Daily campaign report — 2026-08-08, against 2026-08-07

The account bought $745.02 and got 814 installs at $0.915, against $749.94 and 998 at $0.751 the day before. Booked revenue was $1,695.44; within-day was $996.74, up 128.1% on the same spend. Our records read 0.99 and 1.04 of Meta's booked figure on the account's own clock, inside the documented 0.91–1.06 band. The German campaign was off for the whole day, so the worldwide buy is 100% of it against 98.4% the day before, and this is the third complete day of the five-cell split test; definitions and the report clock are in the appendix.

1Overview

Delivery rows are the delivering campaigns summed. The retired earner's booking tail (1 install (Meta), 17 registrations (ours), $3.35 within-day, on $0.00 of spend) is excluded here and named in section 5, because counting it would put installs into a CPI denominator against spend that was never incurred.

	2026-08-07	2026-08-08	move
spend	$749.94	$745.02	−0.7%
impressions	81,142	75,033	−7.5%
CPM	$9.24	$9.93	+7.5%
clicks	5,016	4,202	−16.2%
CTR	6.18%	5.60%	−9.4%
installs (Meta)	998	814	−18.4%
click→install	19.90%	19.37%	−2.6%
CPI	$0.751	$0.915	+21.8%
booked (Meta)	$1,505.16 (2.01)	$1,695.44 (2.28)	+12.6%
within-day	$436.98 (0.58)	$996.74 (1.34)	+128.1%
cohort @ 24h	$472.45 (0.63)	$1,163.66 (1.56)	+146.3%

Cohort value — ARPU and ARPPU at d0, d3 and d7

Keyed on the INSTALL day and measured from each user's own install stamp: dN is their first (N+1) x 24 h. ARPU is Adjust's own series, the one the partner's dashboard reports; ARPPU is our payment ledger, because Adjust exposes no cumulative distinct-payer count. A horizon the window has not lived through yet is left empty rather than written as zero.

app	ARPU (d0)	ARPU (d3)	ARPU (d7)	ARPPU (d0)	ARPPU (d3)	ARPPU (d7)	installs
Vloom	$1.0710	$1.7803	$2.6828	$17.93	$26.19	$37.21	971

Campaign spend split, both days. Worldwide $737.89 (98.4%) → $745.02 (100.0%); German $12.05 (1.6%) → $0.00.

Delivery got worse for the fourth day running and the money side reversed. Same spend, 21.8% more per install, within-day revenue more than doubled.

The country mix moved. US installs 4.7% → 7.1%: +13% of the payer-rate move and +13% of revenue per install, 87% inside countries (section 3).

It rests on 47 payers, one of whom is a seventh of the day. Payer rate 3.84% → 5.99%, ARPPU $11.58 → $20.18.

The split test's ranking did not replicate on its third day. Spearman +0.70 between the first two days, +0.10 against the third (section 6).

2Hourly, 2026-08-08

Hours both days delivered in, on the UTC clock: 00 through 23. Only the shared hours are compared.

Every hour of the day itself, all three revenue measures beside the delivery funnel. Delivery columns are Meta's on both sides of each ratio; ours is our own registrations and every revenue column is ours. cov is that hour's cohort coverage at H=24, which falls to zero in the last hours because those installs have not yet lived a full day.

