattern-evidence / pain-recognition / origin-story frames).

Status: Complete

Family: skeptic-opener

Result vs decision logic: V1 Skeptical Scientist (specific scientist case study) won decisively: $238.52 spend, $654.10 revenue, 2.74x ROAS. V2 Pattern Evidence was delivered but did not convert ($58.03 / $0); V3 Direct Pain Recognition ($19.19) and V4 Personal Origin Story ($25.03) were NO-DELIVERY. Under the amended doctrine the starved arm is a Meta rejection, not an unmeasured rival — the allocation winner IS the frame verdict. The Skeptical Scientist frame is confirmed skeptic-hook doctrine. (This experiment predates the preregistration requirement; retroactively adopted, outcome confirmed on allocation plus conversion.)

Learnings:

• [avatar: Spiritually Curious Skeptic] The Skeptical Scientist frame is confirmed for the skeptic avatar: credit the skeptic's worldview first ('she believed in neurons and probability'), then ask the one question that makes them go quiet ('when did it start?').
• [family] Delivery without conversion (V2 Pattern Evidence was served but produced no revenue) is a second, independent rejection layer — the platform delivered it and the market declined it. Both layers point at the same winner.
• Transfer: Applies to skeptic-avatar hooks across TBF. Cheapest test: open the next skeptic script with the worldview-credit + quiet-question structure and compare against the 14.5 baseline metrics.

Ad set name: 20260708 - EXP - TBF - Exp 14.5 Skeptic Hook Framing - 17 Jetsetters (Spiritually Curious Skeptic)

Ad set ID: 120246407497680539

Variant map:

• 20260708 - EXP - TBF - Exp 14.5 V1 Control Skeptical Scientist - 17 Jetsetters (Spiritually Curious Skeptic) — ad 120246407498990539
• 20260708 - EXP - TBF - Exp 14.5 V2 Pattern Evidence - 17 Jetsetters (Spiritually Curious Skeptic) — ad 120246407499400539
• 20260708 - EXP - TBF - Exp 14.5 V3 Direct Pain Recognition - 17 Jetsetters (Spiritually Curious Skeptic) — ad 120246407500110539
• 20260708 - EXP - TBF - Exp 14.5 V4 Personal Origin Story - 17 Jetsetters (Spiritually Curious Skeptic) — ad 120246407500960539

Result: V1 — top spend $54.35 on 2026-07-18 (latest day). Total adset spend $340.77 over 9 days. Ranking: V1 $54.35 latest / $238.52 total; V4 $3.64 latest / $25.03 total NO-DELIVERY (<$50 lifetime); V2 $1.94 latest / $58.03 total; V3 $0.00 latest / $19.19 total NO-DELIVERY (<$50 lifetime). Auto-called per 4d/$50 doctrine.

Provenance: Launched 2026-07-08 on Meta as "TBF Exp 014" with no register block — number collided with 014 (richard-champa-message-compression). Adopted into the register and renumbered 14.5 per Edward 2026-07-11; live entities renamed the same day. Spec/hypothesis were never logged at design time — reconstruct from ad copy before calling a result.

Element tested: Credibility-bridge frame inside the same SCS Authority UGC script — restrained skeptical-scientist case (14.5 winner, verbatim) vs lost-discovery archive-ledger frame.

Status: Complete

Family: skeptic-opener

Parent: Exp 14.5

Result vs decision logic: Branch 1 fired: V1 delivered ($129 total, $33.65 latest day) and Control is NO-DELIVERY (<$50 lifetime, $10.45). Under the amended doctrine the starved arm is a Meta rejection, not an unmeasured rival — the allocation winner IS the frame verdict. Lost Discovery is the SCS opener family; Banana's prediction validated.

Learnings:

• [avatar] The lost-discovery archive-ledger bridge is a confirmed credibility thread for the Spiritually Curious Skeptic. Future SCS scripts open with the rediscovered-archive frame (case ledgers, generations of observation, names change but the pattern doesn't).
• [universal] A starved losing arm is a rejection by the auction, not a rival that was never measured. Never re-run against a rejected arm; the delivered winner already settled that contest.
• [avatar] The SCS avatar now has a confirmed opener doctrine: Lost Discovery. The skeptical-scientist frame (14.5) is superseded, not invalidated — it remains an alternative bridge family.
• Transfer: Applies to any TBF avatar using Authority UGC monk VSL. Cheapest test: next SCS script opens Lost Discovery and validates the frame on a second avatar or hook variant.

Move: Adjacent territory merged with 14.5's preregistered open-question fresh-pair confirm — impact-ranked #1 under the move-library redesign of design-next-ad-experiment. Banana family model's top unseen SCS combos are Lost Discovery hooks (predicted $500–760).

Ad set name: 20260713 - EXP - TBF - Exp 16 Lost Discovery Bridge - 17 Jetsetters (Spiritually Curious Skeptic)

Live ad set name: 20260713 - EXP - TBF - Exp 16 Lost Discovery Bridge - 17 Jetsetters (Spiritually Curious Skeptic)

Ad set ID: 120246552979210539

Variant map:

• 20260713 - EXP - TBF - Exp 16 Control Lost Discovery Bridge - 17 Jetsetters (Spiritually Curious Skeptic) — 14.5 winner creative reused verbatim (source ad 120246407498990539, video sha 7910ec72…)
• 20260713 - EXP - TBF - Exp 16 V1 Lost Discovery Bridge - 17 Jetsetters (Spiritually Curious Skeptic) — new render, matched batch (locked Champa Dorje photo + avatar_prompt from 14.5 variation-01)

Delivery risk: Fresh-pair design (TBF 009 pattern): both ads enter a brand-new CBO adset with no incumbent delivery history — the Control is a re-uploaded creative with a new ad ID, so neither arm starts with an allocation advantage. If either arm's lifetime spend stays under $50, that arm is NO-DELIVERY per Meta-rejection doctrine; a challenger NO-DELIVERY leaves 14.5's skeptical-scientist result at Early-signal confidence, unconfirmed.

Date: 2026-07-13

Avatar: The Spiritually Curious Sceptic (17 Jetsetters)

Creative type: Authority UGC monk VSL — SoulGem authority_talking_head_3x4, character Champa Dorje

Source observation: 14.5 called V1 (skeptical-scientist frame) per 4d/$50 doctrine, but at Early-signal confidence only — $38.17 lifetime on the winner, V2–V4 all NO-DELIVERY, and the result line preregistered "require a literal fresh pair before treating the framing as confirmed."

Hypothesis: For the Spiritually Curious Skeptic, an evidence-first lost-discovery bridge (rediscovered archive ledgers documenting the same relational pattern across generations) will earn Meta allocation at least comparable to the restrained skeptical-scientist case, because it offers the same restrained, proof-shaped credibility without asking the viewer to identify with a named academic.

Novel variable: The credibility-bridge thread only — two contiguous spans (opening bridge after the diagnosis; closing echo before the CTA). Control: Dr. Priya Mehta live case. V1: 1998 rediscovered Drepung case ledgers (1911 merchant, 1954 schoolteacher, final-entry woman). Everything outside the two spans is byte-identical (span validator: output/adchain/tbf-exp016-scs-lost-discovery-bridge/scripts/validation.json, ok=true; Control 703 words sha 7a6a3d36…, V1 704 words sha aa108eed…).

Hold constant: Opening reversal, relational-interference diagnosis, root-chakra mechanism, think-back block, product/offer copy (150/lunar cycle), restrained CTA; character, locked portrait, avatar prompt, render style and resolution (matched to Control's original 14.5 render — 540p cost default deliberately NOT applied, validity over cost); primary text, headline, destination URL, CTA type, audience/placement template, campaign 120245415438400539.

Decision logic:

• V1 top spend on latest day, Control delivering ≥$50 → Lost Discovery becomes the SCS opener family; Banana's prediction validated; next SCS scripts open lost-discovery.
• Control top spend, V1 delivering ≥$50 → skeptical-scientist frame upgraded Early-signal → Confirmed (14.5's fresh-pair requirement satisfied); it becomes SCS opener doctrine.
• Either arm < $50 lifetime → that arm NO-DELIVERY per Meta-rejection doctrine; no frame confirmation, log deliverability learning only.

Transferable rule: If this resolves, future TBF SCS ads will be built differently: every SCS script opens with the winning credibility-bridge family, applied to all future SCS scripts from AdMaker and AdChain.

Why it matters: The entire SCS avatar currently rests on one $38 early signal. This test either confirms the anchor frame or replaces it with the model's highest-predicted untested frame — before more SCS spend stacks on unverified doctrine.

Spend contract: Fresh CBO adset in campaign 120245415438400539 (SCS testing), no adset budget, daily_min_spend_target $30/day — replicates the approved 14.5 contract (verify.json: hasBudget=false, minSpend=3000). Source: approved experiment spec + Edward's go at the AdChain checkpoint.

Launch status: Live / ACTIVE — 2026-07-14 (adset + both ads verified ACTIVE post-activation; min-spend verified 3000¢ via platform-api-requests raw GET)

Scripts: /root/.hermes/output/adchain/tbf-exp016-scs-lost-discovery-bridge/scripts/ (control 7a6a3d36…, v1 aa108eed…; validation.json ok=true)

Live destination: https://discover.karmaitems.com/tbf-2306-scs-advertorial-ev-001/

Campaign: 120245415438400539 — 20260623 - TBF | Spiritually Curious Skeptic - Testing

Live adset: 120246552979210539 — 20260713 - EXP - TBF - Exp 16 Lost Discovery Bridge - 17 Jetsetters (Spiritually Curious Skeptic) — ACTIVE

Ads:

• Control 120246552984150539 — ACTIVE (14.5 winner creative re-upload, video sha 7910ec72…)
• V1 Corrected 120246554194180539 — ACTIVE (720p matched render, captions re-burned to MarginV=232 matching Control; video sha a9dc412d…) — THE live V1 arm; result lookup must use this ad
• V1 120246552988710539 — PAUSED 2026-07-14 (original render shipped bottom-flush captions, MarginV=0 — SoulGem non-9:16 safe-zone regression, fixed 40ac3b87; ~1 day of delivery on this ad is excluded from the V1 arm's verdict)

Budget contract: CBO daily_min_spend_target 3000¢, no adset budget; campaign daily budget raised $211→$242 by the repair path.

Repair provenance: adset was created at the helper's $20 default (spec said $30); repaired live 2026-07-14 via repair_live_adset_spend_contract (draft 6a55c73573937162175048c9, published, readback 3000¢). Production note: V1 render was killed mid-run by a session exit; resumed surgically (segments 000/001 + TTS reused, only segment 002 re-rendered) — no creative drift, script sha re-verified before both render and resume. Caption swap: original V1 ad replaced 2026-07-14 by V1 Corrected (new ad in same adset, old ad paused; swap artifacts in batch dir swap_v1/).

Batch artifacts: /root/.hermes/output/meta-uploads/tbf-exp016-scs-lost-discovery-bridge-20260714/

Result: V1 — top spend $33.65 on 2026-07-18 (latest day). Total adset spend $139.45 over 4 days. Ranking: V1 $33.65 latest / $129.00 total; Control $0.00 latest / $10.45 total NO-DELIVERY (<$50 lifetime). Auto-called per 4d/$50 doctrine.

Element tested: Proof structure inside the confirmed lost-discovery credibility bridge (Exp 016 winner): named cases with dates vs generational pattern-of-many vs final-entry identification pivot.

