Meta winners and losers · 2026-09-10

Learnings Ledger v1

Every ad with $300 or more of lifetime spend labeled, the carriers decomposed, cross-cuts on the whole set, recommendations ranked by expected CPA impact.

2,355 ads with spend · $282,366 lifetime · 92 labeled · pulled 2026-09-10 read-only

Label Definitions Used

CPA target $35 (account cost cap $37; ADHD-named ads ran $29.43 and the account $34.23 to 9-01).

• BREAKTHROUGH: lifetime spend $2,000+ and CPA at or under $35. Volume and efficiency.
• SPEND WINNER: spend $2,000+ and CPA over $35. Meta likes it, the margin does not.
• KPI WINNER: spend $300 to $2,000, CPA at or under $35, 8+ purchases. Efficient, never scaled.
• LOSER: spend $300+ and neither. Read with real money.
• STARVED: $50 to $300. Unread, not falsified. Listed separately.
• NEVER READ: under $50. 2,073 ads. Not evidence of anything.

PART 1 · STATE OF THE ACCOUNT (the numbers first)

1.1 Lifetime, labeled ads (92 ads with $300+)

• 15 Breakthroughs: $152,540, 5,045 purchases, $30.24 CPA. 54% of all lifetime spend.
• 8 Spend Winners: $47,452, 1,181 purchases, $40.18.
• 33 KPI Winners: $25,546, 893 purchases, $28.61. Efficient ads that were never given volume.
• 36 Losers: $23,199, 507 purchases, $45.76.
• 190 Starved ($50 to $300): $33,628 combined with the never-read tail. 12% of spend went to ads that never got a read.

1.2 Right now

• Last 30 days: $48,833, 1,252 purchases, $39.00 CPA.
• Last 7 days: $9,951, 224 purchases, $44.42 CPA.
• The May cohort (slop 4/7/9, offer ad, independent slop 13) is 55% of 30-day spend at $31.62. The August cohort is 20% at $59.26. September is 5% at $60.09.
• Last 7 days on the three ads that carry the account: slop 9 $39.6 (lifetime $33.4), slop 7 $36.1 (lifetime $30.5), slop 4 $43.6 (lifetime $29.0). All three are above their lifetime CPA. The engine is fatiguing this week, not next quarter.
• Only offer ad Copy 3 is holding: $29.3 last 7 days on $615.

1.3 By launch month (ads with $50+ spend)

• Feb: 48 ads, $93,217, $29.62. 6 Breakthroughs, 8 KPI winners. The account's best month.
• Mar: 64 ads, $48,290, $35.38. 2 B, 8 K, 10 losers. First long-native wave.
• Apr: 27 ads, $19,069, $37.69. 2 B, 2 K, 11 losers.
• May: 69 ads, $76,298, $32.93. 4 B, 11 K. The slop engine.
• Jun: no ad created in June reached $50 of spend. A six-week gap with no new engine started.
• Jul: 18 ads, $14,207, $34.82. 1 B (v10), 2 K (freezer marbles, v10 copy).
• Aug: 38 ads, $14,648, $49.49. 0 B, 0 new K, 2 losers, 34 starved. 27 of 38 were long natives or video.
• Sep: 12 ads, $2,375, $59.38. 0 B, 0 K, 2 losers, 10 starved.

PART 2 · WHAT THE WINNERS ARE (nine-question decomposition)

The account runs on one skeleton written in February and re-skinned in May. Reading the bodies of all 15 Breakthroughs, 14 share it:

1. Named-reader qualifier in line 1 ("If you're still the one responsible for getting your teen out of bed" / "If your ADHD teen 'can't' wake up to alarms, you're not dealing with a discipline problem").
2. Failed-solution stack in one line (louder alarms, more alarms, earlier bedtimes, consequences).
3. Mechanism in two or three sentences (sound is easy to ignore in deep sleep / his brain tunes out repetitive sound / it can't tune out vibration on the wrist).
4. "So you become the alarm" (the reader's behavior named back to her).
5. Product in one sentence.
6. Five checkmark outcomes.
7. 60-night trial, "tap through to see how it works."

Length 880 to 1,100 characters. One still image. Title "The Parent-Free Wake-Up Solution" or an ADHD callout.

