The Monday Board

Two booked re-verdict dates have now passed uncollected, the oldest by 47 days. Section 1's whole argument is that $22,408 of measured loss came from verdicts nobody returned to collect. These are written here rather than rolled forward a third time.

Overdue by 47 days — 2026-08-05, and by 42 days — 2026-08-10. The 2026-07-29 board booked a re-verdict on thirteen RUN ads and on the Bird Lovers campaign 120227961038970726, and both dated overrides expired the same day — the catalog ad sets at $50.00 each, and the $26 cost-per-result goal on AdSet_01_Open. Section 9 says an override past its expiry is the rulebook quietly changing. Source: jobs/ads/decisions/2026-07-29-board-decisions.md.

Bird Lovers closes as NOT TESTED. Campaign 120227961038970726 reads PAUSED live. Reactivated 2026-07-29, it spent $111.71 over 2 days across four ad sets and returned $39.49 — 0.35 back per dollar on 1 purchase — then stopped on 2026-07-31. No economic verdict is permitted: it never came near the $250 gate. It was switched off before it bought the evidence its own dated chance was meant to buy, and NOT TESTED is the honest label — which also means monthly catalog maintenance may not act on any of it.

The Audition-20 override can no longer expire, which section 9 forbids outright. jobs/ads/decisions/2026-07-30-audition20-overrides.md sets its review date as "at the $250 verdict of slot 1 (Jimothy-03)". Slot 1 died under K1 at $68.92, so that verdict will never exist and the date will never arrive. Three deviations from the rulebook — $20.00 a day instead of $8.00, a day-three stall check instead of day-four, and kills that can be resurrected at day seven — are live with no reachable expiry. One dated line closing the pilot fixes it, and the pilot section below has the numbers that line needs.

5 of the six ad sets crossed $54 with zero purchases, which is K1 — and the kill has matured. Section 3.3a's pending-confirmation window is long closed: the test has not spent since 2026-09-01, the seven-day attribution window has run out, and no purchase arrived. Each is an actionable KILL with the record RULED OUT. There is nothing to click — Meta shows every one switched off since 2026-08-31. What is owed is the written record.

Static_NotTodayCardinal_PDP_2026-CW35 (ad set 120249108191930726) is the one design that bought a sale: $102.04 for 1 purchase, a 0.33 return. It tripped no kill rule and never reached its $250 gate, and it has read switched off since 2026-09-01. Section 1 says a pause is a KILL or a stall; it is off, so the call is KILL — but the honest record is evidence incomplete, the same reading as the four Audition-20 slots below, so monthly catalog maintenance must not act on this design. If you would rather it finish its test, the alternative is switching it back on; that is your call, not the board's.

The stall test does not apply. It is a test of a switched-on ad set that is not spending its budget, and nothing here is switched on. For information only: while live, each ad set delivered between 56% and 82% of what its own budget authorised across the 5 days with spend.

The comparison the test was built to make, reported and not judged. Garment context took $294.72 across 6 ads and returned $33.86 on 277 link clicks; full-frame fabric took $180.56 across 6 ads and returned $0.00 on 132 link clicks. The single purchase in the whole test came from a garment-context ad. 5 days and one purchase is not a framing answer — section 2 forbids an economic reading at this spend, and the honest statement is that the test was stopped before it could give one. Cost per add-to-cart is the en-route signal it can legitimately show: $49.12 for garment context against $25.79 for full-frame fabric, on 6 and 7 add-to-carts respectively — too few to lean on, and pointing the opposite way to spend.

2 of the 6 are finished verdicts that can be written in one line each. Slots 73 and 78 both passed $54 with zero purchases, which is K1, an automatic KILL. Slot 73's day-seven attribution resurrection check — granted by the pilot's own override — came due around 2026-08-09 and nothing arrived. Both are RULED OUT.