Status: Complete

Result vs decision logic: Branch 1 fired: Edward called V1 on the last day the ad set spent over $5 (2026-07-27). Specific dated cases with ledger detail are load-bearing; future SCS lost-discovery scripts must carry named cases with dates and ledger detail. Branch 2 not_fired: Pattern of Many did not win. Branch 3 not_fired: Final Entry Pivot did not win. Branch 4 not_fired: the Control blend did not win. Branch 5 not_fired as an experiment-wide void — operator ranked on the last day over $5 rather than the $50 auto-call bar; V1 is the called winner.

Learnings:

• [family] Keep specific dated cases with ledger detail in every future SCS lost-discovery script. Named people, years, and the archive-ledger texture are the load-bearing proof.
• [family] Do not replace those named cases with a generational pattern-of-many. The cases are the asset, not scaffolding to cut.
• [family] Do not close the lost-discovery bridge on a final-entry identification pivot. That closer lost to named-case proof.
• Transfer: Cheapest test: same format — Champa Dorje Authority UGC monk VSL on SCS. Lock the V1 named-case-deep script as Control. Hold the UMP (someone close has shifted) and the named-case proof. Vary only the felt pain: fortune-turned vs unmatched potential vs unexplained setbacks.

Result: V1 — last day over $5 was 2026-07-27; V1 $5.21 that day / $34.25 lifetime. Ranking: V1 $5.21 / $34.25; V2 $3.80 / $15.24; Control $0.28 / $17.55; V3 $0.49 / $7.73. Adset $9.78 that day, $74.77 lifetime through 2026-08-14. Operator-called 2026-08-15 on last day with over $5 (auto-checker withheld: no arm strictly over $50 lifetime).

Family: skeptic-opener

Parent: Exp 016

Date: 2026-07-21

Avatar: The Spiritually Curious Sceptic (17 Jetsetters)

Creative type: Authority UGC monk VSL — SoulGem authority_talking_head_3x4, character Champa Dorje

Design provenance: Exp 016 confirmed the lost-discovery archive-ledger bridge as the SCS opener family (Branch 1, amended doctrine — Control starved = Meta rejection, allocation winner is the verdict). First isolation on this winner; no depth-cap issue. Approved ranking candidate #1 by Edward 2026-07-21.

Transferable rule: If this resolves, future TBF Spiritually Curious Skeptic lost-discovery scripts carry the winning proof structure — applied to every SCS script AdMaker/AdChain writes from here.

Ad set name: 20260721 - EXP - TBF - Exp 26 Lost Discovery Proof Type - 17 Jetsetters (Spiritually Curious Skeptic)

Live ad set name: 20260721 - EXP - TBF - Exp 26 Lost Discovery Proof Type - 17 Jetsetters (Spiritually Curious Skeptic)

Ad set ID: 120246795357040539

AdLauncher request ID: produced-tbf-exp026-scs-lost-discovery-proof-type-20260721-b6ca4c3a

AdLauncher queue state: live (publisher proof; register auto-synced)

AdLauncher request SHA-256: 73999249a24f7e192646db2c44dd71031e87cdabfe34944b28d3c19821add614

Variant map:

• 20260721 - EXP - TBF - Exp 26 Control Lost Discovery Proof Type - 17 Jetsetters (Spiritually Curious Skeptic) — 016 V1 verbatim (704 words, sha aa108eed9187; creative reused from live ad 120246554194180539, no re-render)
• 20260721 - EXP - TBF - Exp 26 V1 Named Cases Deep Lost Discovery Proof Type - 17 Jetsetters (Spiritually Curious Skeptic) — same two named cases, expanded sensory/ledger detail (739 words, sha 43aa119c23a4)
• 20260721 - EXP - TBF - Exp 26 V2 Pattern of Many Lost Discovery Proof Type - 17 Jetsetters (Spiritually Curious Skeptic) — no named cases, generational pattern only (692 words, sha e7d4ad7ce6f4)
• 20260721 - EXP - TBF - Exp 26 V3 Final Entry Pivot Lost Discovery Proof Type - 17 Jetsetters (Spiritually Curious Skeptic) — named cases + 1971 final-entry-woman identification pivot (757 words, sha 070d94600066)

Novel variable: The internal proof structure inside the lost-discovery bridge span only ("In 1998, during a renovation…" → "The pattern does not change."). All four scripts are byte-identical outside this span (asserted in build_scripts.py).

Hold constant:

• Opening reversal ("It is not bad decisions. It is not bad luck.… Someone close to you has shifted") — byte-identical
• Everything after the span ("Luck is what we call it…" through the restrained CTA) — byte-identical
• Format: Authority UGC monk talking head; character Champa Dorje, locked photo + avatar_prompt from 14.5 variation-01 (matched-batch pattern)
• Primary text, headline ("If you keep an open mind, read this"), destination https://discover.karmaitems.com/tbf-2306-scs-advertorial-ev-001/, CTA learn_more — identical across all four ads, cloned from the 14.5/016 manifest
• Audience: 17 Jetsetters SCS; campaign 120245415438400539 (SCS testing)
• Script length band 650–780 words (Control 704 / V1 739 / V2 692 / V3 757)

Decision logic:

• V1 wins → specific dated cases are load-bearing; future SCS lost-discovery scripts must carry named cases with dates and ledger detail.
• V2 wins → the generational pattern-of-many is the asset; named cases are scaffolding — cut them from future scripts.
• V3 wins → the final-entry identification pivot is the load-bearing closer inside the bridge; next test asks whether it carries without named cases.
• Control wins → the current blend (named cases + pattern) is the asset; proof structure is mapped, stop decomposing this span.
• If any challenger spend < $50 → NO-DELIVERY — log as Meta rejection per doctrine; under the amended doctrine (2026-07-21) a starved arm is a rejection verdict, not an unmeasured rival. No re-run.

Delivery risk: Fresh-pair CBO adset (TBF 009 pattern): all four ads enter a brand-new adset with no incumbent delivery history. Control re-enters under a fresh ad ID (video reused, new upload) so no arm carries creative-history advantage. 4-arm champion-challenger adsets are near the 013 starvation ceiling — arms are maximally spaced on the proof-type axis and each carries a distinct decision outcome; NO-DELIVERY branch pre-registered per arm.

Spend contract: Fresh CBO adset in campaign 120245415438400539, AdLauncher fixed CBO floor 2000¢/arm ($80 total floor). Historical $30/3000¢ min-spend from 016 noted as evidence only.

Production route: AdChain experiment mode — SoulGem authority_talking_head_3x4 renders of V1/V2/V3 matched to 016 settings (720p 832x1088, locked Champa Dorje photo + avatar_prompt, captions MarginV=232 per the P11 fix); Control = 016 V1 live video reused verbatim. Checkpoint 2: locked scripts (this spec). Checkpoint 3: pre-launch — nothing goes live on Meta without Edward's go.

Scripts: /root/.hermes/output/adchain/tbf-exp026-scs-lost-discovery-proof-type/scripts/

• control_lost_discovery_verbatim.txt (704 words, sha aa108eed9187)
• v1_named_cases_deep.txt (739 words, sha 43aa119c23a4)
• v2_pattern_of_many.txt (692 words, sha e7d4ad7ce6f4)
• v3_final_entry_pivot.txt (757 words, sha 070d94600066)
• build_scripts.py + spans.json (span-swap construction, outside-span byte-identity asserted)

Verification: All four scripts passed verify_generic_script.py (word count 650–780, no spoken URL, no banned surnames, no disease terms, authority first person, no financial advisor phrases, no provenance leakage). 2026-07-21.

Element tested: Felt pain named in the SCS lost-discovery monk VSL — fortune-turned vs unmatched potential vs unexplained setbacks — holding the UMP (someone close has shifted) and the named-case proof.

Status: Complete

Result vs decision logic: Branch 2 fired (V2 wins): unexplained setbacks is the default SCS felt pain. Branch 3 fired (winner delivers, challengers under the floor): V1 and Control are rejections; V2 is the verdict. Branches 0, 1, and 4 did not fire.

Learnings:

• [family] Name the SCS wound as a pattern of setbacks that will not explain themselves — that felt pain beat fortune-turned and unmatched-potential.
• [avatar] Lead the Spiritually Curious Skeptic with the unexplained-setbacks wound; keep the 'someone close has shifted' UMP and the dated cases as the shared spine.
• Transfer: Cheapest test: carry the unexplained-setbacks wound into the next SCS creative and see whether it holds outside the monk-VSL frame.

Family: skeptic-opener

Parent: Exp 026

Altitude: Rung 4 — angle (felt pain)

Date: 2026-08-15

Avatar: The Spiritually Curious Sceptic (17 Jetsetters)

Creative type: Authority UGC monk VSL — SoulGem authority_talking_head_3x4, character Champa Dorje

Campaign ID: 120245415438400539

Campaign: 20260623 - TBF | Spiritually Curious Skeptic - Testing

Destination URL: https://discover.karmaitems.com/tbf-2306-scs-advertorial-ev-001/

Destination shorthand: 17 Jetsetters (Spiritually Curious Skeptic)

Ad set name: 20260815 - EXP - TBF - Exp 52 Pain Type - 17 Jetsetters (Spiritually Curious Skeptic)

Live ad set name: 20260815 - P - EXP - TBF - Exp 52 Pain Type - 17 Jetsetters (Spiritually Curious Skeptic)

Ad set ID: 120247419401090539

Live ad IDs:

• Control: 120247419411690539, 120247419412970539
• V1: 120247419414120539, 120247419415700539
• V2: 120247419417160539, 120247419419170539

Variant map:

• 20260815 - EXP - TBF - Exp 52 Control Fortune Turned Pain Type - 17 Jetsetters (Spiritually Curious Skeptic) — Exp 26 V1 verbatim (739 words, sha 43aa119c23a4); felt pain = fortune turned
• 20260815 - EXP - TBF - Exp 52 V1 Unmatched Potential Pain Type - 17 Jetsetters (Spiritually Curious Skeptic) — same UMP + named cases; felt pain = life did not match the work (775 words, sha a69899042331)
• 20260815 - EXP - TBF - Exp 52 V2 Unexplained Setbacks Pain Type - 17 Jetsetters (Spiritually Curious Skeptic) — same UMP + named cases; felt pain = setbacks that will not explain themselves (753 words, sha 7abc7655b118)

Result: V2 — top spend $10.16 on 2026-08-23 (latest day). Total adset spend $101.86 over 6 days. Ranking: V2 $10.16 latest / $57.89 total; V1 $4.03 latest / $27.78 total NO-DELIVERY (<$50 lifetime); Control $2.63 latest / $16.19 total NO-DELIVERY (<$50 lifetime). Auto-called per 4d/$50 doctrine.

Design provenance: Exp 016 locked lost-discovery as the SCS opener. Exp 026 isolated proof type; Edward called V1 (named-case-deep) on last day over $5. Operator vetoed a case-count follow-up and a static transfer. Operator correction 2026-08-15: "someone close to you has shifted" is the UMP, not a pain. This test holds that UMP and the dated cases; only the named felt pain changes. Checkpoint 1 copy approved by "adchain to live" 2026-08-15.

Hypothesis: The SCS lost-discovery winner names the wound as fortune turning. A different felt pain — unmatched potential, or unexplained setbacks — may earn more allocation on the same UMP and proof.

Prospect claim: The Spiritually Curious Skeptic already feels a quiet shortfall; the variant works iff the named wound is the one they recognize before the relational cause is named.

Placement: Problem-aware. The avatar already suspects something is quietly wrong; the test names which wrong.

Novel variable: Felt pain named in the viewer-facing reversal, case-interview line, merchant outcome, and "what we call it" line.