hr	spend	impr	CPM	clicks	CTR	inst	clk→i	CPI	ours	cov	booked	bROAS	bPay	within	wROAS	wPay	wARPU	wARPPU	cohort	cROAS	cPay	cARPU	cARPPU
0	$16.04	1474	10.88	100	6.78%	30	30.0%	$0.535	33	100%	$14.99	0.93	3	$5.61	0.35	1	0.170	5.61	$5.61	0.35	1	0.170	5.61
1	$21.25	2163	9.82	147	6.80%	38	25.9%	$0.559	40	100%	$24.35	1.15	5	$82.63	3.89	4	2.066	20.66	$82.63	3.89	4	2.066	20.66
2	$22.99	2373	9.69	132	5.56%	24	18.2%	$0.958	30	100%	$55.93	2.43	6	$7.98	0.35	1	0.266	7.98	$7.98	0.35	1	0.266	7.98
3	$23.47	2891	8.12	175	6.05%	26	14.9%	$0.903	31	100%	$24.23	1.03	3	$11.70	0.50	2	0.377	5.85	$11.70	0.50	2	0.377	5.85
4	$27.72	3027	9.16	172	5.68%	35	20.3%	$0.792	41	100%	$14.29	0.52	3	$248.72	8.97	3	6.066	82.91	$248.72	8.97	3	6.066	82.91
5	$25.04	2891	8.66	143	4.95%	26	18.2%	$0.963	28	100%	$177.85	7.10	6	$0.00	0.00	0	0.000	0.00	$0.00	0.00	0	0.000	0.00
6	$27.75	2728	10.17	149	5.46%	33	22.1%	$0.841	32	100%	$133.10	4.80	8	$56.89	2.05	3	1.778	18.96	$56.89	2.05	3	1.778	18.96
7	$21.19	2528	8.38	144	5.70%	33	22.9%	$0.642	36	100%	$110.07	5.19	6	$4.99	0.24	1	0.138	4.99	$4.99	0.24	1	0.138	4.99
8	$24.75	2878	8.60	146	5.07%	22	15.1%	$1.125	26	100%	$15.44	0.62	2	$9.89	0.40	1	0.380	9.89	$9.89	0.40	1	0.380	9.89
9	$24.82	2939	8.45	158	5.38%	29	18.4%	$0.856	33	100%	$29.14	1.17	4	$18.74	0.76	1	0.568	18.74	$18.74	0.76	1	0.568	18.74
10	$28.79	2958	9.73	141	4.77%	25	17.7%	$1.152	31	100%	$61.49	2.14	3	$5.01	0.17	1	0.161	5.01	$5.01	0.17	1	0.161	5.01
11	$34.41	3227	10.66	166	5.14%	33	19.9%	$1.043	36	100%	$14.89	0.43	3	$0.00	0.00	0	0.000	0.00	$3.56	0.10	1	0.099	3.56
12	$33.84	3202	10.57	186	5.81%	30	16.1%	$1.128	37	100%	$36.43	1.08	5	$43.64	1.29	5	1.180	8.73	$43.64	1.29	5	1.180	8.73
13	$39.62	3329	11.90	178	5.35%	41	23.0%	$0.966	43	100%	$106.09	2.68	9	$60.89	1.54	2	1.416	30.44	$110.12	2.78	2	2.561	55.06
14	$38.01	3059	12.43	191	6.24%	33	17.3%	$1.152	37	100%	$102.97	2.71	7	$18.19	0.48	2	0.492	9.09	$18.19	0.48	2	0.492	9.09
15	$37.26	4143	8.99	250	6.03%	45	18.0%	$0.828	55	100%	$35.58	0.95	8	$49.45	1.33	5	0.899	9.89	$54.44	1.46	6	0.990	9.07
16	$38.02	4231	8.99	239	5.65%	47	19.7%	$0.809	54	100%	$140.63	3.70	10	$30.04	0.79	3	0.556	10.01	$64.77	1.70	4	1.199	16.19
17	$41.53	5225	7.95	239	4.57%	42	17.6%	$0.989	41	100%	$117.67	2.83	10	$113.99	2.74	4	2.780	28.50	$113.99	2.74	4	2.780	28.50
18	$40.08	4614	8.69	232	5.03%	45	19.4%	$0.891	47	100%	$77.00	1.92	5	$47.33	1.18	2	1.007	23.67	$47.33	1.18	2	1.007	23.67
19	$36.45	3363	10.84	209	6.21%	37	17.7%	$0.985	43	100%	$22.82	0.63	4	$11.52	0.32	2	0.268	5.76	$11.52	0.32	2	0.268	5.76
20	$39.49	3465	11.40	202	5.83%	41	20.3%	$0.963	44	100%	$70.23	1.78	5	$32.21	0.82	2	0.732	16.10	$32.21	0.82	2	0.732	16.10
21	$34.65	2951	11.74	173	5.86%	30	17.3%	$1.155	32	100%	$78.03	2.25	7	$50.05	1.44	3	1.564	16.68	$50.05	1.44	3	1.564	16.68
22	$33.89	2774	12.22	192	6.92%	43	22.4%	$0.788	40	100%	$170.71	5.04	8	$81.34	2.40	2	2.034	40.67	$152.20	4.49	3	3.805	50.73
23	$33.96	2600	13.06	138	5.31%	26	18.8%	$1.306	32	100%	$61.50	1.81	3	$5.93	0.17	1	0.185	5.93	$9.49	0.28	1	0.297	9.49
ALL	$745.02	75033	9.93	4202	5.60%	814	19.4%	$0.915	902	100%	$1695.44	2.28	133	$996.74	1.34	51	1.105	19.54	$1163.66	1.56	55	1.290	21.16

Counting which way each shared hour points, which needs no weighting and asks only for a direction:

metric	hours 2026-08-08 ran higher	hours it ran lower	sign test
CPM	17 of 24	7 of 24	p = 0.064
CTR	5 of 24	19 of 24	p = 0.007
click→install	9 of 24	15 of 24	p = 0.307
CPI	19 of 24	5 of 24	p = 0.007

A sign test throws magnitude away and buys weight-free direction with it. CTR ran lower in 19 of the 24 shared hours and CPI higher in 19 of 24, so neither is an artifact of when the day bought. CPM misses 0.05 and click→install points nowhere.

