Kova · Pott Candles · Meta · read-only backtest

Tomato Vine: what moved before the winners wore out, and a watcher that refreshes instead of cutting

Built 21 Oct 2026 from Meta Graph API data pulled the same day (1 Jan–20 Oct 2026, unified attribution). Nothing in the ad account was changed. Every figure below is from the API or computed from it; the method notes say where a number is a judgement.

The answer

The two tomato winners wore out in different ways, so no single signal leads. Ad 551 (video) lost click rate first: CTR net of the account fell from the week of 8 Jul, 5 weeks before its return dropped (12 Aug). Hold rate followed about 2 weeks before; hook rate never moved. Ad 707 (image) never lost click rate. What moved was reach: new people per £ fell 40% in week 2 and about 50% after, 2 weeks before its return dropped (2 Sep). Ad 487 lost click rate about 2 weeks ahead on weekly data (the daily watcher saw it 4 weeks ahead). Frequency was useless: the tomato ad sets never passed 2.45 against a 2.5 ceiling.
The current watcher would not have got a refresh live in time on the biggest one. On Ad 707 its first call was "Cut today" on 29 Aug, 4 days before the drop, after 12 days of "too new" while the ad took 31% of spend. With a 2–3 week build, a refresh briefed then goes live mid-September; the real tomato refresh wave went live 24–25 Sep, 23 days after the drop began. On Ads 551 and 487 it did fire early enough (Highly likely 26 and 29 days ahead), but its stage changed 16–19 times per ad, the account-wide guard hid calls on 4–9 days per ad, it said "Cut today" on Ad 551 during the 15–21 Jul site dip, and it lost Ad 551 entirely when the ad was duplicated on 5 Aug.
New ladder: Monitor → Brief refresh (any winner: 10%+ of spend at or above the account's ROAS, even when nothing has slipped) → Build refresh (the old Highly likely, unchanged) → Launch refresh (the old Fatigued and Close: launch next to the original, never pause it) → Hand over (a refresh is live and proving itself). It adds one check (new reach per £), keeps each stage for 7 days from when it was last earned, and a fresh Launch on an account-wide day drops to Build or Monitor. On the tomato ads it briefs Ad 707 on 19 Aug (14 days ahead) and Ad 551 on 22 Jun (51 days ahead).
It does not fire on everything. Over Apr–Oct it opens 61 refresh items, against 57 old alerts, and 31 Launch entries against 29 Close/Fatigued. Ads it puts at Launch went on to earn 0.70x of baseline over the next 14 days, net of the account; ads at None earned 1.06x.
Pott as of 20 Oct: Launch refresh on Ad 1000 (Holly review card, 26% of spend, 1.95x → 1.30x; the old rules said "Cut today") and Ad 983 (Icicle). Build refresh on Ads 922, 995, 973, 940, 837 and 915. No Hand over: Ad 1035 looked like Ad 1000's refresh but uses the same image (sent to the homepage), so it is a duplicate. Caveat: both Launch ads' baselines include the 6 Oct Christmas launch week.

The pick: top tomato ads by spend, 1 Apr–20 Oct 2026

Matched on "Tomato Vine" in the ad name. Among the account's top 80 spenders, every tomato ad whose link we could read points at /collections/tomato-vine-refillable-candle, and no ad without "Tomato" in its name points there. No tomato ad spent anything before 2 Jun 2026, when the tomato campaign launched. Highlighted rows are analysed. Ad 551 is one video in two ad IDs (£9,498 together, so effectively joint top with Ad 707). Ad 487 is included as a close third.

#	Ad	Ad set	Spend	Purchases	ROAS	Created	Status today
1	Ad 707 Image · Tomato Vine · UGC Testimonial OG - Copy	Ad Set 1 - MAIN - Broad - Exclude Customers	£9,226	230	1.73x	14 Aug	PAUSED
2	Ad 551 Video · Tomato Vine · Unboxing OG - Copy	Ad Set 1 - MAIN - Broad - Exclude Customers	£4,934	109	1.50x	5 Aug	ADSET_PAUSED
3	Ad 551 Video · Tomato Vine · Unboxing OG	Ad Set 2 - MAIN - Broad - Exclude Customers	£4,564	169	2.55x	17 Jun	ADSET_PAUSED
4	Ad 487 Image · Tomato Vine · Refillable Starter Pack OG	Ad Set 1 - MAIN - Broad - Exclude Customers	£4,041	152	2.52x	2 Jun	ARCHIVED
5	Ad 489 Image · Tomato Vine · Three Wick Tower OG	Ad Set 1 - MAIN - Broad - Exclude Customers	£3,598	119	2.47x	2 Jun	ADSET_PAUSED
6	Ad 940 Image · Tomato Vine · Maker So Happy OG	Tomato - Broad - Exclude Customers	£3,565	68	1.26x	25 Sept	ACTIVE
7	Ad 934 Image · Tomato Vine · Customer Quote Whole Hall OG	Tomato - Broad - Exclude Customers	£3,550	63	1.24x	24 Sept	ACTIVE
8	Ad 828 Video · Tomato Vine · Scent Led OG	Ad Set 1 - MAIN - Broad - Exclude Customers	£3,457	91	1.85x	15 Aug	PAUSED

Purchases and ROAS: Meta pixel purchases, unified attribution (the watcher's definition). Pott's breakeven ROAS is 2.46x (margin sheet); the account ran 1.5x–2.7x a week over this period, so every ad here sat below it most weeks.

What else was going on (confounders)

All three ads ran in the CBO campaign "CS | Sales | Cold & Warm | Evergreen (old Tomato vine)". Meta moves money between ads inside it, so a spend fall is often Meta's choice, not a sign the ad is spent.