2.1 neveralarm-pdp family (6 ads, $54,847, 1,772 purchases, $30.95; first 2026-02-01)

• Awareness (hook): problem-aware, named reader ("still the one responsible").
• Valence: mild shame turned into a label ("you have a daily dependency"). Low intensity.
• Desire: relief of responsibility, then independence ("wake themselves up").
• Avatar: any parent who wakes a teen. Not ADHD-specific.
• Angle: you are the system. Stop being the system.
• Mechanism: 2 sentences, sound vs wrist. Nothing more.
• Belief: 60-night trial, "tested in real mornings, not imagined." No authority, no count.
• Positioning: vs louder alarms, more alarms, bedtimes, consequences. Not vs another band.
• Urgency: none stated. Daily pain does the work.
• Hypothesis: the reader recognizes her own behavior in line 1 and gets a name for it. The body promises removal of that behavior, not a transformation. Short enough that the click is a decision, not a read.

2.2 adhd convenience slop 7 / 9 / 4 (one body, four images; $56k, 1,760 purchases, $31 to $33; May)

• Awareness: problem-aware, named reader, ADHD in line 1.
• Valence: blame removal in line 1 ("not a discipline problem"), then the yelling scene ("by 7:14 you've yelled up the stairs three times"). Guilt named, then lifted.
• Desire: vindication first (not laziness, biology), relief second ("you stop being the morning manager").
• Avatar: mother of an ADHD teen who has tried six alarms.
• Angle: it was never discipline, it's wiring, and the wiring has one door.
• Mechanism: 3 lines, tune-out sound / can't tune out wrist. Same as 2.1.
• Belief: "Trusted by 9,000+ parents," 60-night trial, "on a Tuesday."
• Positioning: vs the standard playbook.
• Urgency: none.
• Hypothesis: same skeleton as 2.1 with the ADHD qualifier and one concrete morning scene. The image is the variable that decided volume (slop 7 image "You shouldn't have to wake them five times" took the spend; slop 4 image "sound gets ignored" took the cheapest CPA). This family is the answer to "what converts": identification by symptom, blame removal, simple mechanism, short.

2.3 wakeforcollege / whosgoingtowakehim / never_be_their_alarm_v3 (same body as 2.1, listicle or PDP destination; $28.9k, 979 purchases, $28 to $33)

• Only difference from 2.1: destination (7-reasons listicle) and image. Listicle did not beat PDP for this body (hook-performance 3-18) but did not hurt either.
• Hypothesis: the body is robust to destination and image. It is the skeleton that carries.

2.4 testimonial_v2 ($11,603, 380, $30.5; Feb)

• Awareness: problem-aware, named by consequence ("missing school").
• Valence: blame removal in line 1, then the snowball (missed bus, warnings).
• Desire: reliable wake, then relief.
• Belief: one review line ("hasn't missed school once since starting 11th grade") plus the trial.
• Hypothesis: a single concrete retrospective result inside the skeleton. The ONLY proof device that appears in a Breakthrough beyond the trial and the parent count.

2.5 fable-v10-zero-missed-8ams ($7,557, 241, $31.4; July) and its image twin img-c1-dressed-asleep ($8,267, 161, $51.3; Aug)

• Awareness: problem-aware, first person, retrospective ("did not miss a single 8am class").
• Valence: fear answered. High intensity.
• Desire: future readiness, his independence proven.
• Avatar: mother of an ADHD teen headed to college.
• Angle: college proof, told backwards.
• Mechanism: pediatrician says "her brain treats sound like spam." One authority beat.
• Belief: retrospective result, authority, count.
• Positioning: none.
• Urgency: the college date.
• Hypothesis: the only long native that reached Breakthrough, and it did it at 3.75% LPV-to-purchase versus 5.6% for the skeleton ads. The same body with a "girl dressed asleep" photo went to $51 CPA at 2.8% LPV-to-purchase: the image bought curiosity clicks that did not buy. The lesson is not "natives work," it is "this one story worked once at the top of a college-fear window, and its image is fragile."