The other 4 are the expensive part, and the rulebook has no word for them. They stopped between $107.76 and $165.88 — past the money that buys evidence, short of the money that buys a verdict. Section 1 says a pause is either a KILL or a stall and asks you to say which; they are off for good, so the decision is KILL. But neither record label fits: they all passed $56 so they are not NOT TESTED, and none reached a gate or tripped a kill rule so they are not RULED OUT either. The honest record is evidence incomplete, and its practical consequence is that monthly catalog maintenance must not act on any of these four designs. They were not proven to fail; they were stopped.

What the evidence says about how to fill them. Section 5.3 gives four kinds of slot, and this month has produced a clean read on two of them. Established re-tests of proven winners: three ran, spent $297.67, returned 0.64 against source returns of 2.19 to 2.38 — that lever has been pulled and it did not work. Angle variations: the three Jimothy auditions spent $393.69 and returned $169.33, a 0.43, and none reached its gate. The two untried slots are graduates with an organic sales signal — section 5.3 counts 15 published products that sold with no ad help and have never had a dedicated test — and blind discovery from the newest batch. Forty-eight designs went live on 2026-08-16, so the blind-discovery pool has never been fuller.

One design has earned a slot on its own evidence and it is not a re-test. 11652301_Perfectly-Imperfect_Flatlay_01 (ad 120248308232330726) finished at 2.40 back per dollar on $146.96 across 5 purchases — the best number in the account — and it stopped $3.04 short of its $150 gate when its container 120248308217960726 was switched off on 2026-08-26. It never got a verdict and under section 1 it cannot have one: it is still TESTING, so the honest record is NOT TESTED and monthly catalog maintenance may not act on it. The cheap, correct move is to re-slot it in its own one-ad ad set and let it buy the last three dollars of evidence it was three dollars away from buying.

A design's round is the calendar week its first dollar was spent, and this scorecard counts dedicated tests only — catalog never takes a $250 verdict, so including it would flatter the number. The round that entered the week of 2026-08-31 put 0 of 0 across the gate.

The number is thin because almost nothing entered, and that is the finding. One design entering a round is not a selection rule being tested — it is a selection rule not being run. Look at the three rounds either side: 0 entered the week of 2026-08-17 and 0 crossed; 6 entered the week of 2026-08-24 and 0 crossed; 0 entered the week of 2026-09-07 and 0 crossed. Across the last five rounds 0 of 6 dedicated designs have reached the $250 verdict gate. The rulebook is explicit about what a flat month means: the problem is upstream, in what is being chosen to test — and step 7 is where that gets fixed, not here.

The six designs of the static framing test are counted. They sit in their own campaign 120249101567700726 rather than in 120232440887150726, and each ad set is one design, so they are scored in the round of the week they first spent, 2026-08-24. None reached $250; the most any one spent was $102.04.

For completeness, because this morning's daily digest quotes a different number. It puts the round of Aug 31 at 2 of 2. The digest counts every non-catalog ad in the whole account that first spent that week, including ads outside any dedicated test; this board counts only designs that entered a dedicated test — the testing campaign and the static framing test. The digest's ads that crossed $250 were not dedicated tests, so they do not enter this scorecard; the digest itself records what became of them.

Carousel is buying an add-to-cart 11% cheaper than single image — $14.53 against $16.27 — on near-identical spend ($1,365.80 against $2,164.41). Section 10.1 puts a 28%-or-better same-destination gap in the Decision band. And 56 purchases across 14 arms cannot separate anything: section 10.1's bands were set on the arithmetic that a four-week block buys roughly 27 finished orders per arm, and this block has between 0 and 3. The gap is real in the data and not yet distinguishable from noise.

2 arms sit at zero purchases on more than $120 each — C2 (BundleSave, single image), T1 (Collection, single image). Under a dedicated test each would be dead twice over, killed at $54 and again at $81. Section 5 gives catalog no ad-level kill rules at all, so nothing fires and that is correct rather than an oversight: catalog is infrastructure, and its job is coverage.