Campaign daily budget (from the activity log)

Account ROAS by week (Wed–Tue)

Date	Budget change	By
6 Jun	£100 → £120/day	Casey Snow
11 Jun	£120 → £144/day	Casey Snow
14 Jun	£144 → £172/day	Casey Snow
15 Jun	£172 → £206/day	Casey Snow
3 Jul	£206 → £246/day	Casey Snow
14 Jul	£246 → £290/day	Casey Snow
25 Jul	£290 → £340/day	Casey Snow
4 Aug	£340 → £408/day	Casey Snow
5 Aug	£408 → £488/day	Casey Snow
20 Aug	£488 → £800/day	Casey Snow
30 Aug	£800 → £960/day	Casey Snow
4 Sept	£960 → £1,100/day	Casey Snow
19 Sept	£1,100 → £1,320/day	Casey Snow
25 Sept	£1,320 → £1,820/day	Casey Ai Brain
7 Oct	£1,820 → £2,550/day	Casey Ai Brain
14 Oct	£2,550 → £3,060/day	Casey Snow

Date	Event
2 Jun	Ad 487 created (OG)
17 Jun	Ad 551 created (OG)
5 Aug	Ad 551 created (OG - Copy)
5 Aug	Ad Set 2 - MAIN - Broad - Exclude Customers set inactive
14 Aug	Ad 707 created (OG - Copy)
13 Sept	Ad 487 switched off (Deleted)
19 Sept	Ad 707 switched off (Inactive)
25 Sept	Ad Set 1 - MAIN - Broad - Exclude Customers set inactive
25 Sept	Ad Set 2 - MAIN - Broad - Exclude Customers set inactive
15–21 Jul	Account-wide site dip (account ROAS 1.32x vs ~1.9x; the watcher's 18 Jul check case)
6 Oct	Christmas launch (owned-channel drop; distorts October baselines)

Tomato ad batches (ads created per day): 2 Jun (29), 17 Jun (25), 14 Jul (30), 25 Jul (15), 5 Aug (44), 14 Aug (2), 15 Aug (4), 13 Sept (1), 24 Sept (13), 25 Sept (9), 12 Oct (6), 19 Oct (1). Median gap between batches ≈ 11 days. The 2–3 week brief→launch assumption cannot be checked from Meta: almost every creative file was uploaded the same day its ad was created, so production time leaves no trace in the account. What can be seen: Ad 707's return started falling 2 Sep, it was switched off 19 Sep, and the next tomato batch went live 24–25 Sep (23 days after the drop), consistent with roughly three weeks.

Ad 707 · Image · UGC Testimonial

Ad Set 1 - MAIN, Evergreen (old Tomato vine) CBO. Live 14 Aug to 19 Sep 2026.

Outcome-drop week 2 Sept–8 Sept ROAS 1.60x then 1.06x vs baseline 2.01x (bar 1.61x)Net of account -15%, -34%Sales in the 2 weeks: 61 vs 83 expected Confounders: account ROAS -20%

What moved first

New reach per £1k first moved (and stayed moved) the week of 19 Aug, 2 weeks before the drop (-40%).

Click rate never moved (net of the account it stayed within −8%). What moved was reach: new people per £1k fell 40% in week 2 and about 50% every week after, and CPM vs the account rose 26% in week 2. Conversion rate and basket size fell only once the drop had started (CVR −21%, AOV −12% the week of 9 Sep). Its own weekly frequency stayed at 1.05–1.28, and its ad set's 7-day frequency never passed 2.45 against the 2.5 ceiling. This is an ad using up its pocket of a broad audience, fast, while Meta scaled the campaign (budget £488 → £800 a day on 20 Aug, £960 on 30 Aug, £1,100 on 4 Sep).

Week by week

Week	Spend	Share	ROAS	Base	CPA	CTR	CTR net Δ	CPM vs acct Δ	Freq	New reach/£1k Δ	CVR	AOV	Spend vs base	Acct ROAS Δ	Confounders
12 Aug	£1,041	10%	2.17x	–	£31	1.45%	–	–	1.18	75,374 (–)	2.5%	£66	–	–
19 Aug	£2,696	31%	1.89x	2.17x	£39	1.46%	+0%	+26%	1.24	45,397 (-40%)	2.5%	£73	+159%	+2%	scaled (spend >=1.5x baseline)
26 Aug	£1,658	20%	2.10x	1.97x	£35	1.16%	-8%	+7%	1.28	26,973 (-50%)	3.6%	£73	-11%	+49%
2 Sept drop	£1,603	20%	1.60x	2.01x	£42	1.24%	-1%	+11%	1.20	23,468 (-48%)	3.0%	£67	-11%	-7%
9 Sept	£1,356	16%	1.06x	1.92x	£59	1.44%	+7%	+11%	1.19	18,095 (-55%)	2.2%	£62	-22%	-20%	account ROAS -20%
16 Sept	£873	8%	1.17x	1.72x	£55	1.30%	0%	-5%	1.05	29,176 (-7%)	2.2%	£64	-52%	-23%	pulled back (spend <0.5x baseline); account ROAS -23%

Signal timeline

Each cell is that week vs the ad's previous up-to-4 weeks. Blue = a leading signal past its bar, red = an outcome (ROAS, CPA) past its bar; the outlined column is the outcome-drop week.

Signal (trip bar)	19 Aug	26 Aug	2 Sept	9 Sept	16 Sept
CTR, net of account ≤ −10%	+0%	-8%	-1%	+7%	0%
CPM vs account ≥ +15%	+26%	+7%	+11%	+11%	-5%
Ad weekly frequency ≥ +10%	+5%	+5%	-3%	-4%	-15%
New reach per £1k ≤ −25%	-40%	-50%	-48%	-55%	-7%
CVR ≤ −20%	-3%	+46%	+9%	-21%	-20%
ROAS (outcome) ≤ −20%	-13%	+7%	-20%	-45%	-32%
CPA (outcome) ≥ +25%	+26%	-4%	+19%	+60%	+34%

Daily (7-day rolling)

ROAS, 7-day rolling (dashed: 2.46x breakeven)

Link CTR ÷ account CTR, 7-day

CPM ÷ account CPM, 7-day

New reach per £1k spend, by week (from cumulative reach)

Daily spend

CVR (purchases ÷ link clicks), 7-day

Old vs new watcher, every day

Ad 707 (Image, UGC Testimonial)

old

ssnnnnnnnnnnnAAFFFHHHHmmmmFFmAFAFAAAFAAppsssssss

new

ssnnnbbbbbbbbBBBLLLLLLLLLLLLLLLLLLLLLLLLLsssssss

14 Aug21 Aug28 Aug4 Sept11 Sept18 Sept25 Sept

Ad 551 · Video · Unboxing (one creative, two ad IDs)

OG in Ad Set 2 from 17 Jun; on 5 Aug Ad Set 2 was switched off and the same video copied into Ad Set 1 (the Copy), which ran to 25 Sep.

Outcome-drop week 12 Aug–18 Aug ROAS 1.73x then 0.71x vs baseline 2.25x (bar 1.80x)Net of account -23%, -68%Sales in the 2 weeks: 22 vs 34 expected Confounders: pulled back (spend <0.5x baseline); AOV -26%

What moved first

CTR (net of account) first moved (and stayed moved) the week of 8 Jul, 5 weeks before the drop (-16%).