2.6 offer ad Copy 3 / Copy 2 (5 headlines, DAWN20; $5,249, 170, $30.88; May, still $29 this week)

• Awareness: product-aware / offer-aware. "If your teen sleeps through alarms, misses the bus, or needs you to drag them out of bed, louder noise probably is not the answer."
• Desire: undisturbed house ("Wake them. Not the whole house.").
• Belief: 4 checks and the guarantee. 453 characters.
• Hypothesis: the shortest body in the account converts at 7.07% LPV-to-purchase, the best of any family with real spend. The account has one offer/closer family and it is the only thing not fatiguing. Competitors run a whole closer layer (35% of their creative). We run one ad.

2.7 KPI winners never scaled (the bench)

• Thiswakesthem2 ($3,776 across 3, $23.03): 356-character skeleton, "it's not attitude." Cheapest CPA of any family with $2k+.
• Mothers Day 1 ($1,746, $27.72, 8.76% LPV-to-purchase): a seasonal hook on the skeleton.
• tp723-07-freezer-marbles ($1,711, $28.52): in-group artifact open, 1,244 chars, PDP-direct. Re-promoted 8-28 and printing at $22.5 last 7 days on $180.
• s3-never-late ($8,558 across 3, $30.56, 8.12% LPV-to-purchase): a SELF-BUYER ad (ADHD adult, "your manager noticed") with a melatonin new-information beat. The parked adult lane has a $30 CPA receipt at $8.5k.
• graduated-6-academic ($1,301, $34.2) and 6-academic ($360, $17.1): college self-buyer, long native.
• independent slop 13 ($928, $29.9): the skeleton with the independence/college stake in the body.
• image-notdisciplineproblem ($967, $30.2, 9.55% LPV-to-purchase): a 328-character static to the "real reason" listicle. Highest post-click conversion in the account.

PART 3 · WHAT THE LOSERS ARE (nine-question decomposition of the ones that cost the most since July)

3.1 img-c1-dressed-asleep ($8,267, $51.3): see 2.5. Same body as a Breakthrough. Image swap doubled CPA. Test attribution: image only.

3.2 s2-a-img2-dusk-uniform ($1,926, $47.0, 2.6% LPV-to-purchase): long native, scholarship letter stake, sleep specialist in a webinar as authority, attendance review letter as artifact. Awareness first-person. Desire future-readiness. It read like v10 with the stakes turned up and converted at half the rate. Hypothesis: stakes at the top of the ad recruit readers who are worried, not readers who are buying.

3.3 tp723-01-coffee-shop-exile ($3,286, $41.6, 4.3%): 12,462 characters. In-group open, failed-solutions with prices, long. Got spend, did not clear CPA. Hypothesis: length past ~1,500 characters costs conversion on Meta feed for this buyer; the story is good and the click is late.

3.4 evolve-a4-s1 ($684, $52.6; Sep 4): the skeleton, rebuilt for the sibling-tax sub-avatar. Changes vs slop 7: qualifier narrows to two-kid households; mechanism swapped to the cortisol / fight-or-flight chain (the competitors' spine); "2009 review of 16 studies" authority line added; desire moved to calm mornings. Every one of those four changes points the wrong way in the cross-cuts below. This is the cleanest receipt in the account that the 9-03/9-04 batches changed the things that were working and kept the things that were not.

3.5 ugc904-jess-control ($751, $47.0; Sep 4): UGC video, third-person caption, feature list (15-day charge, no app). Video is 4 for 4 losers in the labeled set at $54.89. Not a creator problem; the format has never converted in feed here.

3.6 charswap-B-m6-driveway ($459, $57.4; Aug 20): short caption on video, "gently at first, then harder." Same.

3.7 s1-adhd-teen ($366, $61.1; May): emoji-static body, "7 alarms," ADHD mechanism with "nervous system literally closes the channel," college in 14 months. Hypothesis: mechanism-heavy and stake-heavy in 1,041 characters with no recognition line.

3.8 Listen Kid v3 ($671 + $478, $60 to $84): grandparent narrator. Falsified twice. Narrator off the mother costs conversion (7-31 L9 said the same on CTR).

3.9 fable-v07-h ($301, $60.2): dad narrator recording his wife. Same lesson as 3.8.