The one destination reading worth carrying to the block's end. BundleSave is last by a distance at 0.27 on $372.84 across both formats, and the Home page — the one destination section 6 rules against with High confidence — sits at 0.98, ahead of three of the other four. That is now several mornings running. One block does not overturn a table built on the account's whole history, and section 8.6 warns that absolute figures are the least trustworthy numbers here — but the block's end date is the right moment to say it out loud.

Catalog carries no kill rules, so nothing in this section is a decision. It is judged monthly on its own cost-per-add-to-cart trend, never against 1.87, and it is reported on the Monday board only so the weekly ritual is not mistaken for the whole account.

The destination incumbent has stopped delivering, and section 6.1 still points at it. Ad set 120248466373140726 reads PAUSED. It holds the /Sale catalog advertisement (ad 120248466416500726) at $504.82 of era spend and a 1.37 return — the best catalog arm outside the protected winner. Section 6.1 settles the account's one live destination conflict by ruling that /Sale holds as the catalog incumbent until a head-to-head settles it at equal budget in the same weeks, and the ten-arm block has no /Sale arm. So the head-to-head section 6.1 promised no longer has an incumbent running in it. NOT AVAILABLE — reason: no receipt records who paused it or when, and the ten-arm authorisation (receipts/ads/2026-08-12-catalog-build-sheet-ten-arm-creation-authorization.json) explicitly did not authorise incumbent changes. One line saying whether it stood down deliberately closes this.

The catalog layer is configured at $368.00 a day against a section 5 layer budget of $60.00 and Tim's own section 14 ruling of $120.00. $368.00 of it is the ten-arm block, and that part is authorised in writing — the 2026-08-12 creation receipt locks a maximum configured daily total of $368.00 for a fourteen-day measurement window. What has no authority behind it is the arithmetic: section 14 says $120.00 a day and the block says $368.00 a day, and nothing on file records which supersedes which. Section 9 would call this an override — it names what changes and who agreed, but not the date it lapses. Section 14 also said the legacy catalog ads retire when their replacement arms launch; the arms launched on 2026-08-13 and $0.00 a day of legacy is still switched on.

Who switched off the dark ads, and on whose decision. 6 of this board's 6 kills in the old testing campaign were already switched off before it ran, and so were 6 of the six static framing test ad sets — Meta's own last-updated stamps put those pauses on 2026-08-31, 2026-09-01. What no receipt under receipts/ads/ records is who made any of those pauses or against which rule. They are reported here as read from the account. NOT AVAILABLE — reason: no receipt written.

Whether recent revenue is final. Attribution matures for up to seven days, so every return on this page is a floor rather than a settled figure. That matters most for the five static framing test ad sets killed on three days of evidence at zero purchases, and for the catalog block's 16-day numbers. A purchase that attributes late would be recorded against the decision rather than hidden — and for the static test it would turn a RULED OUT record into a NOT TESTED one.

Shopify-verified revenue. Everything here is Meta-attributed (section 8.6). Blended Meta-reported return runs 1.18 against a store-wide 1.47, so comparisons between two arms running in the same weeks are far more trustworthy than any absolute figure on this page — which is exactly why the ten-arm block is built the way it is.

Which products the catalog impressions landed on. Section 8.5 records that all sixteen catalog ad identifiers have zero rows in both product-bridge tables. It is the only build dependency in the whole rulebook and it still blocks the two catalog-signal graduation slots in section 5.3. NOT AVAILABLE — reason: no product-level catalog breakdown exists.

The cost-per-result goal on AdSet_01_Open. It sits under a campaign-budget campaign and carries no ad-set budget row, so it cannot be read through the interface this board uses. Its $26 override lapsed 47 days ago and its current value is NOT AVAILABLE — reason: no ad-set budget row.

Whether the account is spending to plan. Section 5 sets $250.00 a day across both layers. The last three complete days averaged $143.97. That is the whole account, not this campaign, and no rule on this board acts on it — it is here because a board that reports only the campaign it prunes would let the number drift unwatched.