The classic wear-out. Click rate net of the account fell from the week of 8 Jul (−16%, −14%, −26%, −11%) while ROAS held at 2.3x–2.6x, then hold rate followed (−11% the week of 29 Jul). Hook rate stayed flat at 27–30% throughout. CPM vs the account only rose at the drop. The drop itself landed on the Copy: once Ad 707 launched on 14 Aug, Meta moved the campaign's money onto it and the Copy fell to £223 a week at 0.71x. So the drop is fatigue that Meta's own budget move exposed. Both watchers lost the history at the copy: the Copy was a new ad ID with no baseline.

Week by week

Week	Spend	Share	ROAS	Base	CPA	CTR	CTR net Δ	CPM vs acct Δ	Hook	Hold	CVR	AOV	Spend vs base	Acct ROAS Δ	Confounders
17 Jun	£688	12%	2.51x	–	£29	1.58%	–	–	27.0%	18.4%	4.1%	£72	–	–
24 Jun	£556	10%	2.82x	2.51x	£22	1.86%	+18%	-4%	28.9%	16.6%	4.3%	£63	-19%	+2%	AOV -13%
1 Jul	£408	7%	3.02x	2.65x	£24	1.78%	+4%	-7%	29.8%	17.0%	3.4%	£72	-34%	-2%
8 Jul	£832	12%	2.53x	2.74x	£29	1.41%	-16%	-7%	29.6%	16.2%	3.5%	£72	+51%	-8%	scaled (spend >=1.5x baseline)
15 Jul	£418	6%	1.51x	2.67x	£42	1.37%	-14%	-14%	28.5%	16.0%	2.1%	£63	-33%	-32%	account ROAS -32%
22 Jul	£598	9%	2.59x	2.50x	£28	1.11%	-26%	-2%	26.7%	15.6%	4.1%	£74	+8%	+8%
29 Jul	£1,061	13%	2.29x	2.44x	£29	1.14%	-11%	-9%	28.0%	14.3%	3.6%	£66	+88%	+8%	scaled (spend >=1.5x baseline)
5 Aug	£1,723	16%	2.30x	2.31x	£32	1.13%	-9%	-3%	27.5%	14.8%	3.8%	£73	+137%	+8%	scaled (spend >=1.5x baseline)
12 Aug drop	£851	8%	1.73x	2.25x	£45	1.19%	+2%	+21%	28.1%	15.3%	3.1%	£78	-10%	-1%
19 Aug	£223	3%	0.71x	2.22x	£74	1.44%	+27%	+54%	26.5%	16.7%	1.9%	£53	-79%	-3%	pulled back (spend <0.5x baseline); AOV -26%
26 Aug	£377	5%	0.75x	2.08x	£63	1.26%	+21%	+46%	27.2%	15.2%	2.7%	£47	-61%	+45%	pulled back (spend <0.5x baseline); AOV -34%
2 Sept	£253	3%	1.07x	1.85x	£51	1.15%	+6%	+26%	23.1%	16.3%	3.5%	£54	-68%	-5%	pulled back (spend <0.5x baseline); AOV -25%
9 Sept	£311	4%	2.67x	1.28x	£24	1.41%	+15%	-6%	25.8%	14.9%	6.2%	£64	-27%	-20%	account ROAS -20%
16 Sept	£750	7%	0.92x	1.32x	£58	1.20%	-6%	-2%	28.3%	13.1%	3.2%	£53	+158%	-23%	scaled (spend >=1.5x baseline); account ROAS -23%
23 Sept	£450	3%	0.34x	1.23x	£225	0.92%	-23%	+17%	28.5%	12.9%	1.3%	£76	+6%	-22%	account ROAS -22%

Weeks run Wednesday to Tuesday so the last one ends 20 Oct. Δ columns compare each week with the ad's previous up-to-4 weeks (ratio of sums, as the watcher does). Shaded = past the trip bar (blue = leading signal, red = outcome). Frequency and new reach are per ad ID and cannot be added across OG and Copy (reach is not additive), so they are shown per ID in the charts below.

Signal timeline

Each cell is that week vs the ad's previous up-to-4 weeks. Blue = a leading signal past its bar, red = an outcome (ROAS, CPA) past its bar; the outlined column is the outcome-drop week.

Signal (trip bar)	24 Jun	1 Jul	8 Jul	15 Jul	22 Jul	29 Jul	5 Aug	12 Aug	19 Aug	26 Aug	2 Sept	9 Sept	16 Sept	23 Sept
CTR, net of account ≤ −10%	+18%	+4%	-16%	-14%	-26%	-11%	-9%	+2%	+27%	+21%	+6%	+15%	-6%	-23%
CPM vs account ≥ +15%	-4%	-7%	-7%	-14%	-2%	-9%	-3%	+21%	+54%	+46%	+26%	-6%	-2%	+17%
Hook rate ≤ −10%	+7%	+7%	+4%	-1%	-9%	-2%	-2%	+1%	-4%	-2%	-16%	-5%	+10%	+6%
Hold rate ≤ −10%	-10%	-3%	-7%	-5%	-5%	-11%	-4%	+3%	+13%	+3%	+8%	-4%	-16%	-10%
CVR ≤ −20%	+4%	-18%	-12%	-44%	+21%	+7%	+12%	-13%	-47%	-24%	+4%	+114%	-13%	-65%
ROAS (outcome) ≤ −20%	+12%	+14%	-8%	-43%	+4%	-6%	0%	-23%	-68%	-64%	-42%	+108%	-31%	-73%
CPA (outcome) ≥ +25%	-22%	-6%	+15%	+60%	+4%	-2%	+6%	+44%	+130%	+84%	+31%	-54%	+34%	+392%

Daily (7-day rolling)

ROAS, 7-day rolling (dashed: 2.46x breakeven)

Link CTR ÷ account CTR, 7-day

CPM ÷ account CPM, 7-day

Hold rate (ThruPlays ÷ 3s views), 7-day

Daily spend

CVR (purchases ÷ link clicks), 7-day

Old vs new watcher, every day

Ad 551 OG (Video, Unboxing) in Ad Set 2

old

snnnnnnnnnnnn···········mmAmm·HHmmAAFACAAHHHHHHHHHHmm·ssssss

new

snnnnbbbbbbbbbbbbbbb···bbbmbbbBBBBBBLLLLLLLLLLLLBBBBBBssssss

17 Jun24 Jun1 Jul8 Jul15 Jul22 Jul29 Jul5 Aug12 Aug

Ad 551 Copy (same video) in Ad Set 1

old

snnnnnnnnnnnnpppssssspppCCCCsssssspppHH····mmmHHAAAAAAFFsssss

new

snnnnbbbbbbbbbbbsssssmmmBLLLssssssLmmBBBBBBBBBBBBBBBBBLLsssss

5 Aug12 Aug19 Aug26 Aug2 Sept9 Sept16 Sept23 Sept30 Sept

Ad 487 · Image · Refillable Starter Pack

Ad Set 1 - MAIN. Live 2 Jun; deleted 13 Sep 2026.