3.10 100% Wake Up Guarantee v1 ($327, $40.8): the skeleton with a guarantee headline. The guarantee as the lead does not convert better than the recognition lead (feedback_guarantee_positioning 8-27 already ruled this).

3.11 The 9-03 / 9-04 / 9-05 / 9-07 waves: evolve-sa3 ($95, $94.6), evolve-a5-s5 ($187, 0 purchases), dawn56 self-buyer statics ($153 to $50, $25 to $115), S02/S03/S04 apologetic mom ($54 to $112, $56 to none), Partner 0908 ($55, 0). 12 ads, $2,375, 40 purchases, $59.38. Ten of twelve starved under $200. Two read as losers. Zero KPI winners.

PART 4 · CROSS-CUT FINDINGS (labeled ads, $300+, unless stated)

F1. The format that wins is short copy plus one image, not the long native.

• Short copy + image: 66 ads, $192,533, $31.20 CPA, 5.63% LPV-to-purchase. 14 of 15 Breakthroughs.
• Long native (2,000+ chars): 22 ads, $53,954, $38.16, 3.84%. 1 Breakthrough, 5 Spend Winners, 10 Losers.
• Video: 4 ads, $2,251, $54.89. 0 winners.
• Natives get more link clicks per impression (2.23% vs 1.32%) and convert them worse. The 7-31 learnings' "natives buy a visit for $0.12" used all-clicks CTR (14 to 44%); on link clicks native CPC is $1.08 to $1.60, the same as the skeleton ads. The click-price advantage was a measurement artifact. What is real: natives read well and buy less.
• Since August the launch mix flipped to the losing format: 27 of 38 August ads were long natives or video. May was 61 of 69 short copy + image.

F2. Recognition converts. Story converts less. Stakes convert least.

• Named-reader problem-aware hooks ("If you're still the one…" / "If your ADHD teen…"): 51 ads, $176,821, $31.84, 5.29%.
• First-person witness hooks: 24 ads, $58,653, $36.16, 4.52%.
• Product/offer-aware hooks: 12 ads, $11,018, $28.04, 7.16%. Cheapest lane, least funded.
• Stakes in the body: none $30.56 (29 ads) and relationship $33.22 (37 ads) beat college/scholarship/attendance $39.80 (4 ads, CTR 2.48% but 3.25% LPV-to-purchase) and job $46.19. High stakes buy clicks from readers who are scared, not readers who are ready. The 8-23 hook taxonomy mandates "High Stakes" in every hook. The account's money says the opposite: symptom recognition in line 1, stakes as one line in the body at most.

F3. The mechanism is table stakes, and elaborating it does not pay. (The sophistication answer.)

• Every labeled ad names the wrist/vibration; all but one (a UGC caption) name sound. There is no ad without the mechanism to compare against, which is itself the finding: the mechanism cannot be the variable that separates winners from losers because it is constant.
• Simple mechanism (sound + touch, or + filter), 2 to 3 sentences: 73 ads, $228,706, $32.6 blended.
• Persist-until-dismissed added: no lift ($32.00 with, $33.39 without).
• Initiation Wall added: 7 ads, $6,791, $30 to $44.6. No lift.
• Startle / cortisol chain (the 9-03 rebuild): 1 ad with a read (evolve-a4-s1), $684, $52.6; a5-s5 $187 with 0 purchases. Worse, small n.
• Study/authority beats (clinician, "2009 review," webinar specialist): 9 ads, $24,352, $39.5 to $52.6, LPV-to-purchase 2.5% to 3.5%. High CTR, low conversion. The Alevia deltas ruled IN on 9-03 (clinician authority, counted results) have no receipt in this account and a negative one where they were tried.
• The sophistication answer. Lock-in §1d called the mechanism claim "Stage 4-5, saturated: four-plus competitors run it, it must be in the ad, it cannot be the hook." The data agrees exactly. The market is past the mechanism. The winners treat it as a three-sentence formality and spend the ad on recognition. The August/September work spent its cycles on the mechanism (uniqueness plan, cortisol chain, Wall, persist) and on authority, which is a Stage-3 play in a Stage-4/5 market. Answer to the question asked: the mechanism is not "not fitting" the market; it is invisible to the market. It cannot differentiate, so work on it cannot move CPA. What is unclaimed (lock-in §1d) is the mother as subject and the sheer recognition line; that is the Stage-5 identity play, and it is what the winners already do in line 1.