Hold constant:

• UMP: someone close has shifted (opening restatement, source-is-a-person block, Think back questions)
• Named-case identities: 1998 chest, 1911 merchant, 1954 schoolteacher, ledger texture
• After-span product / copper / root-chakra / 150-per-cycle / CTA
• Format: Authority UGC monk talking head; Champa Dorje; locked Himalayan-cave photo
• Primary text, headline ("If you keep an open mind, read this"), destination https://discover.karmaitems.com/tbf-2306-scs-advertorial-ev-001/, CTA learn_more
• Audience: 17 Jetsetters SCS; campaign 120245415438400539
• Script length band 650–780 words

Decision logic:

• Control wins → fortune-turned is the default SCS felt pain; keep naming the wound as fortune/luck turning.
• V1 wins → unmatched potential is the default SCS felt pain; future SCS scripts name the wound as life not matching the work.
• V2 wins → unexplained setbacks is the default SCS felt pain; future SCS scripts name the wound as a pattern of setbacks that will not explain themselves.
• Winner delivers and a challenger is under $50 lifetime → that challenger is a rejection; the delivered winner is the verdict; do not re-run the rejected arm.
• All arms under $50 lifetime → early signal; one fresh-pair confirmation.

Delivery risk: Fresh-pair CBO adset. Control reuses the live Exp 26 V1 take (Meta video 2689795551415791) under a new ad ID. V1/V2 are new matched renders. 3-arm champion-challenger set is inside the starvation ceiling. Any arm under $50 lifetime is a Meta rejection. If at least one arm delivers, the allocation winner is the verdict.

Spend contract: Fresh CBO adset in campaign 120245415438400539. AdLauncher fixed CBO floor: 2000¢. $50 challenger spend is the evidence threshold only.

Production route: AdChain experiment mode — Control = live Exp 26 V1 recovered from Meta (no Hedra remake). V1/V2 SoulGem authority_talking_head_3x4 matched to the 014/026 Champa Himalayan-cave lock, 720p 3:4, playback 1.0. Manual/no-Trello enqueue. Operator go: "adchain to live" 2026-08-15.

Transferable rule: If this resolves, future TBF SCS lost-discovery scripts will name the winning felt pain as the default wound, applied to every SCS Authority UGC monk VSL.

Why it matters: Proof type is mapped. The remaining high-leverage question is which wound the Skeptic recognizes, not how many dated cases we keep.

Scripts: /home/runner/.hermes/output/adchain/tbf-scs-pain-type/scripts/

• control_fortune_turned.txt (739 words, sha 43aa119c23a4) — Exp 26 V1 verbatim
• v1_unmatched_potential.txt (775 words, sha a69899042331)
• v2_unexplained_setbacks.txt (753 words, sha 7abc7655b118)
• build_scripts.py + spans.json

Verification: All three scripts passed verify_generic_script.py (word count 650–780, no spoken URL, no banned surnames, no disease terms, authority first person, no financial advisor phrases, no provenance leakage). 2026-08-15.

AdLauncher request ID: produced-tbf-exp052-scs-pain-type-20260815-434c6b86

AdLauncher queue state: live (publisher proof; register auto-synced)

AdLauncher request SHA-256: 72c6fb647dbdca582075f06ebd42c2bf051386aa07ff6e5424520264fa624691

Queue receipt: Request produced-tbf-exp052-scs-pain-type-20260815-434c6b86; state queued; request sha256 72c6fb647dbdca582075f06ebd42c2bf051386aa07ff6e5424520264fa624691. No platform IDs created; no release forced.

Element tested: Complete SoulGem render-format package carrying the exact Exp 052 V2 SCS winner — full-frame Champa talking head versus Champa corner authority over an educational visual bed versus Champa narration over kinetic text slides.

Status: Complete

Result vs decision logic: Branch 1 fired: Control full-frame Champa took the most allocation with $56.19 lifetime and $8.56 on the final day. Branches 2 and 3 not fired: V1 corner authority drew $46.96 and V2 text slides $16.22, both under $50. Branch 4 fired: two arms stayed under $50 while Control delivered, so both are rejections and the delivered winner is the verdict. Branch 5 not fired.

Learnings:

• [family] Keep full-frame visible Champa (authority_talking_head) as the default carrier for the Sceptic unexplained-setbacks message; the corner-authority and kinetic-text packages of the identical script were not funded.
• [avatar] A visible presenter carries the Spiritually Curious Sceptic message better than message-led text; do not move SCS creative to slide formats by default.
• Transfer: Cheapest test: run the same three-package split on the next Cursed monk-message winner to learn whether the full-frame preference is a Champa-frame rule or an SCS-only one.

Family: skeptic-opener

Parent: Exp 052

Altitude: Rung 8 — production polish

Ad set name: 20260826 - EXP - TBF - Exp 59 SCS Format Translation - 17 Jetsetters (Spiritually Curious Skeptic)

Live ad set name: 20260826 - P - EXP - TBF - Exp 59 SCS Format Translation - 17 Jetsetters (Spiritually Curious Skeptic)

Ad set ID: 120247575607310539

Live ad IDs:

• Control: 120247575609990539, 120247575611860539
• V1: 120247575612440539, 120247575613430539
• V2: 120247575615660539, 120247575616120539

Variant map:

• 20260826 - EXP - TBF - Exp 59 Control Talking Head SCS Format Translation - 17 Jetsetters (Spiritually Curious Skeptic) — fresh SoulGem authority_talking_head render with full-frame Champa Dorje.
• 20260826 - EXP - TBF - Exp 59 V1 Corner Authority SCS Format Translation - 17 Jetsetters (Spiritually Curious Skeptic) — fresh SoulGem corner_authority_vsl render with Champa Dorje in the corner over the educational animated bed.
• 20260826 - EXP - TBF - Exp 59 V2 Text Slide SCS Format Translation - 17 Jetsetters (Spiritually Curious Skeptic) — fresh SoulGem slide_vsl render with Champa Dorje's registered voice over kinetic text, distributed contextual B-roll, and exact TBF product photography; no visible presenter.

Result: Control — top spend $8.56 on 2026-09-02 (latest day). Total adset spend $119.37 over 6 days. Ranking: Control $8.56 latest / $56.19 total; V1 $6.51 latest / $46.96 total NO-DELIVERY (<$50 lifetime); V2 $1.90 latest / $16.22 total NO-DELIVERY (<$50 lifetime). Auto-called per 4d/$50 doctrine.

Date: 2026-08-26

Avatar: The Spiritually Curious Sceptic (17 Jetsetters)

Campaign ID: 120245415438400539

Campaign: 20260623 - TBF | Spiritually Curious Skeptic - Testing

Destination shorthand: 17 Jetsetters (Spiritually Curious Skeptic)

Destination URL: https://discover.karmaitems.com/tbf-2306-scs-advertorial-ev-001/

Approved script: Exact Exp 052 V2 winner at output/adchain/tbf-scs-pain-type/scripts/v2_unexplained_setbacks.txt; 753 words; sha256 7abc7655b11880f5b573662d505e2e256cb460a11ae3cb12f3224411e22df23e. The same spoken word sequence is mandatory in every arm. slide_vsl blank-line segmentation is structural metadata only and may not add, delete, reorder, or rewrite spoken words. The script passed verify_generic_script.py at the 701–1000-word Authority UGC Long band on 2026-08-26.

Hypothesis: Exp 052 proved that unexplained setbacks is the SCS wound when the UMP, dated Drepung cases, relational source, Root Chakra mechanism, copper seal, and CTA are retained. This test resolves which complete production format best carries that fixed winning message: visible full-frame authority, authority plus explanatory visuals, or message-led kinetic text.

Prospect claim: The Spiritually Curious Skeptic wants a rational bridge before accepting a spiritual cause; a format wins iff it makes the dated evidence and mechanism easier to absorb without weakening Champa's authority.

Placement: Problem-aware. The viewer recognizes unexplained setbacks but still needs evidence before accepting the relational source and copper solution.

Novel variable: SoulGem render-format package, with three levels: authority_talking_head, corner_authority_vsl, and slide_vsl. Each package's native composition, aspect, caption treatment, pacing, music, and visual grammar move together. This is a format-package comparison, not an isolation of any one subcomponent.

Hold constant:

• Exact 753-word script bytes and spoken word order, including every skeptic-opener family lesson: unexplained setbacks; “someone close to you has shifted”; the 1998 Drepung chest; the 1911 merchant and 1954 schoolteacher; the relational source; Root Chakra mechanism; copper seal; 150-per-cycle scarcity; and CTA.
• Champa Dorje's registered ElevenLabs voice identity across all three arms. Format-native synthesis/assembly may differ and is part of the format package; no narrator substitution is allowed.
• Primary text; headline If you keep an open mind, read this; CTA learn_more; campaign; audience; offer; product; and SCS destination.
• All three arms are freshly produced and launched as new media and ad entities in one fresh CBO adset. No arm may reuse a live post or inherit creative history.

Primary test variants:

• Control — Full-frame Champa talking head: exact approved script through canonical authority_talking_head.
• V1 — Corner Authority VSL: exact approved script through canonical corner_authority_vsl; Champa remains visible while the educational bed carries the evidence and mechanism.
• V2 — Text Slide VSL: exact approved script through canonical slide_vsl; Champa remains the voice while kinetic text carries the message and contextual B-roll/product imagery illustrates selected evidence, mechanism, and offer beats.

Decision logic:

• Control takes the most allocation with at least $50 lifetime spend → full-frame visible Champa is the best carrier for this SCS family; keep authority_talking_head as its default.
• V1 takes the most allocation with at least $50 lifetime spend → explanatory visuals plus visible Champa best carry this SCS message; promote corner_authority_vsl for the next SCS family creative.
• V2 takes the most allocation with at least $50 lifetime spend → the fixed message can carry without a visible presenter; promote slide_vsl for the next SCS family creative.
• One or two arms remain below $50 while another delivers → each starved arm is a Meta rejection; the delivered allocation winner is the verdict and rejected arms are never re-run.
• All three arms remain below $50 → whole-adset rejection; treat as an early signal requiring one fresh three-arm confirmation before changing the SCS format default.

Delivery risk: Fresh three-arm CBO adset, not an incumbent injection. All arms use new media and ad entities so the Exp 052 Control carries no creative-history advantage. Any arm under $50 lifetime spend is a Meta rejection. If at least one arm delivers, Meta's allocation winner is the verdict. Only if all three starve is one fresh confirmation permitted.

Spend contract: Fresh CBO adset in campaign 120245415438400539; no adset budget. AdLauncher fixed CBO floor: 2000¢ per adset, not per arm. The $50 figure is the evidence threshold for interpreting each arm, not a JungleLab-selected spend floor.

Production route: AdChain experiment mode. Use the three canonical SoulGem formats named above, then deterministic script/transcript/voice-identity/media-cardinality checks, live Sheet/product/congruence gates, and one AdLauncher request in Control → V1 → V2 order. Edward approved the three-arm paid production and normal-cadence enqueue on 2026-08-26; no forced release was authorized.