Within-day revenue is one hour. Hour 04 alone carries $248.72 of the day's $996.74, a quarter of it, on three payers.

The full hour-by-hour pairing, 2026-08-07 → 2026-08-08

Every shared hour, 2026-08-07 → 2026-08-08. All columns are Meta's instrument on both sides of each ratio. Revenue is deliberately absent: per-hour payer counts run 0 to 5, so an hourly revenue column would be one person's basket read as a rate.

hr	impressions	CPM	CTR	click→install	CPI
00	2,615 → 1,474	10.65 → 10.88	6.88% → 6.78%	18.9% → 30.0%	0.819 → 0.535
01	3,149 → 2,163	8.18 → 9.82	6.38% → 6.80%	18.9% → 25.9%	0.678 → 0.559
02	3,370 → 2,373	8.15 → 9.69	6.62% → 5.56%	18.8% → 18.2%	0.654 → 0.958
03	3,416 → 2,891	8.47 → 8.12	6.24% → 6.05%	18.8% → 14.9%	0.723 → 0.903
04	3,507 → 3,027	9.71 → 9.16	5.82% → 5.68%	19.1% → 20.3%	0.873 → 0.792
05	3,656 → 2,891	9.79 → 8.66	6.18% → 4.95%	21.7% → 18.2%	0.731 → 0.963
06	3,287 → 2,728	6.53 → 10.17	6.27% → 5.46%	21.4% → 22.1%	0.487 → 0.841
07	3,081 → 2,528	8.24 → 8.38	5.10% → 5.70%	19.1% → 22.9%	0.846 → 0.642
08	3,560 → 2,878	7.99 → 8.60	6.38% → 5.07%	15.4% → 15.1%	0.813 → 1.125
09	3,873 → 2,939	8.76 → 8.45	5.19% → 5.38%	22.9% → 18.4%	0.738 → 0.856
10	3,994 → 2,958	9.20 → 9.73	5.78% → 4.77%	18.6% → 17.7%	0.855 → 1.152
11	4,115 → 3,227	9.57 → 10.66	5.66% → 5.14%	20.2% → 19.9%	0.838 → 1.043
12	4,018 → 3,202	8.91 → 10.57	6.22% → 5.81%	16.4% → 16.1%	0.873 → 1.128
13	4,017 → 3,329	8.45 → 11.90	5.58% → 5.35%	20.5% → 23.0%	0.738 → 0.966
14	3,510 → 3,059	11.44 → 12.43	6.95% → 6.24%	19.7% → 17.3%	0.836 → 1.152
15	4,316 → 4,143	7.88 → 8.99	6.16% → 6.03%	17.3% → 18.0%	0.740 → 0.828
16	3,608 → 4,231	8.03 → 8.99	6.32% → 5.65%	16.7% → 19.7%	0.762 → 0.809
17	4,110 → 5,225	8.41 → 7.95	6.52% → 4.57%	16.8% → 17.6%	0.768 → 0.989
18	3,700 → 4,614	11.26 → 8.69	7.14% → 5.03%	20.1% → 19.4%	0.786 → 0.891
19	3,206 → 3,363	10.71 → 10.84	5.36% → 6.21%	25.0% → 17.7%	0.799 → 0.985
20	2,799 → 3,465	10.31 → 11.40	6.11% → 5.83%	28.1% → 20.3%	0.601 → 0.963
21	2,353 → 2,951	11.49 → 11.74	6.88% → 5.86%	27.2% → 17.3%	0.615 → 1.155
22	2,211 → 2,774	12.42 → 12.22	6.33% → 6.92%	23.6% → 22.4%	0.832 → 0.788
23	1,671 → 2,600	10.76 → 13.06	7.48% → 5.31%	20.8% → 18.8%	0.692 → 1.306

3Country

(i) The portfolio by country, 2026-08-08. On the report clock. Meta serves no hour x country grid, so each country cut is raked from its own Meta day onto UTC; our own columns were already there.

194 further countries are not listed, $291.33 between them (39.1% of this block).

Reconciliation, what is measured and what is fitted. Every per-country cell is an allocation across hours; every total is measured. B1 (UTC-7) $164.26 from Meta 2026-08-07 and $580.76 from Meta 2026-08-08, a measured UTC-day total of $745.02.

Absent from the table above, by name: Nomad Node, KBM1, HR1. Their spend is in every day-grain table; only their country split is missing.