F4. Desire: independence and vindication carry; calm mornings and undisturbed house are a tier lower.

• independence 21 ads $76,844 $30.77 · vindication 19 ads $73,871 $31.66 · reliable-wake 17 ads $23,267 $31.66 · future-readiness 11 ads $30,966 $35.55 · calm-mornings 17 ads $18,336 $35.53 · undisturbed-house 4 ads $22,467 $36.65 · safety 1 ad $46.97 · relief-of-responsibility 2 ads $53.09.
• The 9-03 sub-avatar build (813 lines, SA1/SA2/SA3) and every batch cut from it were built under "I want calmer mornings." That desire converts at $35.5 on $18k. Independence and vindication convert at $31 on $150k. The sub-avatar work was pointed at the third-best desire on the board.
• The lock-in ruled INDEPENDENCE overarching on 9-02. The batches shipped after it did not follow the ruling.

F5. Post-click, not the click, is where losers die.

• Winners: 5.6% to 8.8% LPV-to-purchase. Losers: 2.5% to 3.6%. CTRs overlap (1.0% to 2.9% both). A 2%+ link CTR with a 2.6% LPV-to-purchase (s2-a, img-c1) is a curiosity click. The 8-31 hook learning said the same ("unqualified clicks are the failure mode"); this is the account-wide number for it.

F6. Destination matters and the wrong listicle is carrying volume.

• school-mornings-adhd listicle: 5 ads, $45,900, $31.12, 6.35% LPV-to-purchase. Best converting destination at scale.
• PDP wake-up-band-for-teens: 27 ads, $88,580, $31.93, 5.20%.
• 7-reasons listicle: 8 ads, $24,111, $30.29, 4.78%.
• PDP wake-up-band-adhd: 22 ads, $46,223, $35.07, 4.95%.
• adhd-teen-alarm listicle: 14 ads, $30,469, $36.67, 4.01%. Worst destination with real volume. This is the page the 8-13 rebuild and most 8-23 ppad ads point at.
• the-real-reason listicle: 3 ads, $3,663, $22.75, 7.98%. Tiny n, best number on the board.

F7. Narrator off the mother loses. Grandparent (2 ads, $60 to $84), dad (v07, $60), teen-first-person natives (s1-adhd-teen $61, 6-academic is the exception at $17 to $34 on the self-buyer PDP). 7-31 L9 found the same on CTR; it now holds on CPA.

F8. Proof: the trial and a parent count are enough. Retrospective result helps once (testimonial_v2, v10). Study/authority hurts (F3). Guarantee as the lead does not beat recognition as the lead (100% Wake Up Guarantee v1 $40.8 vs the same body with the dependency lead at $31).

F9. Length: the two winners over 2,000 characters (v10, 3-partner-alarm) are the only two, and 3-partner is a Spend Winner at $36.5. Every winner under 1,300 characters converts at 5%+. Every native over 2,500 characters converts under 4.3%. tp723-01 at 12,462 characters is the longest ad in the account and the worst-converting Spend Winner.

F10. Starvation is structural and got worse. 190 starved + 2,073 never-read. Since 8-01: 44 of 50 ads with $50+ never reached $300. Batches of 10 to 56 ads into CBO give 80% of budget to one ad by day 3 (the 8-10 synthesis said 281 of 1,399 never delivered; it is now 2,263 of 2,355 under $300). The 9-05 self-buyer batch launched 56 statics; the top one got $153. Nothing launched since August has had a fair read except img-c1, s2-a, ugc904 and a4-s1, which lost.

F11. The bench has never been scaled. 33 KPI winners at $28.61 blended on $25,546. Thiswakesthem2 ($23), Mothers Day ($27.7), freezer marbles ($28.5), s3-never-late self-buyer ($30.6), notdisciplineproblem ($30.2). The account looked for new winners while sitting on efficient ones that never got past $2k.