Production receipt (2026-08-26): The original V1 preflight block was superseded when Edward approved bed reuse. corner_authority_vsl now selects eight evenly distributed storyboard anchors (source indexes 0, 10, 20, 30, 39, 49, 59, 69), holds each for 7 seconds, and cycles that complete 56-second educational sequence six times across the narration. Control: 233.64s, 576×704, sha256 fdc9914531e44d8036104c03fdfa79d79b961df4ba7e351152eaf1915882ab33. V1: 237.25s, 720×1280, sha256 7b810b01c3c91d773f5d6bf2c108e8a11f43cb901fc970706979a74981fa2d11. Corrected V2: 341.994667s, 720×1280, 67 slides with 13 distributed inserts — 10 contextual B-roll and 3 exact TBF product photographs — covering 21.6169% of the timeline; 75.2298% of insert runtime is non-product B-roll; sha256 fae00db12c2dfd926c40d6c37005b1e3db26f89ebebab1ef6f365cd081f1809c. All three passed stream/decode, visual, caption/timeline, and voice-identity QA; corrected V2 additionally passed one isolated vision check for text boundaries and each of its 13 inserts. Every arm retains the exact normalized 753-word sequence and source hash 7abc7655b11880f5b573662d505e2e256cb460a11ae3cb12f3224411e22df23e, so every mapped skeptic-opener lesson is held constant.

AdLauncher request ID: produced-tbf-exp059-scs-format-translation-v2-corrected-20260826-b65f8ece

AdLauncher queue state: live (publisher proof; register auto-synced)

AdLauncher request SHA-256: ecbb0977547660402d50e941cd3929f6f976334885a65eff1220e23c7535e009

Queue receipt: The corrected request contains exactly three queue-owned videos in Control → V1 → V2 order. Live Sheet, product identity, fresh batch-bound ad↔page congruence, template targeting/page/pixel, strict request, source→staged→queue custody, and 2000¢ CBO-floor checks passed. Control and V1 bytes are unchanged; only V2 changed. The original request produced-tbf-exp059-scs-format-translation-20260826-1aadc012 was proven to have zero remote Meta side effects, then cancelled with superseded_by pointing to the corrected request. Enqueue created zero Meta objects and no release was forced.

Why it matters: This determines whether the Exp 052 SCS win belongs to Champa's full-screen presence, benefits from explanatory visuals, or survives as a message-led format. The result sets the production default for the next ads in the skeptic-opener family without reopening its mapped copy.

Element tested: SoulGem render format as a bundle for the same monk-authority script — corner_authority_vsl (full-frame educational 3D animated bed + corner PIP lip-synced Champa Dorje) versus authority_talking_head (full-frame Hedra 540p Champa Dorje talking head).

Status: Superseded — invalid/confounded cadence defect; adset 120247417386340539 paused 2026-08-20 and excluded from format conclusions; replaced by Exp 054

Production state (2026-08-15): Authority captions repaired (Drapung→Drepung; output_captioned.mp4 reburned 91.52s). Enqueued request_id produced-tbf-exp047-corner-vs-authority-20260815-fa9f299c; request_hash da322687aeef71463983d504fcf3b46214e527c0298a09aa49efd24b887d7ce8; batch output/meta-uploads/tbf-exp047-corner-vs-authority-20260815/; state=queued; no Meta entities yet.

Family: ted-talk

Move: Decompose — render-format isolation of a scaled winner ($882 lifetime at 1.43 ROAS, trailing-7d 2.19). No numbered parent: decomposes the 2026-07-12 manual corner launch (adset 120246510301450539).

Avatar: The Cursed

Campaign ID: 120237762972300539

Campaign: 20260222 - TBF | Curses - Testing

Template adset: 120246510301450539 — incumbent Corner Authority VSL Ted Talk V1; targeting/pixel/page cloned; fresh ACTIVE readback required at packaging.

Destination shorthand: 17 Jetsetters (Cursed)

Destination URL: https://discover.karmaitems.com/tbf-1206-curses-advertorial-ev-001

Destination: https://discover.karmaitems.com/tbf-1206-curses-advertorial-ev-001

Ad set name: 20260808 - EXP - TBF - Exp 47 Corner vs Authority UGC Render - 17 Jetsetters (Cursed)

Live ad set name: 20260808 - P - EXP - TBF - Exp 47 Corner vs Authority UGC Render - 17 Jetsetters (Cursed)

Ad set ID: 120247417386340539

Live ad IDs:

• Control: 120247417402630539, 120247417407380539
• V1: 120247417411110539, 120247417415630539

AdLauncher request ID: produced-tbf-exp047-corner-vs-authority-20260815-fa9f299c

AdLauncher queue state: live (publisher proof; register auto-synced)

AdLauncher request SHA-256: da322687aeef71463983d504fcf3b46214e527c0298a09aa49efd24b887d7ce8

Variant map:

• Control — corner_authority_vsl: Champa Dorje corner PIP over animated educational bed, 9:16, fresh render.
• V1 — authority_talking_head: Champa Dorje full-frame talking head, Hedra 540p, delivered 9:16, fresh render.

Script (fixed, verbatim both arms): ads_data_dedup.creative_id = 217 — "20260429 - TBF Ted Talk Style UGC | V1 - 17 Jetsetters (Curses)", 218 words. Sha-locked copy at adchain/runs/2026-08-08_tbf-exp047-corner-vs-authority-render/script.txt, sha256 39de23e5ae705223d7e97cd79b04606f1aa0ec61937d42d1f9fd03e370d01d22 (byte-verified against Supabase). No edits permitted in either arm.

Hypothesis: The corner format's educational bed is load-bearing for monk-authority diagnosis scripts — it visualizes the mechanism while Champa speaks and will take allocation and beat the plain talking head. Counter-outcome: the script carries; plain UGC matches or wins and the corner bed is decorative cost. Cost asymmetry is the payoff: corner renders a full Kling I2V bed plus a Hedra PIP layer; authority_talking_head is a single cheap Hedra pass. The loser's tier is retired as default for monk-authority scripts.

Novel variable: Render format bundle (bed visuals, captions style, PIP scale, pacing move together — pre-registered as a bundle; no single-element attribution).

Hold constant: Script verbatim (sha above); speaker Champa Dorje + registered voice; delivered aspect 9:16 both arms (218 words ≈ 90s, under the 120s aspect-policy threshold); both arms rendered fresh in the SAME SoulGem session (Exp 024/038 matched-production precedent — the cheaper reuse-live-corner-creative Control was considered and rejected as a five-week pipeline-drift confound); ad copy identical both arms, cloned verbatim from the live Ted Talk V1 corner ad (headline "Wear It 2 Weeks and Watch Your Luck Turn", primary text per meta-uploads/tbf-cursed-corner-tedtalk-v1-20260712/manual_manifest.json, CTA learn_more); targeting cloned from template adset; one fresh CBO adset; both arms new ad entities with no delivery history.

Decision logic:

• Control (corner) takes allocation and ROAS ≥ 1.45 bar → corner bed is load-bearing; monk-authority scripts default to corner render despite credit cost.
• V1 (plain UGC) matches or beats corner → script carries; future monk-authority renders default to the cheap tier; corner reserved for scripts needing mechanism visualization.
• Either arm below $50 lifetime → that arm is NO-DELIVERY under doctrine; load-bearing conclusions stay Early-signal until a fresh-pair re-run confirms.
• Both starve or both under bar → format is not the lever on this script; signal returns to script/hook territory.

Spend contract: Fresh CBO adset in campaign 120237762972300539; no adset budget. AdLauncher fixed CBO floor: 2000¢ per adset, not per arm.

Production route: AdChain experiment mode → SoulGem matched session (corner_authority_vsl + authority_talking_head, dry-run preflight both before paid) → Checkpoint 2 locked script (trivially the sha-locked verbatim script) → Checkpoint 3 pre-launch review (nothing goes live without it) → AdLauncher queue, next release cohort.

Invalidation / repair lineage (2026-08-20): Edward approved a normalized, fully punctuated narration and a fresh-pair rerun as Exp 054. Both replacement formats must use the same newly generated narration-master bytes; parity proof must include narration SHA-256, script SHA-256, duration, dimensions, captions, and fresh media hashes before queue custody.

Element tested: SoulGem render-format bundle for one matched monk-authority narration — corner_authority_vsl versus authority_talking_head. This is the clean replacement for invalid/confounded Exp 047.

Status: Complete

Result vs decision logic: Branch 2 fired (V1 below the floor while Control delivered): Control is the allocation winner and V1 is a Meta rejection. Branch 1 did not fire; Branch 3 did not fire. Branch 0 (ROAS >= 1.45) is indeterminate — Control took allocation but the result does not report conversion, so the load-bearing claim is confirmed on delivery and the ROAS bar stays open.

Learnings:

• [family] Render monk-authority diagnosis as the corner PIP over the animated educational bed, not the plain full-frame talking head.
• [family] Keep the educational bed under the corner PIP — the full-frame talking head does not carry the diagnosis.
• Transfer: Cheapest test: read the corner format's conversion on the next monk-authority diagnosis ad to settle the ROAS bar, so keeping the corner bed is confirmed on conversion and not delivery alone.

Family: ted-talk

Parent: Exp 047

Move: Replication / production-integrity repair

Date: 2026-08-20

Avatar: The Cursed

Campaign ID: 120237762972300539

Campaign: 20260222 - TBF | Curses - Testing

Template adset: 120246510301450539 — incumbent Corner Authority VSL Ted Talk V1; fresh ACTIVE readback required at packaging.

Destination shorthand: 17 Jetsetters (Cursed)

Destination URL: https://discover.karmaitems.com/tbf-1206-curses-advertorial-ev-001

Ad set name: 20260820 - EXP - TBF - Exp 54 Corner vs Authority Cadence Repair - 17 Jetsetters (Cursed)

Live ad set name: 20260820 - P - EXP - TBF - Exp 54 Corner vs Authority Cadence Repair - 17 Jetsetters (Cursed)

Ad set ID: 120247477897820539

Live ad IDs:

• Control: 120247477899820539, 120247477900210539
• V1: 120247477902730539, 120247477904610539

Variant map:

• 20260820 - EXP - TBF - Exp 54 Control Corner vs Authority Cadence Repair - 17 Jetsetters (Cursed) — fresh corner_authority_vsl render: Champa Dorje corner PIP over animated educational bed, 9:16.
• 20260820 - EXP - TBF - Exp 54 V1 Corner vs Authority Cadence Repair - 17 Jetsetters (Cursed) — fresh authority_talking_head render: full-frame Champa Dorje, Hedra 540p, 9:16.

Approved script: adchain/runs/2026-08-20_tbf-exp054-corner-vs-authority-cadence-repair/script.txt; 226 words; sha256 219233bd0ba4a909ce599b225cad5b1a5ad8f3f33a641357e73bfdd5820c63c3. The approved copy is fully punctuated prose. No visual-format arm may rewrite, reflow, or resynthesize it independently.

Hypothesis: The corner format's educational bed is load-bearing for the monk-authority diagnosis: with one clean narration held byte-identical, it will earn more allocation than the plain talking head. Counter-outcome: the narration carries and the cheaper full-frame authority treatment matches or wins.

Novel variable: Render-format bundle only: educational bed plus corner PIP versus full-frame talking head. Captions and visual composition move with the format; the narration-master audio bytes do not.

Hold constant: Approved script bytes; a single newly generated Champa Dorje narration master copied into both fresh SoulGem workdirs; narration SHA-256 and duration; registered voice; 9:16 aspect; 1.3× playback treatment; headline; primary text; destination; CTA; campaign; template targeting; one fresh CBO adset; both arms as new media and ad entities. A parity verifier must fail closed on any narration-hash, script-hash, aspect, duration, caption, or media-cardinality mismatch.

Decision logic:

• Control (corner) takes allocation and ROAS ≥ 1.45 → corner bed is load-bearing; keep corner as the default for monk-authority diagnosis scripts despite its higher render cost.
• V1 (plain authority) matches or beats Control → the narration carries; use the cheaper full-frame authority tier by default.
• Either arm below $50 lifetime while the other delivers → classify that arm NO-DELIVERY and call the delivered allocation winner.
• Both below $50 lifetime → whole-adset rejection; no format conclusion.