Outcome-drop week 12 Aug–18 Aug ROAS 1.48x then 0.59x vs baseline 2.24x (bar 1.80x)Net of account -34%, -73%Sales in the 2 weeks: 7 vs 11 expected Confounders: pulled back (spend <0.5x baseline); AOV -21%

What moved first

CTR (net of account) first moved (and stayed moved) the week of 29 Jul, 2 weeks before the drop (-13%).
CVR (purchases/click) first moved (and stayed moved) the week of 15 Jul, 4 weeks before the drop (-77%).

Click rate net of the account was −20% the week of 8 Jul and ran −9% to −23% from 29 Jul. The outcome drop on 12 Aug came with Meta already pulling spend (−63% vs baseline) and a smaller basket (AOV −21%), the week after 44 new tomato ads went live in the same campaign on 5 Aug. CVR's fall on 15 Jul was the account-wide site dip (account ROAS −32% that week), not this ad. The 15 Jul and 22 Jul weeks also fail the over-scaled guard (spend 2-2.5x baseline), so they cannot be the drop week.

Week by week

Week	Spend	Share	ROAS	Base	CPA	CTR	CTR net Δ	CPM vs acct Δ	Freq	New reach/£1k Δ	CVR	AOV	Spend vs base	Acct ROAS Δ	Confounders
27 May	£2	0%	0.00x	–	–	0.00%	–	–	1.05	68,553 (–)	–	–	–	–
3 Jun	£2	0%	0.00x	0.00x	–	0.50%	–	-4%	1.12	76,522 (+12%)	0.0%	–	+45%	-12%
10 Jun	£243	4%	2.76x	0.00x	£27	1.55%	+404%	+4%	1.28	65,887 (-10%)	2.8%	£74	+12382%	-24%	scaled (spend >=1.5x baseline); account ROAS -24%
17 Jun	£109	2%	3.46x	2.71x	£22	1.73%	+29%	-1%	1.16	57,337 (-13%)	3.0%	£75	+32%	-13%
24 Jun	£84	1%	3.62x	2.94x	£17	1.51%	+3%	+17%	1.19	39,967 (-37%)	5.1%	£61	-5%	-9%	AOV -18%
1 Jul	£228	4%	3.21x	3.08x	£19	1.71%	+8%	-5%	1.32	51,673 (-12%)	3.0%	£61	+108%	-5%	scaled (spend >=1.5x baseline); AOV -14%
8 Jul	£323	5%	5.75x	3.14x	£12	1.25%	-20%	+5%	1.35	42,877 (-24%)	7.0%	£66	+95%	-7%	scaled (spend >=1.5x baseline)
15 Jul	£651	10%	1.41x	4.39x	£54	1.46%	-1%	-9%	1.36	58,599 (+24%)	1.1%	£76	+250%	-32%	scaled (spend >=1.5x baseline); account ROAS -32%
22 Jul	£682	10%	2.56x	2.96x	£27	1.45%	+3%	-2%	1.32	55,394 (+6%)	2.1%	£70	+112%	+8%	scaled (spend >=1.5x baseline)
29 Jul	£589	7%	2.61x	2.78x	£22	1.17%	-13%	-2%	1.37	51,758 (-4%)	3.2%	£57	+25%	+8%	AOV -17%
5 Aug	£334	3%	2.60x	2.69x	£26	1.23%	-9%	-11%	1.20	57,570 (+7%)	2.7%	£67	-40%	+8%
12 Aug drop	£210	2%	1.48x	2.24x	£35	1.03%	-23%	+17%	1.30	36,404 (-35%)	2.9%	£52	-63%	-1%	pulled back (spend <0.5x baseline); AOV -21%
19 Aug	£115	1%	0.59x	2.46x	£115	1.01%	-21%	+39%	1.36	20,885 (-60%)	1.0%	£69	-75%	-3%	pulled back (spend <0.5x baseline)
26 Aug	£196	2%	2.38x	2.23x	£39	1.07%	+2%	+35%	1.43	16,907 (-65%)	3.0%	£93	-37%	+45%
2 Sept	£190	2%	1.37x	2.01x	£63	0.87%	-17%	+12%	1.36	27,410 (-28%)	2.2%	£87	-11%	-5%
9 Sept	£82	1%	1.08x	1.56x	£82	0.88%	-11%	+3%	1.27	13,152 (-50%)	2.0%	£89	-54%	-20%	pulled back (spend <0.5x baseline); account ROAS -20%

Signal timeline

Each cell is that week vs the ad's previous up-to-4 weeks. Blue = a leading signal past its bar, red = an outcome (ROAS, CPA) past its bar; the outlined column is the outcome-drop week.

Signal (trip bar)	3 Jun	10 Jun	17 Jun	24 Jun	1 Jul	8 Jul	15 Jul	22 Jul	29 Jul	5 Aug	12 Aug	19 Aug	26 Aug	2 Sept	9 Sept
CTR, net of account ≤ −10%		+404%	+29%	+3%	+8%	-20%	-1%	+3%	-13%	-9%	-23%	-21%	+2%	-17%	-11%
CPM vs account ≥ +15%	-4%	+4%	-1%	+17%	-5%	+5%	-9%	-2%	-2%	-11%	+17%	+39%	+35%	+12%	+3%
Ad weekly frequency ≥ +10%	+7%	+17%	-10%	-4%	+7%	+7%	+5%	-2%	+2%	-11%	-2%	+3%	+9%	+6%	-6%
New reach per £1k ≤ −25%	+12%	-10%	-13%	-37%	-12%	-24%	+24%	+6%	-4%	+7%	-35%	-60%	-65%	-28%	-50%
CVR ≤ −20%			+6%	+79%	-6%	+122%	-77%	-25%	+31%	+3%	+35%	-61%	+4%	-17%	-18%
ROAS (outcome) ≤ −20%			+27%	+23%	+4%	+83%	-68%	-14%	-6%	-3%	-34%	-76%	+7%	-32%	-30%
CPA (outcome) ≥ +25%			-21%	-34%	-18%	-46%	+265%	+21%	-11%	+5%	+20%	+351%	+48%	+85%	+73%