F12. Fatigue is now. slop 9 / 7 / 4 at $36 to $44 in the last 7 days versus $29 to $33 lifetime. They are 55% of spend. There is no successor in the pipeline that is (a) in the winning format, (b) under the winning desire, (c) with a fair budget. August and September built successors in the losing format under the third desire with starved budgets.

PART 5 · RECOMMENDATIONS (ranked by expected CPA impact per dollar, each tied to a finding)

R1. Build the successor to slop 7 on the skeleton, not on a new architecture. (F1, F2, F4, F12)

Hold: the 7-beat skeleton, 900 to 1,100 characters, one still image, PDP teens or school-mornings-adhd destination, desire = vindication or independence, no stakes above one line, mechanism at three sentences, proof = trial + "10,000+ parents."

Variable: the line-1 qualifier only. Write 8 line-1 recognitions by observable symptom from the survey (sleeps through every alarm 31%, ADHD mornings 25%, independence / new schedule 17%, daily fights 14%, missing the bus 5%) and the 9-03 harvest's verbatim behaviors, each one a sentence she would say. Examples of the shape, not the copy: "If you set his alarm on your own phone because his doesn't work" / "If you've stopped sitting down for coffee because you're at his door by 6:40."

Then: image fan-out on the two best line-1s, 5 images each, nothing else moves.

Budget: one ABO ad set per 4 ads, $105/day floor (3 × $35), 7 days, no verdict before day 7.

R2. Fund the offer/closer layer. (F2, 2.6)

Offer ad Copy 3 is the only ad holding CPA this week and product-aware hooks convert at 7.16%. Write 4 more closers in the 450-character shape (two-line qualifier, four checks, guarantee, code), each with a different lead: "Tried every alarm already?" / "Wake them, not the whole house" / "No app, no noise" / the guarantee. Put them in their own ad set at the floor. Competitors run 35% of their creative as closers; we run one ad.

R3. Scale the bench before writing anything new. (F11)

Thiswakesthem2, Mothers Day 1 (re-skin the seasonal line for back-to-school / first report card), freezer marbles, s3-never-late (self-buyer), notdisciplineproblem → the-real-reason listicle. Five ads, one ad set, $150/day, 14 days. These are $23 to $31 CPA receipts sitting at under $2k lifetime.

R4. Rebuild the sub-avatar work under the ruled desire. (F4)

The 9-03 SA1/SA2/SA3 cards are built under calm mornings. Redo the three cards under independence (his and hers) and vindication. Same Core 5 method, same verbatim discipline. The sibling-tax and first-hour angles can stay as body beats; they are not qualifiers.

R5. Freeze the mechanism. Stop the mechanism work. (F3)

Three sentences, sound / tunes out / can't tune out the wrist. No cortisol, no fight-or-flight, no Initiation Wall as a hook, no "persists until," no study citations, no clinician. Retire the uniqueness-vibration plan as a copy input (keep it for the PDP and the wear/FitSleeps objection). Strike the 9-03 Alevia deltas (clinician authority, counted results) from the default brief; allow them only as a named test arm. This is the direct answer to "is our mechanism not fitting the sophistication": the market is Stage 4-5 on the mechanism, the winners already treat it that way, and every attempt to make it the differentiator has scored worse.

R6. Drop the High-Stakes hook mandate. (F2, F5)

Stakes are one sentence in the body (rung 2 or 3 at most: the bus, the tardies), never the hook, never the headline. The scholarship/attendance-letter family converts at 3.25% LPV-to-purchase on a 2.48% CTR: the clicks are curiosity. Update hook-taxonomy.md: Emotion + Curiosity Gap stay; High Stakes moves to optional body beat.

R7. Fix the destination routing. (F6)

New ADHD ads go to school-mornings-adhd or PDP teens. Stop routing to adhd-teen-alarm ($36.67, 4.01%) until it is audited against school-mornings-adhd side by side (same ad, two destinations, one week). Give the-real-reason listicle one real test (it is $22.75 on $3.6k).

R8. Length cap and format default. (F1, F9)

New concepts ship as short copy + image by default, 900 to 1,300 characters. Long natives are a named test arm, one per batch, never the batch. Video and UGC do not ship to feed until one converts anywhere (4 for 4 losers; the 7-31 Reels/Stories-only note is the only open door and it was never funded).