(ii) The same cut per account, 2026-08-08.

B1

194 further countries are not listed, $291.33 between them (39.1% of this block).

Nomad Node — no country block: it delivered nothing on this day, a measured zero.

KBM1 — no country block: it delivered nothing on this day, a measured zero.

HR1 — no country block: it delivered nothing on this day, a measured zero.

The prior day, 2026-08-07, the same two tables.

(i) The portfolio by country, 2026-08-07. On the report clock. Meta serves no hour x country grid, so each country cut is raked from its own Meta day onto UTC; our own columns were already there.

187 further countries are not listed, $298.37 between them (39.8% of this block).

Reconciliation, what is measured and what is fitted. Every per-country cell is an allocation across hours; every total is measured. B1 (UTC-7) $201.27 from Meta 2026-08-06 and $548.67 from Meta 2026-08-07, a measured UTC-day total of $749.94.

Absent from the table above, by name: Nomad Node, KBM1, HR1. Their spend is in every day-grain table; only their country split is missing.

(ii) The same cut per account, 2026-08-07.

B1

187 further countries are not listed, $298.37 between them (39.8% of this block).

Nomad Node — no country block: it delivered nothing on this day, a measured zero.

KBM1 — no country block: it delivered nothing on this day, a measured zero.

HR1 — no country block: it delivered nothing on this day, a measured zero.

4Aggregate, both days

Pooled over the 24 hours both days delivered in, on the UTC clock. Meta's instrument on both sides of every ratio.

	spend	impressions	CPM	clicks	CTR	installs	click→install	CPI
2026-08-07	$749.94	81,142	9.24	5,016	6.18%	998	19.90%	$0.751
2026-08-08	$745.02	75,033	9.93	4,202	5.60%	814	19.37%	$0.915

• click → install 19.90% → 19.37% (−2.6%, p = 0.528) — did not separate.
• CTR 6.18% → 5.60% (−9.4%, p = 0.000) — fell.

Pooling across hours weights each hour by what it delivered. Section 2's pairing is the unweighted read of the same two days; here they agree.

Mix-neutral account line. Account CPM $9.13 → $9.93 (+8.7%), and $9.93 held to 08-07's campaign mix: the whole move is inside the worldwide buy.

The three measures side by side:

measure	2026-08-07	2026-08-08	moves after the day closes?
booked	$1,505.16 (2.01) (Meta 2.06)	$1,695.44 (2.28) (Meta 2.21)	settled
within-day	$436.98 (0.58)	$996.74 (1.34)	never
cohort @ 24h	$472.45 (0.63), 100% covered	$1,163.66 (1.56), 100% covered	frozen, both days

08-07's cohort figure has matured and barely moved. The 08-07 report published $468.89 at 51% coverage; fully covered it is $472.45, +0.8%.

Booked revenue by the UTC day its payer installed. The top row is the within-day figure; the gap to booked is the inherited cohort.

installed	booked on 08-08	share
2026-08-08	$996.74	58.8%
2026-08-07	$70.44	4.2%
2026-08-06	$94.57	5.6%
2026-08-05	$88.21	5.2%
2026-08-04	$57.87	3.4%
2026-08-03	$43.22	2.5%
older than 08-03	$344.39	20.3%

Today's own cohort is 58.8% of booked revenue against 29.0%. 08-07 lived off inherited payers; 08-08 earned most of its booked figure inside the day.

Eight days, because two days cannot show whether the account is regressing

Every row from measures.py on one config and one payment cut, UTC clock. Within-day is sealed on every day here and is the column to read across them; cohort@24h is complete except on 08-08.

⚠ 08-01 and 08-02 are a different campaign (_260730, the retired earner) and 08-02 is the billing-stop day. The like-for-like window is 08-03 onward.

Delivery has regressed and not recovered. The same spend bought 1,963 installs on 08-01 and 814 on 08-08, cost per install up 57% since 08-03.

Return on spend has not regressed. Within-day ROAS oscillates between 0.58 and 1.42 across the eight days with no direction; 08-08's 1.34 is also 08-01's.

Revenue per install rose in step with cost per install. $0.500 on 08-01 → $1.225 on 08-08, CPI $0.374 → $0.915: both up 145%.

Two thirds of the cost rise is inside countries, a third is the mix moving up-market. Each day's own per-country prices, held to 08-03's mix:

The India share of spend has halved and the US share tripled across the same window, 31.5% → 18.3% and 5.3% → 16.7% on the Meta clock. That is the mix term, and it is real and large. But the larger term is the other one: held to a fixed mix the account still pays +$0.220 more per install than on 08-03, and that component only turned positive on 08-07 and then doubled. The up-market shift explains the smaller share, and it is also the part that pays for itself in revenue per install; the within-country price rise is not.