Daily (7-day rolling)

ROAS, 7-day rolling (dashed: 2.46x breakeven)

Link CTR ÷ account CTR, 7-day

CPM ÷ account CPM, 7-day

New reach per £1k spend, by week (from cumulative reach)

Daily spend

CVR (purchases ÷ link clicks), 7-day

Old vs new watcher, every day

Ad 487 (Image, Refillable Starter Pack)

old

ssssssssssssssnnnnn·sssssssssssssm·····mmmHHHAmmmmmmmmmmmmmmmmmmmmHHmHCHHAAAssssssssssssssssssssssssssssssssss

new

ssssssssssssssnnnnn·sssssssssssssm·····mmmBBBBBBBBBBBbbbbbbbbbbbbmBBBBLLLLLLssssssssssssssssssssssssssssssssss

2 Jun9 Jun16 Jun23 Jun30 Jun7 Jul14 Jul21 Jul28 Jul4 Aug11 Aug18 Aug25 Aug1 Sept8 Sept15 Sept

The current watcher, replayed every day 1 Apr–20 Oct

The watcher's own code (backed up as watch.cjs.backup-21-oct-before-refresh) was run unchanged for each of 203 days against cached API data, with its ledger carried forward as the daily cron would. Before trusting it, the replay was checked against the four real-API backtests from 20 Oct (30 Jul, 27 Aug, 10 Sep, 24 Sep): 873 of 873 ad judgements identical. The new code's live run for 20 Oct also matched its replay (213 of 213).

H Highly likely (brief)C Close (refresh this week)F Fatigued (Cut today)A Monitor: account-wide suppressedp Monitor: pulled backm Monitor· Nonen Too news Too smallb Brief refreshB Build refreshL Launch refreshO Hand over outcome-drop week

Ad	Outcome-drop week	Old: first alert	Lead	Old: "Cut today"	New: Brief	New: Build	New: Launch	Stage changes old → new
Ad 707	2 Sept	29 Aug	4 days	29 Aug (4 days)	19 Aug (14 days)	27 Aug (6 days)	30 Aug (3 days)	16 → 5
Ad 551 OG	12 Aug	17 Jul	26 days	23 Jul (20 days)	22 Jun (51 days)	17 Jul (26 days)	23 Jul (20 days)	18 → 10
Ad 551 Copy	12 Aug	29 Aug	-17 days	28 Sept (-47 days)	10 Aug (2 days)	29 Aug (-17 days)	30 Aug (-18 days)	14 → 12
Ad 487	12 Aug	14 Jul	29 days	never	14 Jul (29 days)	14 Jul (29 days)	11 Aug (1 day)	19 → 12

Lead = days from the first call to the first day of the outcome-drop week (negative = after it). Ad 551's drop is the creative's (12 Aug, OG and Copy combined). The Copy is a new ad ID, so both watchers judged it from scratch: too new until 17 Aug, then "pulled back".

Which thresholds were too loose or too tight

Too-new gate, too tight for big winners. Ad 707 was "too new" for its first 12 days while taking 31% of spend. Its return fell on day 19. Fixed by the winner brief.
Account-wide guard, too blunt. With 4–11 material ads, "over half trip one family" happened on 101 of 203 days. It hid Ad 707 on 9 days and switched on and off daily. On the merits it was right for cut-level calls: the calls it held back went on to 0.92x of baseline net of the account, like any ad. So it now holds back only a fresh Launch, and never a Brief or Build.
Breakeven line, too loose for a cut. At 2.46x against an account running ~1.8x, nearly every ad is "below breakeven", so "Cut today" came down to a hard return fall. It fired on Ad 551 OG on 23 Jul, during the account-wide site dip, and on Ad 1000 on 20 Oct. Moot now that nothing is cut.
Ad set frequency 2.5 / 3.0, never reachable here. Tomato ad sets peaked at 2.45 (Ad Set 1) and 1.41 (Ad Set 2). Kept, but it is the new-reach check that sees saturation on broad prospecting.
CTR −10% and hold −10%, about right. They caught Ad 551 five and two weeks ahead. Hook −10% never tripped on Ad 551 (flat 27–30%).
"Return moved, no leading signal = maybe attribution lag", too lenient. Those Monitor calls went on to 0.73x of baseline (57% under 0.8x). Not changed here (minimal changes); worth a look next.

The refresh ladder (implemented)

Stage	Action	Trigger	Basis
m Monitor	Watch	Unchanged: one signal alone, return moving with no leading signal, over-scaled, pulled back, check site, audience-not-creative (ad set frequency only).	[EXISTING]
b Brief refresh	Brief the refresh now: same angle and offer, new hook, opening frame, thumbnail, first line or format	A winner: 10%+ of 7-day spend, ROAS at or above the account's 7-day ROAS, 10+ sales this week. Too-new ads qualify.	[EVIDENCE: tomato] Ad 707 took 31% of spend in week 2 and its return fell in week 4, faster than a 2-3 week build. Share bar = existing bigShare [EXISTING]; ROAS ≥ account and 10 sales [JUDGEMENT].
B Build refresh	Put the refresh into production now, live next to the original within 2 weeks	Old Highly likely, unchanged: both leading families tripped, one Strong, or one still tripped a week later. Return not yet hard.	[EXISTING] trigger. [EVIDENCE: tomato] fired 26-29 days before the drop on Ads 551 and 487; across Pott these ads earned 1.14x baseline net of the account over the next 14 days, so the stage is early, not wrong.
L Launch refresh	Launch the refresh in the same ad set as the original now. Never pause the original.	Old Fatigued or Close: return hard plus a leading family, Strong leading plus Moderate return, purchases −50%, or a big spender under half its best week. On an account-wide run a fresh Launch drops to Build (if its leading signals alone justify it) or Monitor.	[EXISTING] triggers. Pott Apr-Oct: old Fatigued and Close calls went on to 0.44x and 0.55x of baseline over the next 14 days, net of the account; new Launch (held days included) 0.70x. Account-wide demotion [EXISTING], now only for a fresh Launch.
O Hand over	Let the refresh take the budget; wind the original down once the refresh carries most of the concept for 7 days	A refresh is live: same campaign, same product + angle name fields, different ad number and different image/video, launched 3+ days later (or named "Refresh of 707", or mapped in config). It has £100+ and 5+ sales this week, ROAS ≥ 0.9x the original's this week and ≥ 0.8x the original's baseline.	[JUDGEMENT] No tomato refresh was named so it could be matched; Ad 1035 showed the duplicate trap (same image as Ad 1000).