Spend contract: Fresh CBO adset in campaign 120237762972300539; no adset budget. AdLauncher fixed CBO floor: 2000¢ per adset, not per arm.

Production route: AdChain experiment mode → both zero-cost SoulGem preflights → generate one fresh authority narration master → render full-frame authority from that master and seed the exact same master into the fresh corner workdir → deterministic media/audio/script/caption parity gates → live Sheet/product/congruence gates → one AdLauncher request with Control + V1 → normal release cadence. No forced release.

Production integrity: Both arms use narration master sha256 e38779009dd77c1f4735a9897329069b0f694ea6ce6cb74651a4f948808a6427 (80.53551s) from one ElevenLabs request. Control final sha256 3caa50ec60cfdbe8346f3e73c6fd936fed9b0ee906e50b8a91d990fca179f848; V1 final sha256 80edac77e4f003c49e27a32a9d95f93f5363244789ba99b221444d060cf31769. Both final files are H.264/AAC, approximately 62s, caption-sensecheck PASS. Transcript timing confirms Americans → mostly and and → within each have a 0.0s word gap.

AdLauncher request ID: produced-tbf-exp054-corner-vs-authority-cadence-repair-20260820-6dbbbb07

AdLauncher queue state: live (publisher proof; register auto-synced)

AdLauncher request SHA-256: fa5a2eaec0746a23af031a236ee557b2804645a13c7c2d5bffc400b4ea0563dd

Queue receipt: One canonical request contains exactly two queue-owned media in Control → V1 order, with hashes matching the registered production files. Strict request, custody, targeting, copy, $20 CBO-floor, exact-one-target, and no-platform-ID checks all passed. No publisher receipt exists and no release was forced.

Result: Control — top spend $18.15 on 2026-08-24 (latest day). Total adset spend $83.66 over 4 days. Ranking: Control $18.15 latest / $63.62 total; V1 $4.62 latest / $20.04 total NO-DELIVERY (<$50 lifetime). Auto-called per 4d/$50 doctrine.

Element tested: The big-idea territory for The Ailing avatar — a Lane A exploration batch of five maximally different concepts, each a full first-person video script, run for Meta's allocation ranking.

Status: Complete

Result vs decision logic: C5-delivers branch fired: Health Is Fortune won the last day over $5 (2026-08-05, C5 $7.71 / $19.08). Fortune-redefinition is the first-class Ailing territory. C1 peer-testimonial not_fired ($0.31 / $2.77). C2 caregiver-drain not_fired ($1.37 / $3.40). C3 failed-remedies not_fired (second, $7.05 / $8.46). C4 family-healing not_fired ($3.07 / $9.31). Incumbent-comparison branch not_fired — C5 is the batch winner; the monk-authority probe remains the conversion benchmark, not the territory winner inside this first-person set.

Learnings:

• [avatar] Write Ailing first-person scripts in the health-is-fortune territory: fortune is a morning without the weight; money luck arrives last.
• [family] Do not default new Ailing characters to peer testimonial, caregiver drain, or household-blessing. Those territories lost.
• [avatar] Failed-remedies (crystals are passive, copper conducts) is the only runner-up worth a later isolation. Peer testimonial is parked.
• Transfer: Cheapest test: rewrite the Health Is Fortune thesis as a Champa monk-authority script and rank it against the live monk-message Control (Exp 44).

Result: C5 Health Is Fortune — last day over $5 was 2026-08-05; C5 $7.71 that day / $19.08 lifetime. Ranking: C5 $7.71 / $19.08; C3 Failed Remedies $7.05 / $8.46; C4 Family Healing $3.07 / $9.31; C2 Caregiver Drain $1.37 / $3.40; C1 Peer Testimonial $0.31 / $2.77. Adset $19.51 that day, $43.02 lifetime through 2026-08-14. Operator-called 2026-08-15 on last day with over $5.

Family: ailing-territory

Thread origin: New thread — no predecessor experiment, so Parent: is intentionally omitted (the field is optional and takes only Exp NNN / CODE Exp NNN). Source observation is the unregistered 2026-07-25 AdChain translate probe (adchain/runs/tbf-ailing-monk-message-remake-20260725), itself derived from the cursed-monk-message winner.

Ad set name: 20260730 - EXP - TBF - Exp 40 Ailing Territory Batch - TBF Sales Page (Ailing)

Live ad set name: 20260730 - P - EXP - TBF - Exp 40 Ailing Territory Batch - TBF Sales Page (Ailing)

Ad set ID: 120247051861210539

AdLauncher request ID: produced-tbf-exp040-ailing-territory-batch-20260730-9ac76320

AdLauncher queue state: live (publisher proof; register auto-synced)

AdLauncher request SHA-256: c1cc444757857bee1bf02a197a6e73deb75ace3a74eeec66541dea95c949e953

Avatar: The Ailing (Avatar 7)

Campaign ID: 120246833001370539

Campaign: 20260725 - TBF | Ailing - Testing (resolved 2026-07-30 from the incumbent's AdLauncher request.json + publisher revalidation receipts; the earlier "resolve from adset 120246901226400539" note misparsed as a campaign ID)

Destination shorthand: TBF Sales Page (Ailing)

Creative type: Video, first-person female speaker (five new characters, none exist yet for Ailing). SoulGem format matched to the incumbent probe's production pipeline — resolve from the incumbent run's receipts at AdChain intake, never from the adset name token.

Variant map: (no Control — see Design note)

• 20260730 - EXP - TBF - Exp 40 C1 Peer Testimonial Ailing Territory Batch - TBF Sales Page (Ailing) — Ginther-skeleton ill-peer testimonial (Doris; uncelebrated peer, population-level survey proof 34/57, modest dignified outcomes). c1-peer-testimonial.txt sha256 8084b10c5f482c5143ad209fd2afc4a2f55f3054fa1a4d9cebff72fb05d04e19
• 20260730 - EXP - TBF - Exp 40 C2 Caregiver Drain Ailing Territory Batch - TBF Sales Page (Ailing) — caregiver spent her fire on everyone else; root chakra as a well drawn down by giving. c2-caregiver-drain.txt sha256 5d130dfc33a686e36e4d2a241af87d2e0e51c086716ceb066cb2a88fcadc44e1
• 20260730 - EXP - TBF - Exp 40 C3 Failed Remedies Ailing Territory Batch - TBF Sales Page (Ailing) — crystals/sound healing are passive, a block needs a conductor; copper conducts. c3-failed-remedies.txt sha256 08900d7dd2a66eeba8b5247dbf0bee559b29232ac506e80284beaf6f13985df0
• 20260730 - EXP - TBF - Exp 40 C4 Family Healing Ailing Territory Batch - TBF Sales Page (Ailing) — blessing flows through the household, enters through whoever holds the family up. c4-family-healing.txt sha256 7ca0e289c7b557ad59b860aec2cec2ba416039752d54549732668bbf3f383409
• 20260730 - EXP - TBF - Exp 40 C5 Health Is Fortune Ailing Territory Batch - TBF Sales Page (Ailing) — attacks the money-luck framing: fortune is a morning without the weight; money luck arrives last. c5-health-is-fortune.txt sha256 7ca568f703936ab87f5fccc41cccfefd3ef9a2ed4d13ebccd779be9f5fa7a571
• Scripts: output/junglelab/2026-07-30-tbf-ailing-territory-batch/ (brief.md carries the avatar grounding, claims gates and hold-constants)

Design note — no control arm: Lane A exploration batch — all five arms are new creatives so no arm starts with a creative-history allocation advantage. The live incumbent probe (adset 120246901226400539, monk-message antagonist-swap, ~1.6 ROAS at $158 spend as of 2026-07-30) stays untouched in its own adset and rides to its own 14-day verdict; it is the informal benchmark, never an in-adset arm. Approved by Edward 2026-07-30 with the batch.

Source observation: The Ailing surface has exactly one creative ever run — the 7/25 monk-message translate probe — and it cleared the 1.45 ROAS bar early ($158 spend, ~$254 revenue, $124 of it on 7/29 as Meta ramps). The avatar converts, but only one big-idea territory has been sampled. No Banana rows exist for this avatar (fully unseen).

Hypothesis: The monk-authority frame is not necessarily the Ailing avatar's best territory — the avatar doc (long-running pain, failed remedies, caregiver burden, family sickness, "health is the real fortune" core belief) suggests at least five distinct big ideas, and which ones Meta's allocation validates determines where all future Ailing creative effort goes.

Novel variable: The big idea / territory only. Each arm is a genuinely different thesis in the avatar's own idiom.

Hold constant: Mechanism core (copper mined in Tibet, hand-worked, Drepung full-moon blessing, dominant hand), offer close (500 blessed bracelets, "click below to see if any of these bracelets is still available"), first-person voice, 399–487 word band, claims posture (fortune promise; physical improvements only as softened reported experience), destination LP (match the live incumbent's exact destination at enqueue), audience/campaign lineage, production pipeline identical across all five arms.

Delivery risk: Five all-new arms in one fresh CBO adset — Lane A by design; allocation ranking IS the deliverable. Pre-registered per arm: lifetime spend < $50 = NO-DELIVERY = Meta rejection of that territory. Expected outcome is 1–2 delivering arms. All five < $50 lifetime → whole-adset rejection; the read is void, not negative (the failure is the ad set, and the incumbent's continued delivery tells us the avatar itself still converts).

Decision logic:

• C1 (peer testimonial) delivers → the Ginther formula transfers to Ailing; the 017 character factory gains a second avatar to produce for.
• C2 (caregiver drain) delivers → the caregiver sub-segment is reachable and self-sacrifice framing carries it; future Ailing hooks can target the giver identity directly.
• C3 (failed remedies) delivers → the failed-spiritual-tools empathy frame is the entry point; future hooks open on the drawer of crystals, not the pain.
• C4 (family healing) delivers → buying-for-another is a live purchase motive; unlocks household/gift framing across TBF.
• C5 (health-is-fortune) delivers → the avatar rewards reframing the product promise itself; fortune-redefinition becomes a first-class TBF angle.
• Winner comparison vs incumbent (informal, cross-adset): any arm whose delivered ROAS approaches the monk-message probe's justifies an isolation follow-up in that territory; the monk-authority frame keeps the crown only if no batch arm delivers comparably.

Transferable rule: If this resolves, future TBF Ailing ads will be built differently: creative production for this avatar is aimed only at Meta-validated territories from this map — applied to every subsequent Ailing script, character and hook — instead of defaulting to more monk-message derivatives.

Why it matters: Ailing is a brand-new proven-converting surface with zero territory map. One adset buys the creative roadmap for the entire avatar and decides whether the 017 factory, the caregiver angle, or the incumbent's authority frame gets the next production dollars.

Spend contract: Fresh CBO adset in the Ailing testing campaign, no adset budget. AdLauncher fixed CBO floor: 2000¢ — per adset, not per arm; five arms do not multiply the floor.

Production route: AdChain experiment mode — five renders, five new female speaker characters (character creation required; no Ailing characters exist). Production format and portrait/voice policy matched across all arms. Checkpoint 2: locked scripts (cleared at log time — SHAs above). Checkpoint 3: pre-launch — nothing goes live on Meta without Edward's go.