Watch the join between the two. If within-country cost keeps rising while revenue per install stops, ROAS falls on about two days' warning.

5The campaigns, side by side

Delivery is Meta's instrument on both sides of every ratio; every revenue measure is ours. Cohort horizon H=24h. Report clock UTC. Rank on CPM, cost per click and CPI only — geography is the treatment here, which is what makes those comparable across campaigns, while revenue still needs 16 payers (DECISION_LOG.md #37, #48) and a campaign under that bar carries a direction and never a magnitude.

campaign	spend	share	impressions	impr share	CPM	clicks	CTR	cost/click	installs	CPI	within-day	wROAS	payers
Worldwide	$745.02	100.0%	75,033	100.0%	9.93	4,202	5.60%	$0.177	813	$0.916	$993.39	1.33	50

Booking with no delivery, and excluded from every rate above: German (DE/AT) 0 installs (Meta), 0 registrations (ours), $0.00 within-day; Earner (retired 08-03) 1 install (Meta), 17 registrations (ours), $3.35 within-day. Counting them would put installs into the CPI denominator against spend that never happened.

The three campaign charts are not drawn. Spend share, CPM and CPI by campaign each need two delivering campaigns; the German buy spent nothing.

6The worldwide campaign

bailingxia_meituan_ww_cvr_260802, the whole of the day's spend.

Structure on this day

Six ad sets, five delivering. ..._d100_mugshot has been off since 08-04 12:32 PT and took no spend. The ..._d100_spill set holds two ads, vid_ww_spill_8s_9x16_v1 and its Copy, a replacement pair: only the Copy delivered on this day. So one delivering ad per ad set on this day, which makes the ad-set row the ad row for the daily cut, and the rows below are labelled by asset. That identity holds for this day's cut alone; the object model does not carry it.

All five delivering arms carried $150/day throughout, unchanged since 08-06 04:06 PT. They are budget-matched by construction, which is what lets a cross-arm comparison inside this stretch avoid DECISION_LOG.md #47.

A. Per asset — delivery

Meta's instrument on both sides of every ratio. Our registrations never enter this table.

asset	spend	impressions	CPM	clicks	CTR	installs	click→install	CPI
E2/bodysuit	$162.42	17,970	9.04	979	5.45%	210	21.45%	$0.773
E2/bikini	$158.42	12,874	12.31	750	5.83%	149	19.87%	$1.063
E2/pool5	$155.92	21,276	7.33	1,054	4.95%	171	16.22%	$0.912
E2/spill2	$148.42	12,234	12.13	785	6.42%	173	22.04%	$0.858
E2/lounge	$119.84	10,679	11.22	634	5.94%	110	17.35%	$1.089
E2/mugshot	$0.00	0	—	0	—	0	—	—

B. Per asset — the three revenue measures

Ours on every column. Meta can express only booked, so it rides bracketed beside that column and never as a row. Payers are distinct people, never transactions. Cohort horizon H=24h, 100% covered on this day. Report clock UTC.

asset	booked	bROAS (Meta)	within-day	wROAS	cohort@H	cROAS	payers	payer%	ARPU	ARPPU	largest payer
E2/bodysuit	$412.36	2.54 (3.39)	$341.10	2.10	$375.06	2.31	15	6.4%	1.458	22.74	41.0%
E2/bikini	$116.51	0.74 (1.34)	$102.16	0.64	$117.23	0.74	3	1.7%	0.574	34.05	53.8%
E2/pool5	$480.09	3.08 (2.89)	$357.78	2.29	$407.01	2.61	17	9.2%	1.944	21.05	16.3%
E2/spill2	$114.75	0.77 (0.65)	$106.76	0.72	$171.86	1.16	7	3.9%	0.600	15.25	36.3%
E2/lounge	$132.55	1.11 (1.97)	$85.60	0.71	$89.15	0.74	8	7.3%	0.778	10.70	33.6%
E2/mugshot	$0.00	— (—)	$0.00	—	$0.00	—	0	0.0%	0.000	—	—

The 16-payer bar (#37): any row above whose payer count is under 16 is a direction and never a magnitude. Only pool5 clears it on this single day.

C. Per asset — country mix

⚠ This table is cut on the META day, while every other table on this page is on the UTC clock. Meta serves no hour × country grid, so per-country delivery exists only on the ad account's own UTC-7 day and cannot be restitched. It is a seven-hour-offset window against tables A and B above; read it for mix, never for level against them.