Other changes

Change	Rule	Basis
New reach check (audience family)	New people per £1k vs baseline weeks, from cumulative reach since the ad was created. Moderate −45%, strong −55%.	[EVIDENCE: tomato] Ad 707's click rate never moved; new reach per £ ran −45% to −55% from 27 Aug. Ad 551 held within ±20%. Bars set at the worst ~10% / ~4% of Pott's material ad-days, because new reach falls with age anyway (median −26% under 3 weeks).
Sticky stages	An ad keeps the highest stage it earned in the last 7 days. Held days do not count, so a hold cannot extend itself.	[JUDGEMENT] [EVIDENCE: tomato] the old stages changed 16-19 times per tomato ad (Ad 707: monitor/fatigued/monitor on 11-13 Sep). New: 5-12.
Audience-not-creative	Only when the ad set's frequency trips. A CPM or new-reach move on one ad in a broad ad set is the creative saturating, so it goes up the ladder.	[EVIDENCE: tomato] Ad 707 (broad, frequency ~1.2).
Young-ad and no-breakeven caps	Still computed, but both Fatigued and Close map to Launch, so they no longer change the stage (they only guarded a cut).	[EXISTING]
Everything else	Thresholds, windows, chance test, bigger-baskets guard, check-site, over-scaled, pulled back, persistence, families: unchanged.	[EXISTING]

Wording in every action, card, email line and report heading now says refresh/launch/hand over. Nothing anywhere says cut or pause. Cron and email are unchanged and still off.

Sanity check: does it fire on everything?

Every material Pott ad, every day, 1 Apr–20 Oct (about 6 material ads a day). A refresh stage lasts longer than an old alert, because it holds for 7 days, so the count of ads opening a refresh is the fairer comparison.

Month	Old: ads entering Highly likely+	Old: entering Close/Fatigued	New: ads entering Brief+	New: entering Launch
2026-04	14	5	15	5
2026-05	7	5	10	5
2026-06	8	3	7	3
2026-07	10	6	9	8
2026-08	8	6	7	6
2026-09	2	0	5	0
2026-10	8	4	8	4
Total	57	29	61	31

An entry = an ad reaching the stage for the first time in 30 days. Daily replay, 1 Apr to 20 Oct 2026.

Share of ad-days at each stage

Old stage	Ad-days	New stage	Ad-days
Fatigued (cut today)	36 (3%)	Launch refresh	196 (15%)
Close	39 (3%)	Hand over	3 (0%)
Highly likely	228 (19%)	Build refresh	402 (32%)
		Brief refresh	152 (12%)
Monitor	557 (46%)	Monitor	310 (24%)
None	338 (28%)	None	206 (16%)
Material ad-days	1198	Material ad-days	1269

New Build/Launch include held days (about a third of all flagged days). On a typical day: ~6.3 material ads, ~1 at Launch, ~2 at Build, under 1 at Brief.

What happened next: ROAS over the following 14 days, by stage

Weekly samples (Tuesdays 7 Apr–6 Oct), each ad's next-14-day ROAS divided by its own 28-day baseline. "Net of account" divides again by the account's own move over the same windows, so a falling account does not flatter or punish a stage.

Old rules

Stage	Calls	Still spending	Next 14d ROAS ÷ baseline	Same, net of account	Share under 0.8x (net)
fatigued	5	4	0.47x	0.44x	75%
close	7	6	0.43x	0.55x	67%
highly-likely	31	29	1.12x	1.14x	38%
monitor (account-wide: held-back Fatigued/Close)	12	9	0.88x	0.92x	33%
monitor	54	49	0.73x	0.73x	57%
none	38	35	0.96x	1.04x	29%

New rules

Stage	Calls	Still spending	Next 14d ROAS ÷ baseline	Same, net of account	Share under 0.8x (net)
launch-refresh	26	23	0.63x	0.70x	65%
build-refresh (account-wide day: held back from Launch)	8	5	0.68x	0.83x	40%
build-refresh	42	40	1.12x	1.14x	33%
brief-refresh	20	14	0.88x	0.82x	50%
monitor	33	29	0.63x	0.66x	69%
none	23	22	0.96x	1.06x	14%

Small samples (5–47 calls per stage). Read the ordering, not the decimals. Launch and Monitor ads went on to fall; Build and None ads kept earning. Build is an early stage by design.

Pott's top spenders since 1 Apr, old vs new, every day

Ad 1000 Image · Holly · Customer Quote Tree Without Needle Drop £13,304 since 1 Apr at 1.60x · 7 Oct to 20 Oct

old

snnnnnnnnnnnnF

new

snbbbbbbbbbbOL

First old Highly likely –, Close –, Fatigued 20 Oct · First new Brief 9 Oct, Build 19 Oct, Launch 19 Oct, Hand over 19 Oct

Ad 922 Video · Seasalt Starter Pack · Buy Once Refill Forever £10,335 since 1 Apr at 1.41x · 25 Sept to 20 Oct

old

snnnnnnnnnnnnCC···········

new

snbbbbbbbbnnnLLLLLLLLLBBBB

First old Highly likely –, Close 8 Oct, Fatigued – · First new Brief 27 Sept, Build 8 Oct, Launch 8 Oct

Ad 707 Image · Tomato Vine · UGC Testimonial £9,226 since 1 Apr at 1.73x · 14 Aug to 30 Sept

old

ssnnnnnnnnnnnAAFFFHHHHmmmmFFmAFAFAAAFAAppsssssss

new

ssnnnbbbbbbbbBBBLLLLLLLLLLLLLLLLLLLLLLLLLsssssss

First old Highly likely 1 Sept, Close –, Fatigued 29 Aug · First new Brief 19 Aug, Build 27 Aug, Launch 30 Aug

358 Video · Starter Pack · Grown-Up Candles £5,337 since 1 Apr at 1.16x · 4 Apr to 23 Jun

old

ssssssnsnnnnnHHHHHHHHH··mmm···········mppHHCAHsssssssssssssspHHHHHHHHHHAAAAAsssss

new

ssssssnsnnnnnBBBBBBBBBBBBBBBBbbbbbbbbbbbbBBLLLssssssssssssssmBBBBBBBBBBBBBBBsssss