Interim read (2026-08-06, day 3 of delivery): Batch lifetime $42.99 across all 10 ads (5 arms x 2 fan-out destinations), zero purchases. Per-arm allocation: C5 $19.08 > C4 $9.31 > C3 $8.46 > C2 $3.37 > C1 $2.77 — every arm far below the $50 no-delivery bar. Same window (8/3–8/6), the incumbent monk-message adset took $605.21 at 1.81 ROAS ($1,385.14 lifetime at 1.57). No verdict called — the experiment rides to its own call — but the emerging shape is whole-batch starvation against a scaling incumbent in the same CBO campaign. Design confound noted for the follow-up: all five arms changed territory AND speaker/format simultaneously (new female first-person speakers vs the proven Champa monk-authority frame), so starvation cannot distinguish territory rejection from format rejection. Exp 44 is the disambiguation.

Element tested: The big-idea territory for The Ailing avatar, retested inside the avatar's one proven creative frame — top Exp 40 territories rewritten as Champa Dorje monk-authority scripts, ranked by Meta allocation against the live incumbent script as in-adset Control.

Status: Complete

Result vs decision logic: Branch 1 indeterminate: V1 Health Is Fortune cleared $50 at $74.72 lifetime but drew $0.00 on the final day while Control took $38.57 and $220.37 lifetime; ROAS is not in the register, and its allocation never approached the Control's. Branch 2 not fired: V1 cleared $50, so this was not a full double rejection; V2 Family Healing $27.23 and V3 Failed Remedies $20.87 were NO-DELIVERY. Branch 3 not fired: Control delivered, so the canary held and the read is valid. Branch 4 (cross-adset benchmark) indeterminate: the incumbent's own ROAS is not recorded here. The incumbent hard-years script keeps the crown; Health Is Fortune delivered in the monk frame but lost.

Learnings:

• [avatar] The incumbent hard-years monk script remains the Ailing Control; Health Is Fortune delivered in the monk frame but lost to it, so keep it as a secondary territory rather than the lead.
• [family] Family-healing and failed-remedies were not funded in the monk frame either, after starving as first-person scripts in Exp 40; retire both territories for the Ailing.
• [avatar] Sample new territories inside the avatar's proven frame before funding new speakers; the monk-frame retest separated territory from format where Exp 40 could not.
• Transfer: Cheapest test: hold the incumbent script byte-identical and swap only the three pain-recognition sentences (body-specific versus low-days) to learn whether sharper pain naming outdraws the generic incumbent.

AdLauncher request ID: produced-tbf-exp044-ailing-monk-frame-territory-20260816-c37888fd

AdLauncher queue state: live (publisher proof; register auto-synced)

AdLauncher request SHA-256: 79963ddeee8fc4a5edd562b528144fa0415f96820cd4002a488828d552b0ec62

Queue receipt: Enqueued 2026-08-16 from batch output/meta-uploads/tbf-exp044-ailing-monk-frame-territory-20260816/. Queue custody holds exactly four media in Control → V1 → V2 → V3 order: Control sha256 fbdee260eab772eb8a2c9eb3cc7e183aabaed30accf8a3c0fe3e9754b21072b5 (incumbent live bytes verbatim — sha equals the declared media sha of live request produced-tbf-ailing-monk-message-remake-20260725-authority-captioned-v2-d4ba8996); V1 cfd8c1981a9a804a2a84cd98c4bfdc14849f33fdd33007f4793832356f7e4381; V2 fae767da28a7721ccdc1aafaf5299be1f163c4faf8a363e18dadf8a90eb78e6b; V3 1ab6495b26a861ed5ed47d6debd7e852cf4e87aaad3b0f4bc209211ba6900c5a. Renders completed 2026-08-16 (V2/V3 resumed after the 2026-08-06 crash; no paid side effects had occurred), all arms 576x704 captioned with caption sensecheck pass. Ad↔page congruence gate: pass (V-arms 3/3 votes; split-aware vs live hero_solution_first_v1). Template adset fresh ACTIVE readback: incumbent 120246901216480539 (Exp 40's template 120246833001390539 is now ARCHIVED). Negative guard proof: avatar mutation raised JUNGLELAB_REGISTER_MISMATCH, queue fingerprint unchanged. No platform IDs created; no release forced.

Family: ailing-territory

Parent: Exp 40

Move: Format transposition / Lane B disambiguation — Exp 40 changed territory AND speaker/format simultaneously; this rerun holds the proven format constant so territory is the only live variable.

Avatar: The Ailing (Avatar 7)

Campaign ID: 120246833001370539

Campaign: 20260725 - TBF | Ailing - Testing

Destination shorthand: TBF Sales Page (Ailing)

Live destination: https://discover.karmaitems.com/tbf-2507-ail-sales-page-ev-001/ (match the incumbent's exact destination at enqueue — resolve from receipts, never from the adset name token)

Creative type: Video, Champa Dorje authority talking head — the incumbent's exact production lineage (SoulGem authority_talking_head, captioned; resolve locked photo/avatar-prompt/voice/caption constants from the incumbent run's receipts at AdChain intake: adchain/runs/tbf-ailing-monk-message-remake-20260725, live request produced-tbf-ailing-monk-message-remake-20260725-authority-captioned-v2-d4ba8996).

Ad set name: 20260806 - EXP - TBF - Exp 44 Ailing Monk Frame Territory - TBF Sales Page (Ailing)

Live ad set name: 20260806 - P - EXP - TBF - Exp 44 Ailing Monk Frame Territory - TBF Sales Page (Ailing)

Ad set ID: 120247419990990539

Live ad IDs:

• Control: 120247419997120539, 120247419998330539
• V1: 120247419999350539, 120247420001320539
• V2: 120247420002820539, 120247420004130539
• V3: 120247420005420539, 120247420006320539

Variant map (CP2 locked 2026-08-06 — run dir output/junglelab/2026-08-06-tbf-exp044-ailing-monk-frame-territory/, brief.md sha256 14983e9726ce331dc1243e20ca99ff972ff90349243162a1fd403edb6107c228):

• Control — the live incumbent monk-message Ailing creative reused verbatim (exact captioned bytes, new ad entity, no re-render) per Exp 43 Control doctrine: byte-identical media under a new ad ID carries no ad-level history advantage.
• V1 Health Is Fortune — Exp 40 C5 territory (fortune redefined as a morning without the weight; money luck arrives last) rewritten as a Champa monk-authority script. v1-health-is-fortune.txt sha256 5b69c5551728444dbee5c7d816d5ee5f5232e3affdef10cefd0b8c339159e1ca · 430 words
• V2 Family Healing — Exp 40 C4 territory (blessing flows through the household, enters through whoever holds the family up) rewritten as a Champa monk-authority script. v2-family-healing.txt sha256 4d0c68a789b70c32f0464265873f7e4e0e5d58b69d980d4f9d2140e184500c14 · 470 words
• V3 Failed Remedies — Exp 40 C3 territory (crystals/sound healing are passive; a block needs a conductor; copper conducts) rewritten as a Champa monk-authority script. v3-failed-remedies.txt sha256 c36bc266242d628a02a6c81b027ab76f66444b894f610ced935cdc48682cd1d1 · 444 words
• Script validation (2026-08-06): verify_generic_script --wc-min 399 --wc-max 487 → PASS all 3 (all_pass=true). Main-session gate greps: banned-word/medical/Cursed-leakage lists zero occurrences in all arms; Champa Dorje identification, root chakra, copper, Drepung, full moon, dominant hand, 500-bracelets scarcity, and the exact click-below close present once in each; softeners ≥2 per script with physical outcomes only downstream of a softener; no arm reuses the incumbent's accumulated-years thesis.
• Excluded: C1 Peer Testimonial (structurally untransposable — a peer testimonial requires the peer speaker; parked with its near-zero Exp 40 allocation) and C2 Caregiver Drain (last in Exp 40 allocation at $3.37; parked, revivable if V2 family-framing delivers).

Source observation: Exp 40 interim (2026-08-06, day 3): whole-batch starvation — $42.99 lifetime across five arms, zero purchases, no arm near the $50 bar — while the incumbent monk-authority adset in the same CBO campaign took $605.21 at 1.81 ROAS in the same window ($1,385.14 lifetime at 1.57, scaling). Exp 40's arms differ from the incumbent on TWO axes at once: big-idea territory and speaker/format (five all-new female first-person speakers vs the proven Champa monk-authority frame). The starvation is therefore ambiguous between "Meta rejected the territories" and "Meta rejected the unproven speaker/format."

Hypothesis: At least part of Exp 40's starvation is format rejection, not territory rejection. Territories re-expressed in the avatar's proven monk-authority frame will earn allocation; their ranking against the same-frame Control produces the clean territory map Exp 40 was designed to buy.

Novel variable: Big-idea territory only. Speaker, format, and production constants are identical across all arms including Control.

Hold constant: Champa Dorje character and the incumbent's full production lineage (locked portrait photo, avatar prompt, voice, caption style, aspect, playback policy — exact values resolved from incumbent receipts at intake); mechanism core (copper mined in Tibet, hand-worked, Drepung full-moon blessing, dominant hand); offer close (500 blessed bracelets, "click below to see if any of these bracelets is still available"); 399–487 word band (same band as Exp 40 and the incumbent gates, for cross-experiment comparability); claims posture per the Ailing brief gates (no "illness", no medical vocabulary, fortune promise, physical improvements only as softened reported experience, no Cursed-avatar leakage); destination, Sheet-canonical headline (Let Fortune Return After Years of Running on Empty) and primary text from the incumbent's live request; campaign lineage; one fresh CBO adset; intrinsic destination fan-out at DEFAULT (ENABLED — matches Exp 40's cardinality: 4 arms x 2 destinations = 8 Meta ads).

Delivery risk: The same CBO campaign hosts the scaling incumbent adset and (until called) the Exp 40 batch adset — the strongest headwind is budget concentration on the incumbent. The verbatim-bytes Control doubles as a delivery canary: if even byte-identical proven creative starves in the fresh adset, starvation reads as adset/CBO dynamics, not creative rejection, and the read is void per Exp 40's whole-adset doctrine. Pre-registered per arm: lifetime spend < $50 = NO-DELIVERY = Meta rejection of that territory-in-frame.

Decision logic:

• Any territory arm ≥ $50 with ROAS approaching Control → territory validated; Exp 40's starve was (at least partly) format rejection. The validated territory enters the Ailing roadmap; first-person female production for this avatar happens only in Meta-validated territories, if at all.
• Control delivers, all territory arms < $50 → double rejection: those territories failed in both frames. The territory map closes; monk-message keeps the crown and Ailing production defaults to monk-frame script derivatives (an Exp 41-style Champa script-territory batch adapted for Ailing becomes the natural follow-up).
• Control itself < $50 → delivery-canary trip: whole-adset read void (proven bytes starving = CBO/adset dynamics), not a creative verdict; no territory is condemned on this read.
• Cross-adset informal benchmark stays the live incumbent's own delivered ROAS, as in Exp 40.

Transferable rule: If this resolves, batch exploration for a new avatar changes fleet-wide: never sample new territories and a new speaker/format in the same arms when the avatar has exactly one proven creative — transpose territories into the proven frame first (cheap: same character, no character creation), and only fund new-speaker production inside territories the proven frame has validated.

Why it matters: Ailing is the fleet's newest converting surface and its incumbent is scaling ($605 at 1.81 in 3.5 days). Exp 40 spent its read on a confounded design; this adset buys the same territory roadmap for three renders and zero new characters, and simultaneously stress-tests whether fresh-adset starvation in this campaign is even a creative signal.