The last column is the share of that asset's own within-day revenue that came from the United States: where it far exceeds the US spend share, the return is a property of the geography the asset bought.

asset	spend	India %	US %	other %	US % of its revenue
E2/bodysuit	$161.74	22.8%	16.5%	60.7%	69.2%
E2/bikini	$177.37	20.3%	16.4%	63.3%	53.8%
E2/pool5	$149.30	17.8%	17.1%	65.1%	14.8%
E2/spill2	$151.81	14.7%	18.1%	67.2%	4.7%
E2/lounge	$126.06	14.8%	15.4%	69.8%	0.0%

Table C changes how table B reads. bodysuit and bikini took most of their within-day revenue from the United States; pool5 did not.

The split test, on its three complete days

The test E2_ww_creative_5cell_260806 runs 08-06 00:00 → 08-11 00:00 PT. Its pre-registered primary is payers per install at a 65% confidence bar, pooled over the run. Everything below is on our own registrations and our own payers, matured to 24 h at the payment cut, so numerator and denominator share one instrument.

Pooled over the three days, one arm separates and it is the loser.

Four of the ten pairs clear the 65% bar decisively and all four are the same comparison: lounge (95.7%), pool5 (94.6%), bodysuit (93.2%) and spill (88.4%) each beat bikini. Among the top four, only lounge over spill (70.9%) and pool5 over spill (65.1%) reach the bar at all, and a 65% bar means roughly one call in three is wrong. The honest read is that bikini is worse than the other four and the other four are not separated from each other.

Four arms now clear the 16-payer bar. bodysuit 29, pool5 26, spill 21, lounge 21, bikini 15: most carry a magnitude for the first time.

The ranking did not replicate, and that is the day's finding.

Rank order, best first: 08-06 spill > lounge > pool5 > bikini > bodysuit; 08-07 spill > lounge > bodysuit > pool5 > bikini; 08-08 pool5 > lounge > bodysuit > spill > bikini. Spearman on the rate itself is +0.70 between 08-06 and 08-07, +0.10 between 08-06 and 08-08, and +0.10 between 08-07 and 08-08. spill went first, first, fourth. pool5 went third, fourth, first. Only bikini holds a stable position, at the bottom on all three days.

Held against the first two days pooled, two arms moved on day three and three did not: pool5 3.05% → 9.32% (one-sided p = 0.001) and bodysuit 3.52% → 7.14% (p = 0.023); bikini, spill and lounge are unchanged within noise. Both movers moved on the day the whole campaign's payer rate moved, from 3.84% to 5.99%, which is why the report treats the day effect as the thing that happened and the arm differences on it as riding on top.

This amends DECISION_LOG.md #57. #57 recorded Spearman +0.90 between cost per install and payers per install over two days; over three it is +0.10:

The two metrics are not inverted; they are unrelated. The operational half of #57 survives untouched and for a better reason. Cost per install carries no information about which arm pays, so it cannot be used to cull one. But the stated mechanism, that the dearest installs are the best payers, does not hold on three days and should not be quoted again.

Cost per install replicated: bodysuit < pool5 < bikini ≈ spill < lounge on all three days, nine of ten pairs clearing 65%, eight 94%.

The co-primary is not computable and closes as unanswerable by construction. §7a wanted install→registration as a funnel stage; ours exceed Meta's on every arm.

The quantity it actually is can be tested. Under a Poisson model, R ~ Binomial(R+M, p) with p = R/(R+M):

Homogeneity across the five arms: χ² = 1.81 on 4 df, p = 0.771. The arms do not differ.

A validity check on every other endpoint. The 6.4% excess is an instrument gap common to all five arms, so no Meta-denominator figure is biased.

How loose the 65% bar is: these five arms give six of ten pairs "clearing" 65% on data whose omnibus test finds nothing.

The CTR fall has a third in-test point now

The 08-07 report left this open because two points cannot separate a settling level from a trend. Pooled CTR over matched hours, one instrument across all four days:

Each step is smaller than the one before it, by 21% then 42%; the asymptote is near 4%, 1.4 points below where the day ended.

Fatigue is ruled out on this step. Reach/frequency over the four days: 46,284/1.23, 44,191/1.58, 49,549/1.53, 50,814/1.50 — reach up 2.6%, frequency down, CTR down 0.48.