R9. Batch hygiene. (F10)

Maximum 4 ads per new ad set. ABO. $105/day floor. 7-day gate. No new batch while more than 2 test ad sets are open. 56-ad launches end. The approval gate gets a 48-hour SLA or the batch shrinks until it can meet one.

R10. Narrator stays the mother. (F7)

No grandparent, dad, teen-first-person or partner narrator in a parent-lane ad. The self-buyer lane is a separate lane with its own PDP and has two $17 to $31 receipts (s3-never-late, 6-academic); reopen it as its own test, not as a narrator swap.

R11. Write the two hypotheses down and test them as such. (Part 2)

Slop 7 hypothesis: symptom recognition in line 1 + blame removal + three-sentence mechanism + short = the buyer. v10 hypothesis: a retrospective college-proof story worked once, at 3.75% LPV-to-purchase, and is image-fragile. They agree on line 1 and disagree on everything after it. R1 tests the first; R8 quarantines the second.

R12. Re-run this ledger every 14 days from the same script. (Part 8 of the directory)

output/dawn-learnings-ledger-2026-09-10/pull.py then classify.py. Append the new labels to this file. The angle ledger and hook-performance files are retired; this is the record.

PART 6 · NEXT TESTS (in order)

T1. Skeleton line-1 sniper. Hold everything in R1. Variable: 8 line-1 qualifiers. 2 ad sets × 4 ads, $105/day each, 7 days. Gate: any ad under $35 CPA on 8+ purchases graduates to an image fan-out.

T2. Closer set. R2. 4 ads, one ad set, $105/day, 7 days. Gate: same.

T3. Bench scale. R3. 5 ads, one ad set, $150/day, 14 days.

T4. Destination split. R7. slop 9 body, PDP teens vs school-mornings-adhd vs adhd-teen-alarm, one ad set each, $70/day, 7 days.

T5. Sub-avatar rebuild under independence. R4. Three line-1s (one per card), skeleton body, 2 images each. 6 ads, 2 ad sets.

T6. Self-buyer lane, properly. R10. s3-never-late + 2 new skeleton bodies for the ADHD adult, self PDP, one ad set, $105/day, 7 days.

Part 7 · What This Changes In The Directory Scores

• 8.1 Learnings on losers: 1 → 2 (this file). 8.2 winners: 1 → 2. 8.4 labels: 0 → 2 (every $300+ ad is labeled in the table below). 8.5 tracking surface: 1 → 2 (ledger_rows.csv is the row-per-ad record; the sheet and hub still need the same columns).
• 1.3 sophistication: the mechanism question is answered with data (F3).
• 5.5 image ads: the account's format is short copy + image; the score stands at 2 but the direction of August's work was wrong.
• 6.1 test campaign: F10 makes it a 1.

PART 8 · THE LABEL TABLE (every ad with $300+ lifetime spend, 92 ads)

Columns: label · lifetime spend / purchases / CPA · link CTR · LPV→purchase · format · desire · created · status · ad name · ad set

PART 9 · STARVED SINCE 8-01 (unread, $50 to $300; not falsified)

Columns: spend / purchases / CPA · link CTR · LPV→purchase · format · created · ad name · ad set

Receipts

output/dawn-learnings-ledger-2026-09-10/{pull.py, classify.py, meta_raw.json, ledger_rows.csv, cuts.txt, bodies.txt} · adhd-mom-lockin.md §1d, §3, §9 · research/winners-evolve-bins-2026-09-03.md · research/evolve-subavatars-calmer-mornings-2026-09-03.md · output/meta-account-audit-2026-07-31/LEARNINGS-week-7-19.md · ad-batch-hub site/launches/data/learnings.json (8-10 synthesis) · hook-taxonomy.md · uniqueness-vibration-plan-2026-09-04.md · evolve-brand-directory-2026-09-10.md · scratchpad audit notes: subagent-audit-research.md, subagent-audit-organize-plan.md, subagent-audit-execution-ops.md · Evolve Avatar Training export: output/adhd-self-buyer-evolve-report-2026-09-05/receipts/evolve-download