Sequencing note: Production (CP2 scripts → renders) can proceed immediately — no dependency on Exp 40. Launch is the decision point: recommend holding release until Exp 40 is called (its arms are pre-registered against the $50 bar; earliest natural call ~2026-08-17 at 14 days) so the incumbent, Exp 40, and Exp 44 do not split one CBO purse three ways. Edward may override and launch earlier; the Control canary makes the read partially self-correcting either way.

Spend contract: Fresh CBO adset in campaign 120246833001370539 (20260725 - TBF | Ailing - Testing), no adset budget. AdLauncher fixed CBO floor: 2000¢ — per adset, not per arm.

Production route: AdChain experiment mode — three new SoulGem authority_talking_head renders (Champa Dorje exists; zero character creation) + Control verbatim media reuse (zero renders). Checkpoint 2: locked scripts with SHAs logged here. Checkpoint 3: pre-launch — nothing goes live on Meta without Edward's go.

Result: Control — top spend $38.57 on 2026-09-02 (latest day). Total adset spend $343.19 over 15 days. Ranking: Control $38.57 latest / $220.37 total; V1 $0.00 latest / $74.72 total; V2 $0.00 latest / $27.23 total NO-DELIVERY (<$50 lifetime); V3 $0.00 latest / $20.87 total NO-DELIVERY (<$50 lifetime). Auto-called per 4d/$50 doctrine.

Element tested: Social proof on the Ailing monk frame — whether adding a first-party proof beat (the monks' own longevity as living evidence) to the avatar's best monk-authority script earns more Meta allocation than the script without it, and if so which proof kind (living proof vs external credentialed witness) carries it.

Status: Complete

Result vs decision logic: Branch 2 fired: Control (the proof-free Exp 44 winner) won allocation with $60.61 lifetime while both proof beats stayed under $50 and were rejected — the authority frame alone suffices for Ailing, and neither proof beat is load-bearing. Branch 1 (any beat arm ≥ $50 and outallocating Control) did not fire. Branch 3 (Control wins with a delivered beat arm) did not fire — no beat cleared the bar. Branch 4 (Control itself under $50 canary trip) did not fire — Control delivered $60.61, so the read is valid.

Learnings:

• [avatar] Don’t bolt a proof beat onto the Ailing monk-authority script — the monk’s own authority is already the proof, and both proof kinds (living proof and the visiting doctor’s disbelief) lost to the proof-free control; keep Ailing scripts proof-free by default.
• [family] In this family the monk-authority frame is self-proving — proof is a burden, not an asset, so don’t import the proof-heavy structure from other avatars’ winners into the monk frame.
• Transfer: Cheapest test: run the same proof-beat insertion on a non-monk authority frame for Ailing (a doctor or researcher authority) — if the beat helps there, proof-need is keyed to the authority source (monk = self-proving, researcher = needs proof), which widens the rule from this avatar to a family-wide doctrine.

Family: ailing-territory

Parent: Exp 44

Move: Translate (Overlooked proof doctrine ported to Ailing) / dimension sweep: social proof — untested for this avatar per the Dimension Coverage table.

Avatar: The Ailing (Avatar 7)

Campaign ID: 120246833001370539

Campaign: 20260725 - TBF | Ailing - Testing

Destination shorthand: TBF Sales Page (Ailing)

Ad set name: 20260815 - EXP - TBF - Exp 51 Ailing Proof Beat - TBF Sales Page (Ailing)

Live ad set name: 20260815 - P - EXP - TBF - Exp 51 Ailing Proof Beat - TBF Sales Page (Ailing)

Ad set ID: 120247530570720539

Live ad IDs:

• Control: 120247530571900539, 120247530572500539
• V1: 120247530573350539, 120247530573940539
• V2: 120247530574580539, 120247530575450539

Variant map:

• 20260815 - EXP - TBF - Exp 51 Control Ailing Proof Beat - TBF Sales Page (Ailing) — Exp 44 V1 Health Is Fortune verbatim: control-health-is-fortune.txt sha256 5b69c5551728444dbee5c7d816d5ee5f5232e3affdef10cefd0b8c339159e1ca · 430 words; reuse the exact live captioned media sha256 cfd8c1981a9a804a2a84cd98c4bfdc14849f33fdd33007f4793832356f7e4381 under a fresh ad entity.
• 20260815 - EXP - TBF - Exp 51 V1 Ailing Proof Beat - TBF Sales Page (Ailing) — Control + living-proof beat (monks' own longevity, Champa first-person observation): v1-living-proof.txt sha256 a4fe9eb352e2520285c6b203bb6bf7b30b74103dbbc0acf14fb514f5e0c003c4 · 495 words. Inserted beat sha256 174b2cf47631f7e3f0c6b480765583cb11ea006aa61f3ab6904573d992d3944f · 65 words.
• 20260815 - EXP - TBF - Exp 51 V2 Ailing Proof Beat - TBF Sales Page (Ailing) — Control + external-witness beat (visiting doctor's disbelief; credentialed-outsider proof kind, no invented statistics): v2-external-witness.txt sha256 6d12bde5bd2051c8307b70588f2c306c07d77a84818b057c36cdaab5ef550168 · 488 words. Inserted beat sha256 96bc3facc69057b9c4a1dad38315112e4feaf23aee2a859d1f3a4388c9159926 · 58 words.

Control bind receipt: output/junglelab/2026-08-15-tbf-exp051-ailing-proof-beat/control_bind.json proves the Control source SHA, all three full-script SHAs, and byte identity outside the single inserted span. Generic script validation passed all three at 400–520 words. SoulGem challengers lock the Exp 44 V1 portrait (c9768735641ed…), avatar prompt (f5485855fdab…), Champa registry voice, 540p, 1.0x playback, and neutral-box captions.

• Beat spans + bind procedure: output/junglelab/2026-08-15-tbf-exp051-ailing-proof-beat/ (design.md documents the seam and the byte-identity assertion outside the span).

Render receipt: V1 living proof completed at 174.84s, 576x704 H.264/AAC, caption sensecheck 1.0, media sha256 ae9537b9d378a994e66aadbff120e50860f6a6972e05cd4a05eaca4180e331bc. V2 external witness completed at 178.08s, 576x704 H.264/AAC, caption sensecheck 1.0, media sha256 fe39f9918c95b5297238b0e2edf1f1fdac2e3f4054b16191a6cdb7d74158ef3c. Both outputs are faststart; visual contact-sheet review passed.

AdLauncher request ID: produced-tbf-exp051-ailing-proof-beat-20260822-86e6e4b3

AdLauncher queue state: live (publisher proof; register auto-synced)

AdLauncher request SHA-256: 9278f57806af37ab7dbac9ed1daf4d83fbbd7ab85b3a28c1eb589dd336e19de7

Queue receipt: Enqueued once on 2026-08-22 from output/meta-uploads/tbf-exp051-ailing-proof-beat-20260822/. Queue custody contains exactly three media in Control → V1 → V2 order: Control sha256 cfd8c1981a9a804a2a84cd98c4bfdc14849f33fdd33007f4793832356f7e4381; V1 sha256 ae9537b9d378a994e66aadbff120e50860f6a6972e05cd4a05eaca4180e331bc; V2 sha256 fe39f9918c95b5297238b0e2edf1f1fdac2e3f4054b16191a6cdb7d74158ef3c. Fresh Sheet/product/ad↔page congruence gates passed; strict post-enqueue readback passed all checks, including exact request hash, bytes, names, targeting, $20 CBO floor, exactly one matching request, and absence of a publisher receipt.

Source observation: Every Ailing candidate winner already carries customer-report proof ("many people tell me..."), but no Ailing creative has ever used the avatar doc's stated marketing appeal — the story of the Tibetan monks' own health and longevity — as first-party living proof. The Dimension Coverage table (seeded 2026-08-15) shows social proof fully untested for this avatar. Cross-avatar precedent cuts both ways: proof is load-bearing for Overlooked (Exp 6/7: removing the researcher/statistical block stopped delivery), while the Cursed monk-message thrives without any proof beat.

Hypothesis: A proof beat inserted into the proven monk-authority frame earns more allocation than the same script without it, because the Ailing avatar (burned by failed remedies, fearful nothing will work) needs evidence the mechanism is real before accepting the promise.

Novel variable: Presence and kind of an inserted proof beat. Each V-arm differs from Control by exactly one declared span inserted immediately before the shared seam sentence ("At Drepung Monastery, we made this bracelet so that people can wear such copper on them"), verified present exactly once in all four Exp 44 candidate winners. The added length (~60-65 words) is part of the declared variable; nothing is removed in compensation.

Hold constant: Everything outside the inserted span, byte-identical: the Exp 44 winner's full script, Champa Dorje character and the incumbent production lineage (SoulGem authority_talking_head, locked portrait/avatar-prompt/voice/caption constants resolved from receipts at AdChain intake), mechanism core, offer close and 500-bracelets scarcity, claims posture per the Ailing brief gates (no medical vocabulary, no "illness", physical outcomes only as softened reported experience — both beats are observational with softeners: "I cannot tell you it is one thing" / the doctor's disbelief in place of numbers), destination, Sheet-canonical headline and primary text from the live winner's request, campaign lineage, one fresh CBO adset, intrinsic destination fan-out at DEFAULT.

Delivery risk: Fresh CBO adset in the same campaign as the scaling incumbent — budget concentration is the headwind. The verbatim-bytes Control doubles as the delivery canary. Pre-registered per arm: lifetime spend < $50 = NO-DELIVERY = Meta rejection of that arm — a real verdict, not an unmeasured rival; delivered-winner allocation IS the verdict and rejected arms are never re-run.

Decision logic:

• Any beat arm ≥ $50 and outallocated over Control → the proof beat is load-bearing for Ailing; the winning beat kind becomes a default block in every future Ailing script brief. If both beat arms deliver, their ranking calls the proof KIND (living vs external witness).
• Control wins allocation and both beat arms < $50 → Meta rejected the beats: the authority frame alone suffices for Ailing; future scripts do not add proof beats; never re-run the rejected beats.
• Control wins allocation with at least one beat arm delivering ≥ $50 → proof is not load-bearing but not rejected; keep the frame proof-free by default, park the delivered beat kind as revivable.
• Control itself < $50 → delivery-canary trip: whole-adset read void (Exp 44-winner bytes starving = CBO/adset dynamics, not a creative verdict); no beat is condemned on this read.

Transferable rule: If this resolves, future TBF Ailing ads will be built differently: every Ailing script brief either carries a proof beat of the validated kind as a standard block, or is confirmed proof-free — applied at [avatar] scope to all future Ailing creative.

Why it matters: Social proof is the highest-precedent untested dimension for the fleet's newest converting avatar — load-bearing for one sibling avatar, absent from another's winner. One three-arm adset with zero new characters and two span insertions settles whether every future Ailing brief includes a proof block, and which kind.

Sequencing note: Launch only after Exp 44 is called — the Control does not exist until then, and launching earlier would split the Ailing CBO purse three ways with the incumbent and Exp 44. Production is two span insertions + renders once the Control binds; no character creation.

Spend contract: Fresh CBO adset in campaign 120246833001370539 (20260725 - TBF | Ailing - Testing), no adset budget. AdLauncher fixed CBO floor: 2000¢ — per adset, not per arm.

Production route: AdChain experiment mode — Control verbatim media reuse if the Exp 44 winner's render is reusable byte-identical (per Exp 43 Control doctrine), else re-render from bound bytes; two new authority_talking_head renders for the beat arms (Champa exists; zero character creation). Checkpoint 2 at bind: full arm scripts with SHAs logged here plus byte-identity-outside-span assertion. Checkpoint 3: covered by the standing GO of 2026-08-15 — after deterministic gates pass, enqueue and launch without asking Edward again.