7Caveats

• 08-08's cohort@24h is frozen at 100% coverage. The day's last install passed its 24-hour horizon at 23:59 UTC on 08-09, and the figure has not moved from the $1,163.66 published here: the 2026-08-09 report carries the same number at full coverage. Against 08-07's completed $472.45 the day reads +146.3%. This page shipped on the 2026-08-09 21:02 UTC cut, which had reached 89% of the day's installs and declined that comparison.
• 47 payers carry the whole revenue story, and one of them is 14.8% of it. The largest single payer contributed $139.97 of the campaign's $948.58 of matured within-24h revenue on our instrument, and the top three are 31.8%. On 08-07 the same figures were 9.2% and 26.7%. A doubling of within-day ROAS built on that base is a direction.
• Every per-arm payer row on the single day is under 16 except pool5. The three-day pooled rows in section 6 are the readable ones; the single-day column in table B is not.
• The 65% bar is the test's own and it is loose. At 65% confidence roughly one call in three is wrong. The four bikini comparisons sit at 88–96% and are the only ones this report treats as findings. Run pairwise on the ours ÷ Meta ratios, the same bar "clears" bikini over lounge at 89.6% and over spill at 81.7%, on data whose omnibus test finds nothing at p = 0.771.
• Table C and section 3 are META-day cuts and sit seven hours offset from every other table.
• The regime table had been stale for four days. REGIMES_260802 ended at 08-05 00:00 PT, so every cell from 08-05 onward carried regime = None and the estimator's regime-matched views were empty across the whole split test. It was re-read from the activity history for this report and extended through the test's end. The 08-05, 08-06 and 08-07 reports were written against the stale table; nothing in them depended on a regime-matched cut, which was luck.
• spill is measured on the Copy. The delivering ad in the spill set is vid_ww_spill_8s_9x16_v1 - Copy; the original has not delivered since 08-03. They are one arm over time, which is why they are not split.
• The within-day ROAS series is noisy at ±0.4, which two adjacent days cannot resolve.
• The country cost decomposition runs on the ~95% of spend whose cells are present on both days, and it is not sensitive to small ones: repeating it with a floor of 5 and of 20 installs per country moves the mix share only between 28% and 40%, because per-country cost is estimated on installs and the weighting is by spend.
• The one cost-per-install pair the test did not separate is spill against bikini, at 53.9% on $0.829 and $0.834.

8What would settle the open questions

• Whether the payer-rate ranking has any stable component at all. Three days give three orderings with a stable bottom and an unstable top. The test's last two days (08-09, 08-10) will double the evidence on the top four, and the pre-registered read-out pools the whole run. If the top four are still unseparated at the end, the answer is that these four assets do not differ on the primary, which is itself a result and a usable one.
• What happened on 08-08 inside countries. The mix is excluded and the effect is large, so the candidates are a product, pricing or store-side change on our side, or a change in who the auction delivered to that country mix does not capture. Only the first is checkable by us, and it is a question for the product side.
• Where the CTR decay settles. A fourth and fifth in-test point extend the same series with no new method. The asymptote estimate here rests on three differences and should be re-fitted before it is quoted.
• The co-primary needs a denominator on one clock, unchanged.

9What this hands to the next report

• The read-out at the 08-11 horizon, pooling the whole run on the pre-registered primary. On the three-day read the decision rule that fires is not the one the 08-07 report expected: spill does not beat the same-shot arms, so §7a's "copy the competitor's anatomy" branch is not currently supported by the primary. bikini is the only arm the test has separated, and cutting it is the one action the evidence carries.
• Cost per install as the reproducible endpoint. It is the test's secondary and it replicated across three days at high confidence while the primary did not. That is worth a pre-registration in its own right on the next test.
• 08-08's cohort@24h will be complete by the next report and should be restated there at full coverage, as 08-07's was here.
• The 08-07 report's section 2 carried no 24-row own-day table, which this one restores per daily_report_format.md. The series is inconsistent on that one element across those two days.

10Reproduce

Run directory _runs/2026-08-09_0808_daily/. Report clock UTC; the ad account is fixed at UTC-7, so a UTC day is Meta day 08-07 17:00–23:59 plus Meta day 08-08 00:00–16:59.

Exports, all from AdsPower Profile 23 (k1elevon), B1's bound environment, pulled 2026-08-09 14:11–14:14 PT:

file	breakdown	window
exports/ads_hourly_aug8_settled.csv	Time of day	Meta day 08-08, 24 buckets, 130 rows
exports/ads_day_jul1_aug9.csv	Day	Jul 1 – Aug 9
exports/ads_country_day_jul1_aug9.csv	Day × Country	Jul 1 – Aug 9

The hourly file was pulled twice, 7.8 h apart, as a settling check: spend +0.17%, impressions +0.11%, and clicks, leads, purchases and conversion value all 0.00%. Meta files a conversion on the hour it converted, so a past day's revenue does not grow.