First old Highly likely 17 Apr, Close 17 May, Fatigued – · First new Brief 17 Apr, Build 17 Apr, Launch 17 May

Ad 551 Video · Tomato Vine · Unboxing £4,934 since 1 Apr at 1.50x · 5 Aug to 4 Oct

old

snnnnnnnnnnnnpppssssspppCCCCsssssspppHH····mmmHHAAAAAAFFsssss

new

snnnnbbbbbbbbbbbsssssmmmBLLLssssssLmmBBBBBBBBBBBBBBBBBLLsssss

First old Highly likely 11 Sept, Close 29 Aug, Fatigued 28 Sept · First new Brief 10 Aug, Build 29 Aug, Launch 30 Aug

Ad 551 Video · Tomato Vine · Unboxing £4,564 since 1 Apr at 2.55x · 17 Jun to 15 Aug

old

snnnnnnnnnnnn···········mmAmm·HHmmAAFACAAHHHHHHHHHHmm·ssssss

new

snnnnbbbbbbbbbbbbbbb···bbbmbbbBBBBBBLLLLLLLLLLLLBBBBBBssssss

First old Highly likely 17 Jul, Close 25 Jul, Fatigued 23 Jul · First new Brief 22 Jun, Build 17 Jul, Launch 23 Jul

Ad 554 Image · Roses Refill · Round 2 £4,316 since 1 Apr at 2.28x · 24 Jun to 11 Sept

old

ssssssnnnsssssssssssmmmHHHHAmmmmmmmmmmmmmmmmmmmmmmmmmmmAAAFFCCCCAAssssssssssssss

new

ssssssnnnsssssssssssmmmBBBBBBBBBBBBmmmmmmmmmmmmmmmmmmmmmmBLLLLLLLLssssssssssssss

First old Highly likely 17 Jul, Close 23 Aug, Fatigued 21 Aug · First new Brief 17 Jul, Build 17 Jul, Launch 21 Aug

Ad 465 Video · Cook Candle · Refill Lead £4,167 since 1 Apr at 2.35x · 20 May to 4 Jul

old

sssnnnnnnnnnn····mAm···mmmHHHHHmAAAmmmppssssss

new

sssnnnnnbbbbbbbbbbmbbbbbbbBBBBBBBBBBBBBBssssss

First old Highly likely 15 Jun, Close –, Fatigued – · First new Brief 28 May, Build 15 Jun, Launch –

Ad 366 Video · Starter Pack · Buy One You Can Refill £4,151 since 1 Apr at 1.12x · 3 Apr to 7 Jul

old

sssssssssnnnnHHHHHHHHHHHHHHHHHHHHmmHHHHH·······mmmHHssssssssspppHHHHAAAAAAHAAHH···ssssAACCssssss

new

sssssssssnnnnBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBmmmBBsssssssssmmmBBBBBBBBBBBBBBBBBBssssBBLLssssss

First old Highly likely 16 Apr, Close 30 Jun, Fatigued – · First new Brief 16 Apr, Build 16 Apr, Launch 30 Jun

Ad 487 Image · Tomato Vine · Refillable Starter Pack £4,041 since 1 Apr at 2.52x · 2 Jun to 19 Sept

old

ssssssssssssssnnnnn·sssssssssssssm·····mmmHHHAmmmmmmmmmmmmmmmmmmmmHHmHCHHAAAssssssssssssssssssssssssssssssssss

new

ssssssssssssssnnnnn·sssssssssssssm·····mmmBBBBBBBBBBBbbbbbbbbbbbbmBBBBLLLLLLssssssssssssssssssssssssssssssssss

First old Highly likely 14 Jul, Close 11 Aug, Fatigued – · First new Brief 14 Jul, Build 14 Jul, Launch 11 Aug

Ad 489 Image · Tomato Vine · Three Wick Tower £3,598 since 1 Apr at 2.47x · 2 Jun to 5 Oct

old

ssssnnnnnnnnnmAAssAAAHAAACAAFFFAAmsssssssssssssmmmmmmmmAsssssssssssssssssAAssssssssssssssssssssssssmmmsssssssssss·····ssssssss

new

ssssnnnnnnnnnmBBssBBBBBBBLLLLLLLLLsssssssssssssmmmmmmmmmsssssssssssssssssmmssssssssssssssssssssssssmmmsssssssssss·····ssssssss

First old Highly likely 23 Jun, Close 27 Jun, Fatigued 30 Jun · First new Brief 16 Jun, Build 16 Jun, Launch 27 Jun

Ad 940 Image · Tomato Vine · Maker So Happy £3,565 since 1 Apr at 1.26x · 25 Sept to 20 Oct

old

snnnnnnnnnnnn··········ppp

new

snnnnnnnnnnnn···mmmBBBBBBB

First old Highly likely –, Close –, Fatigued – · First new Brief 14 Oct, Build 14 Oct, Launch –

Pott today: the refresh list as of 20 Oct 2026

The new watcher run against the live API (dry run, in a sandbox so Pott's real ledger, state and last-result files were not touched), as of 20 Oct, with the ledger carried from the replay. Account: 7-day spend £27,174, ROAS 1.72x → 1.67x. All ads listed are ACTIVE today.

L Launch refresh (2)

Ad	7d spend	Share	ROAS base → 7d	Signals	Old watcher said	Action
Ad 1000 Image · Holly · Customer Quote Tree Without Needle Drop Christmas - Broad - Exclude Customers · 13 days · preview	£7,193	27%	1.95x → 1.30x	ctr:moderate, cpa:strong, peak:moderate	fatigued	Launch the refresh in the same ad set now. Do not pause the original. No refresh of it is live yet.
Ad 983 Image · Icicle · New Glaze Fresh From The Kiln Christmas - Broad - Exclude Customers · 13 days · preview	£1,433	5%	2.32x → 1.70x	ctr:moderate, cpa:moderate, peak:strong	close	Launch the refresh in the same ad set now. Do not pause the original. No refresh of it is live yet.