Result: Control — top spend $18.90 on 2026-08-28 (latest day). Total adset spend $103.10 over 4 days. Ranking: Control $18.90 latest / $60.61 total; V2 $5.24 latest / $16.05 total NO-DELIVERY (<$50 lifetime); V1 $3.84 latest / $26.44 total NO-DELIVERY (<$50 lifetime). Auto-called per 4d/$50 doctrine.

Element tested: Closing structure — whether a reflective valedictory before the CTA costs more than it buys.

Status: Planned — Checkpoint 1 variant prose approved by Edward 2026-08-11; production authorized; next matched three-arm render, then launch under this standing GO after deterministic gates pass

Family: miranda-close-structure

Altitude: Rung 7 — winner defense / close structure

Ad set name: 20260725 - EXP - TBF - Exp 31 Abrupt Close vs Winddown - Expert Healer (Overlooked)

Destination shorthand: Expert Healer (Overlooked)

Variant map:

• 20260725 - EXP - TBF - Exp 31 Control Abrupt Close vs Winddown - Expert Healer (Overlooked) — live Miranda VSL verbatim re-upload under a fresh ad ID (875w)
• 20260725 - EXP - TBF - Exp 31 V1 Abrupt Close Abrupt Close vs Winddown - Expert Healer (Overlooked) — five-sentence valedictory close deleted, CTA rejoined as "Act now." (826w)
• 20260725 - EXP - TBF - Exp 31 V2 Relocated Close Abrupt Close vs Winddown - Expert Healer (Overlooked) — same valedictory MOVED to sit immediately before the 50-bracelets-per-month scarcity block, same CTA rejoin (872w)

Result: Pending

Date: 2026-07-25

Avatar: The Overlooked

Creative type: First Person Contemporary VSL, ~875w

Source observation: Ad 120246521495270539 — 20260713 - MIRANDA ITER - TBF - The Overlooked Avatar - Testimonial-Style UGCs - Miranda - Expert Healer (Overlooked), $11,314.32 90-day spend, launched 2026-07-13, currently the scaled Overlooked workhorse. 875 words, sha256 461aceaf0143b2585e529b7a518dca64e5a5815b4db61628d49d29a0935e4be6, source on disk at /home/runner/.hermes/output/junglelab/2026-07-25-mrbeast-derived-exp-031-033/sources/exp031-control-miranda.txt.

Hypothesis: For long-form first-person VSLs, signalling that the story is over before the CTA sheds prospects who would otherwise convert; the cost is in the END-SIGNAL, not in the word count.

Prospect claim: The Overlooked currently feels the story is being told TO them and stays while it is unresolved; the variant works iff a reflective closure reads as permission to leave rather than as earned resolution.

Placement: Product-aware — deep in the asset; the prospect already has the full mechanism.

Novel variable: Presence and position of the valedictory close.

Declared coupled variable: V1 loses 49 words against Control (875 → 826); V2 holds length (875 → 872, the 3-word CTA rejoin only). That asymmetry is exactly why V2 exists — it separates end-signal position from length.

Hold constant: Everything before the valedictory is byte-identical across all three arms; same Miranda render/production, same audience, same canonical Sheet copy/destination for Expert Healer (Overlooked), same CBO floor.

Decision logic:

• V1 AND V2 both beat Control → the end-signal position is load-bearing independent of length; strip or relocate valedictories across all long-form VSLs.
• Only V1 beats Control → the effect is length, not end-signalling; the finding is about compression and belongs with Exp 14 (richard-champa-message-compression).
• Neither beats Control → reflective closure is earning its place for this avatar; stop testing tails on Overlooked.
• Any arm under $50 lifetime spend → NO-DELIVERY per Meta-rejection doctrine; no re-run of the starved arm.

Measurement caveat: A tail change only affects prospects who already reached the tail, so allocation-based reads are weak here. Meta video-view metrics (25/50/75/95%) are the closest analogue to a retention curve and are NOT present in ads_data_master — pulling them is a prerequisite for a confident read on this experiment.

Delivery risk: Fresh-pair CBO adset, never injected into the live Miranda adset, so Control does not inherit an allocation advantage. NO-DELIVERY threshold under $50 lifetime per arm.

Variant scripts on disk: /home/runner/.hermes/output/junglelab/2026-07-25-mrbeast-derived-exp-031-033/variants/exp031-v1-abrupt-close.txt and /home/runner/.hermes/output/junglelab/2026-07-25-mrbeast-derived-exp-031-033/variants/exp031-v2-relocated-close.txt

Status: Complete

Result vs decision logic: No pre-registered Decision logic. V2 (valedictory relocated to sit immediately before the 50-bracelets-per-month scarcity block) won the last day over $5 (2026-08-12, V2 $8.43 / $16.01 vs Control $3.03 / $5.12 vs V1 deleted close $2.07 / $12.19). Keep the valedictory, move it to the scarcity seam, do not delete it.

Learnings:

• [family] Keep Miranda's five-sentence valedictory and place it immediately before the scarcity block. That close structure won.
• [avatar] Do not delete the Overlooked valedictory close. Cutting it lost to relocating it.
• [family] Do not leave the valedictory at the original end position. The scarcity-seam placement is the winning close.
• Transfer: Cheapest test: apply the same valedictory-before-scarcity move to the next Overlooked factory VSL versus its original-end Control.

Result: V2 relocated valedictory — last day over $5 was 2026-08-12; V2 $8.43 that day / $16.01 lifetime. Ranking: V2 $8.43 / $16.01; V1 deleted close $2.07 / $12.19; Control $3.03 / $5.12. Adset $13.53 that day, $33.32 lifetime through 2026-08-14. Operator-called 2026-08-15 on last day with over $5.

AdLauncher request ID: produced-tbf-exp036-overlooked-miranda-close-relaunch-20260807-2d59d63b

AdLauncher queue state: live (publisher proof; register auto-synced)

AdLauncher request SHA-256: 6d8f93b63fe14d44f06e4ac0fd2f2e36207cf55a9a18e4ee4a47e9a997894a7b

Family: miranda-close-structure

Parent: Exp 031

Ad set name: 20260807 - EXP - TBF - Exp 36 Overlooked Miranda Close Structure - Expert Healer (Overlooked)

Live ad set name: 20260807 - P - EXP - TBF - Exp 36 Overlooked Miranda Close Structure - Expert Healer (Overlooked)

Ad set ID: 120247246173160539

Original ad set ID (control-less launch, paused 2026-08-07): 120246961747190539

Avatar: The Overlooked

Campaign ID: 120237768514430539

Destination URL: https://discover.karmaitems.com/tbf-1206-healer-overlooked-advertorial-ev-001/

Destination shorthand: Expert Healer (Overlooked)

Source Trello card ID: 6a64c2f87f3b3ef61d6ca524

Variant map:

• 20260807 - EXP - TBF - Exp 36 Control Overlooked Miranda Close Structure - Expert Healer (Overlooked) — live Miranda VSL verbatim re-upload under a fresh ad ID (875w, per Exp 031 design)
• 20260807 - EXP - TBF - Exp 36 V1 Overlooked Miranda Close Structure - Expert Healer (Overlooked) — five-sentence valedictory close deleted, CTA rejoined as "Act now." (826w, per Exp 031 design)
• 20260807 - EXP - TBF - Exp 36 V2 Overlooked Miranda Close Structure - Expert Healer (Overlooked) — same valedictory relocated to sit immediately before the 50-bracelets-per-month scarcity block, same CTA rejoin (872w, per Exp 031 design)

2026-08-07 control-arm relaunch:

• original adset ID 120246961747190539 launched 2026-07-30 via the Trello daily drain WITHOUT its Control arm because the Frame.io delivery folder lacked the CONTROL VIDEO file at drain time (control-arm doctrine violation)
• old live ad set name: 20260729 - EXP - TBF - Exp 36 Overlooked Miranda Close Structure - Expert Healer (Overlooked)
• original launch carried V1+V2 only; V1 SHA-256 afa8c9a62e075a676198fe3c83b9fa666edbddfec92047d5393fa9d0c0cf8f58 (365,346,575 bytes) and V2 SHA-256 7aa678338f3b37f873839f22bf3d88adaeb920e97d68f2a723e25a0a41ecedc4 (383,539,325 bytes), both byte-identical to the original launch media
• Control SHA-256 dbbbc90a1088fded14a78d26869d352bf5088620da3ee7b1c04f1b99f3fbd02d (383,624,305 bytes), probe 1920x1920, 30 fps, 377.7 s — duration matches V2 exactly, consistent with the verbatim live-ad re-upload design
• relaunch batch path /home/runner/.hermes/output/meta-uploads/tbf-exp036-overlooked-miranda-close-relaunch-20260807/
• old adset paused 2026-08-07 ~04:00Z (adkit readback: paused) and replacement enqueued 2026-08-07 (receipt above); release on normal AdLauncher cadence, no forced release

Element tested: Hook threat framing — which type of curse threat stops the scroll for 17 Jetsetters

Status: Complete

Family: cursed-monk-message

Ad set name: 20260702 - ITER - TBF Curses Monk Message | New Leads - 17 Jetsetters

Date: 2026-07-21

Avatar: 17 Jetsetters

Creative type: Authority UGC monk VSL — SoulGem authority_talking_head_3x4, character Champa Dorje

Context: Six-script batch testing different curse-threat hooks on the same monk-authority skeleton. Four NL variants (A-D) with threat-focused hooks vs two controls (E-F) with myth-busting educational hooks. All scripts converge to identical body copy after the hook/narrative setup (~70% shared text).

Winner: Ad A (NL4) — "Words wrapped in a smile" social-threat hook

• Spend: $441.15 (68.6% of ad set)
• Revenue: $917.18
• ROAS: 2.079
• Hook: "if words can curse you" / "yes, especially when they come wrapped in a smile"
• Post-hook: Social comments as energetic attacks — names actual phrases people say ("Must be nice. You're so lucky. I could never afford that.")

Runner-up: Ad B (NL2) — "Fastest way to go poor" achievement-threat hook

• Spend: $86.81 (13.5%)
• Revenue: $188.50
• ROAS: 2.171 (best in set)
• Hook: "what's the fastest way to go poor?" / "get cursed by someone who resents your success"

Meta rejected (under $50):

• Ad D (NL3, relational betrayal): $30.67 spend, 3.029 ROAS — "friend who secretly envies you" too niche/paranoid
• Ad C (NL1, pattern-oriented): $68.36 spend, 0.492 ROAS — "bad luck makes no sense" too vague
• Ad E (Control V1, myth-busting): $12.32 spend, 0.000 ROAS — educational framing rejected
• Ad F (Control, myth-busting): $3.89 spend, 10.643 ROAS — insufficient spend to judge

Learnings:

• Hook threat specificity beats threat intensity. Ad A's "words wrapped in a smile" is less violent than Ad B's "fastest way to go poor" but more specific and personally relevant. Specific examples ("Must be nice") beat abstract descriptions ("when someone resents you").

• For 17 Jetsetters, social threat beats intellectual curiosity. This avatar responds to "people are cursing you with their words" not "let me explain how curses really work." The myth-busting controls were rejected immediately.

• The monk authority figure works, but the hook must be threatening. Authority is established, but the hook determines whether anyone listens. Educational openings don't stop the scroll for this avatar.

• ROAS and spend allocation diverged. Ad B had the best ROAS (2.171) but Ad A got 5x the spend. Meta's algorithm prioritized reach/scale over immediate conversion efficiency.