Attribution comes from a hand-taken Telegram export of the LumUser and LumPayment channels, driven on the laptop, parsed with ad_ops/lum_notifications/parse_export.py at --tz America/Los_Angeles: 17,716 registrations, 1,988 payments, 14 payments with no matching registration (0.7%). The three clock checks all pass. The parser corrected 56,303 timestamps for the Telegram DST bug; registered_at agrees with the corrected message time to a median of +0.1 s; and correlating our hourly registrations against Meta's hourly Leads wins at lag 0 (r = 0.877 against 0.587 and 0.531).

The instrument gate was run on the META clock, where it is valid: our records read 0.99 and 1.04 of Meta's booked conversion value on the two days, inside the documented 0.91–1.06 band. On the UTC clock the same comparison is cross-clock and measures.py refuses it by name.

python -m ad_ops.measures      --config _runs/2026-08-09_0808_daily/run_config.json --day 2026-08-08 --prev 2026-08-07
python -m ad_ops.day_compare   --config _runs/2026-08-09_0808_daily/run_config.json --today
2026-08-08 --prev 2026-08-07 --h 6
python -m ad_ops.asset_estimator hourly --lum _runs/2026-08-09_0808_daily/lum \
    --meta <the four hourly exports 08-05..08-08> --fx _runs/2026-08-02_today_read/fx_rates.json \
    --ads ww_bikini_10s,ww_bodysuit_10s,ww_pool5,ww_mugshot,ww_spill,ww_lounge \
    --campaign bailingxia_meituan_ww_cvr_260802 --horizon 24
python -m ad_ops.cohort_arpu   --config _runs/2026-08-09_0808_daily/run_config.json --days 2026-08-04,…,2026-08-08
python -m ad_ops.report_tables --config _runs/2026-08-09_0808_daily/run_config.json --day 2026-08-08
--prev 2026-08-07 [--campaign-table | --campaign ww]
python -m ad_ops.report_charts --config _runs/2026-08-09_0808_daily/run_config.json --day 2026-08-08
--prev 2026-08-07 \
    --days 2026-08-05,2026-08-06,2026-08-07,2026-08-08 --out ad_ops/figures/2026-08-08

Two probes in the run directory carry the analyses the instruments do not: e2_65pct.py (the test's own endpoints at its 65% bar, copied from the 08-07 run and repointed to the third day) and payer_jump.py (the mix-neutral decomposition of the day's payer-rate and revenue-per-install move).

REGIMES_260802 in ad_ops/asset_estimator/cohorts.py was extended for this report from manage/campaigns/history, range string asserted "Aug 4, 2026 – Aug 9, 2026".

11Appendix — definitions

The report clock. A report day is a UTC day, because Adjust reports in UTC and a UTC-bucketed report lines up with the attribution dashboard without anyone converting in their head. Meta's ad account is fixed at UTC-7, so each UTC day is stitched from two Meta days and a UTC pull needs the previous Meta day's hourly file as well as the day's own.

Three things the clock does not reach, each labelled where it appears: the country tables cannot follow it, because Meta serves no hour × country grid, so they stay a META-day cut; the instrument check is invalid off the META clock, because Meta's conversion value sits on its conversion hour and cannot be honestly restitched; and the estimator runs on Meta's clock deliberately, because its subject is Meta-side delivery windows.

The three revenue measures, never one, never mixed, always labelled:

measure	counts	moves after the day closes?
booked	revenue that arrived inside the day, whatever day its payer installed. Meta's own number is this basis	settles ~4 h after midnight, then fixed
within-day	installed that day and paid before that day closed	never. Sealed at midnight
cohort at H	that day's installs, counted within H hours of each person's own install. House horizon H=24	rises until every install has lived H hours, then frozen. Always state coverage

All three are read on our own payment records. Meta can express only booked, so it appears bracketed beside that column and never as a row of its own: that is one measure on two instruments, not two measures.

Two install counts exist and are never crossed. Meta's Leads and our own registrations disagree by 5–10%. Meta's is used for the delivery funnel (CPI and click→install), so numerator and denominator are both its. Ours is used for payer rate and every revenue measure, so the payers and the installs they came from are the same population.

A regime is a stretch over which an ad set's daily budget did not change. It belongs in a daily report because the budget is a geography dial on this account: raising it buys wider, cheaper countries, so two creatives measured across a budget change are compared on different audiences (DECISION_LOG.md #47). All of 08-08 sits inside one regime, E2-R6, which began 08-06 04:06 PT when the last two arms were levelled to $150/day and runs to the test's end. Every delivering arm carries the same budget throughout it.

The 16-payer bar (DECISION_LOG.md #37): below roughly 16 payers a revenue figure is a direction and never a magnitude. Coverage, payer count and the denominator's instrument are the three columns to read before the one you came for.