B Build refresh (6)

Ad	7d spend	Share	ROAS base → 7d	Signals	Old watcher said	Action
Ad 922 Video · Seasalt Starter Pack · Buy Once Refill Forever Everything Else - Broad - Exclude Customers · 25 days · held from earlier run · preview	£1,598	6%	1.38x → 1.60x	–	none	Put the refresh into production; live within 2 weeks. No refresh of it is live yet. Held: it earned this stage within the last 7 days.
Ad 995 Image · Icicle · Meet Icicle Limited Edition Christmas - Broad - Exclude Customers · 13 days · preview	£850	3%	2.14x → 1.76x	ctr:strong	highly-likely	Put the refresh into production; live within 2 weeks. No refresh of it is live yet.
Ad 973 Video · Fir · Smells Like A Christmas Tree Christmas - Broad - Exclude Customers · 13 days · preview	£648	2%	2.14x → 4.50x	ctr:strong, hold:moderate	highly-likely	Put the refresh into production; live within 2 weeks. No refresh of it is live yet.
Ad 940 Image · Tomato Vine · Maker So Happy Tomato - Broad - Exclude Customers · 25 days · preview	£525	2%	1.34x → 0.79x	reach:moderate, cpa:moderate, peak:strong, purch:strong	monitor	Put the refresh into production; live within 2 weeks. No refresh of it is live yet.
Ad 837 Image · Fig · Scent Led Seasonal - Broad - Exclude Customers · 26 days · held from earlier run · preview	£470	2%	2.05x → 2.27x	–	none	Put the refresh into production; live within 2 weeks. No refresh of it is live yet. Held: it earned this stage within the last 7 days.
Ad 915 Video · Cook Candle · Review At The Hob Ad Set 1 - MAIN - Broad - Exclude Customers · 31 days · held from earlier run · preview	£323	1%	2.35x → 1.69x	peak:moderate	monitor	Put the refresh into production; live within 2 weeks. No refresh of it is live yet. Held: it earned this stage within the last 7 days.

m Monitor (2)

Ad	7d spend	Share	ROAS base → 7d	Signals	Old watcher said	Action
Ad 1008 Video · Starter Pack · Gifted Last Christmas Still Using It Gifting - Broad - Exclude Customers · 13 days	£717	3%	1.09x → 1.98x	cpm:moderate	monitor	Watch; no action yet.
Ad 1010 Image · Christmas Collection · Favourite Gift Last Christmas Gifting - Broad - Exclude Customers · 13 days	£353	1%	1.02x → 1.55x	ctr:moderate	monitor	Watch; no action yet.

· None (3)

Ad	7d spend	Share	ROAS base → 7d	Signals	Old watcher said
Ad 934 Image · Tomato Vine · Customer Quote Whole Hall Tomato - Broad - Exclude Customers · 26 days	£788	3%	1.20x → 1.40x	–	none
Ad 1006 Video · Starter Pack · Bought It For My Girlfriend Gifting - Broad - Exclude Customers · 13 days	£538	2%	1.70x → 2.25x	–	none
Ad 964 Video · Seasalt · Standing By The Sea Everything Else - Broad - Exclude Customers · 16 days	£398	2%	1.64x → 1.74x	–	none

n Too new (5)

Ad	7d spend	Share	ROAS base → 7d	Signals	Old watcher said
Ad 1020 Video · Cook Candle · One Candle For Your Kitchen Ad Set 1 - MAIN - Broad - Exclude Customers · 8 days	£1,056	4%	1.28x → 1.88x	–	too-new
Ad 1025 Video · Tomato Vine · Only Tomato Candle You Need Tomato - Broad - Exclude Customers · 8 days	£582	2%	3.40x → 2.91x	–	too-new
Ad 1011 Image · Christmas Collection · Gifted To Lots Of People Gifting - Broad - Exclude Customers · 13 days	£574	2%	0.00x → 1.20x	–	too-new
Ad 970 Video · Christmas Collection · Christmas Scents Are Here Christmas - Broad - Exclude Customers · 13 days	£413	2%	1.50x → 0.67x	–	too-new
Ad 1035 Image · Holly · Customer Quote Tree Without Needle Drop Christmas - Broad - Exclude Customers · 3 days	£345	1%	– → 1.49x	–	too-new

Read before acting. Ads 1000, 983, 995 and 973 went live on 7 Oct, the day after the 6 Oct Christmas launch. Their "best week" and part of their baseline is launch week, when the owned-channel drop inflated Meta-attributed revenue, so part of the fall is the launch spike ending. Check Shopify daily orders against the send calendar before reading these as fatigue. The refresh still makes sense for Ad 1000 at 26% of spend. Preview links are Meta's shareable previews.

Method and limits

Data. Graph API v23.0, account act_552116272670678, read-only. Ad-level daily insights 27 Jan–20 Oct (35,040 rows, 1,065 ads), ad set 7-day reach/frequency for every window ending 18 Mar–20 Oct (217 windows), account daily and weekly, per-ad weekly reach (time_increment=7), cumulative reach per ad from its created date (146 calls, 126 ads), the activity log 25 May–21 Oct (11,081 events). Raw files in data/raw/.
Definitions follow the watcher: purchases = "purchase", else offsite_conversion.fb_pixel_purchase; CTR = inline link clicks ÷ impressions; hook = 3-second views ÷ impressions; hold = ThruPlays ÷ 3-second views; CVR = purchases ÷ link clicks.
Outcome-drop week [JUDGEMENT]: the first Wed–Tue week where it and the next week both ran ROAS ≤ 0.8x the ad's previous-4-week baseline, both at least 10% worse than the account's own move, with enough spend to have bought 8+ sales at the baseline cost per sale, and not on an over-scaled week (spend ≥ 1.5x baseline, the watcher's own guard).
"First moved" = past its bar this week and again in at least one of the next two weeks, vs the ad's previous up-to-4 weeks.
Not verifiable from Meta: when refresh briefs started (asset upload dates equal ad creation dates), stock and site status (no Shopify pull in this run), whether any ad was meant as a "refresh" of another (naming does not say).
Replay caveats. In the replay every ad that spent is judged (as the watcher's own --as-of mode does); a live run judges only ACTIVE ads. The replay's ad metadata has no creative, so the duplicate-asset check only applies live: that is why the replay strip shows Ad 1000 at Hand over on 19–20 Oct (matched to Ad 1035, the same image), while the live run correctly keeps it at Launch. Weekly reach for Ad 551 cannot be combined across its two IDs.
Small numbers: three creatives and, for the forward test, 5–47 calls per stage. These are directions, not proofs.

Files: REPORT.md, CHANGES.md, data/ (raw + computed), scripts/ (every step, rerunnable). De-indexed meta tag included; not published anywhere.