Daily campaign report — 2026-08-07, against 2026-08-06

1Overview

Delivery rows are the delivering campaigns summed. The retired earner's booking tail (1 install (Meta), 21 registrations (ours), $9.32 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.

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.

Campaign spend split. Worldwide $676.53 (85.2%) → $737.89 (98.4%); German $117.41 (14.8%) → $12.05 (1.6%), the largest structural change between the two days.

The account bought more impressions for less money and converted them worse. Impressions rose 15.2% on 5.5% less spend, and installs still fell 14.8%.

2Hourly, 2026-08-07

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

Counting which way each shared hour points:

A sign test throws magnitude away and buys weight-free direction with it. CTR ran lower in 23 of the 24 shared hours and CPM in 21 of 24, so neither is a mix artifact. CPI and click→install do not clear 0.05.

3Country

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.

• CTR 7.67% → 6.18% (−19.4%, p = 0.000). Fell.
• click → install 21.69% → 19.90% (−8.3%, p = 0.024). Fell.

Account CPM $11.28 → $9.24, −18.0%; held to 2026-08-06's mix, $10.17 or −9.8%. Half the fall is the German campaign leaving, half cheaper worldwide buying.

The three measures side by side, and the split of booked revenue by the day its payer installed:

Only 29.0% of the day's booked revenue was earned by the day's own installs. Booked rose 7.4% while within-day fell 29.6%, and the whole of that divergence is older cohorts paying.

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.

The German campaign gets no tab, 1.6% of spend and 2 installs against a 2%/20-install gate. Its $12.05 came from Meta 08-06 hours 17–23.

Booking with no delivery, and excluded from every rate above: Earner (retired 08-03) 1 install (Meta), 21 registrations (ours), $9.32 within-day. Meta stamps a conversion on the conversion hour, so a campaign switched off days ago still lands rows on this day.

6The worldwide campaign

One campaign carries 98.4% of the day, so this section is its whole battery.

Structure on this day

Five ad sets, one delivering ad each, all budgets level at $150/day since 04:06 PT on 08-06. The ad set row is the ad row on any single day, so the rows below are labelled by asset.

A. Per asset — delivery

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

B. Per asset — the three revenue measures

Ours on every column. Payers are distinct people, never transactions. Cohort horizon H=24h, 51% covered on this day. Report clock UTC.

The 16-payer bar (#37): every row above is under it, at 4–10 payers. pool5's 2.86 booked is one payment at 60.1% of its revenue.

C. Per asset — country mix

⚠ A META-day cut, seven hours offset from tables A and B. Read it for mix, never for level.

The split test, on its two complete days

The test runs on the ad account's calendar, "Aug 6, 2026, 12:00 AM to Aug 11, 2026, 12:00 AM Pacific Time", so it is read over the hours it actually ran. Day 1 is Meta 08-06 hours 04 to 23 PT, from the first settled hour after the arms went level; day 2 is the whole of Meta 08-07.

The cost-per-install order is identical on both days, Spearman +1.00, while every arm's CPI is higher on day 2 than day 1.

Day 1's spend confound does not replicate, closing test_register.md's open question. CPI against realised spend ran Spearman +0.90 on day 1, −0.60 on day 2.

The split still does not equalise spend. Realised spend runs $105.42 to $167.28 on five equal $150 budgets, a 1.59× range against day 1's 1.43×.

The pre-registered primary is still not readable. Payers per install needs full coverage; 08-07 sits at 51%, 08-06's completed cohort under the 16-payer bar.

7Caveats

The cohort figure below was published at 51% coverage and has since completed. At the full 24-hour horizon 2026-08-07's cohort is $472.45, the figure the 2026-08-08 report carries in its prior-day column; what this page shows is a lower-coverage read of the same cohort, not a different number. Against 08-06's completed $715.22 the day reads −33.9% — the comparison this page declined. Within-day and booked are unaffected; they were sealed and settled when this page shipped.

Cohort@24h is 51% covered on this cut. It is reported and not compared here; Caveats carries the completed figure. The 08-06 column beside it is at 100%, so the two are not the same kind of number and the fall from 0.90 to 0.63 is mostly coverage.

The country and per-asset country tables are META-day cuts and are seven hours offset from every other table here.

The CTR fall is now on its second consecutive day, and the split test is no longer a candidate explanation for it. Both days compared here are inside the test, which is what removes the structural change the 08-06 report could not rule out. Pooled CTR ran 8.73% on 08-05 before the test, 7.24% on the first test day and 6.05% on the second, measured on the account clock. The step from pre-test to test day 1 is the one the split explains; the further step from day 1 to day 2 is not, and it has no explanation in this report.

The split assigns each person to exactly one arm, which does not hold delivery equal.

Frequency is still elevated and is not falling back. Across the five arms it ran 1.23 before the test, 1.59 on day 1 and 1.54 on day 2. The account is still buying repeat exposure against the same partitioned pool.

Turkey's $278.95 booked on $25.03 of spend is one earlier cohort's payment and should not be read as a Turkish result. Its within-day figure is $0.00.

8What would settle the open questions

• Whether the second CTR step is decay or something else: a third in-test day, 08-08, read the same way. Two points cannot distinguish a settling level from a trend.
• Payers per install, the pre-registered primary: the 08-06 cohort is now at 100% coverage and readable, but at 4 to 10 payers per arm it is under the 16-payer bar. It needs the test's remaining days pooled, not one more day.
• Whether the US premium returns: it is absent on this day for the first time in the series. One day with a large foreign booking in it is not evidence that it has gone.

9What this hands to the next report

The test has three days left and 08-08 is the third in-test day, so the day-over-day comparison stays clean. The account is single-campaign, so section 5 will carry one row plus whatever booking tails land. The CPI ranking has now replicated twice and its spend confound is closed, which makes the remaining question about the arms a revenue question, gated on payer counts.

10Reproduce

python -m ad_ops.measures        --config _runs/2026-08-08_0807_daily/run_config.json --day
2026-08-07 --prev 2026-08-06
python -m ad_ops.day_compare     --config … --today 2026-08-07 --prev 2026-08-06 --h 6
python -m ad_ops.cohort_arpu     --config … --days 2026-08-01,…,2026-08-07
python -m ad_ops.asset_estimator hourly --meta <08-05,08-06,08-07 hourly> --lum <run>/lum --fx … \
    --ads bodysuit,pool5,bikini,spill,lounge --campaign bailingxia_meituan_ww_cvr_260802
python -m ad_ops.report_tables   --config … --day 2026-08-07 --prev 2026-08-06 [--campaign-table | --campaign ww]
python -m ad_ops.report_charts   --config … --day 2026-08-07 --prev 2026-08-06 --days 2026-08-04,…,2026-08-07 --out …

Exports pulled 2026-08-08 05:11 to 05:15 PT: ads × hour for Meta day 08-07 (24 buckets, single-day window confirmed on the file), ads × day and ads × country × day for Jul 1 → Aug 8. The hourly and day exports agree at $712.93 for Meta 08-07, and the country export sums to the same figure. Lum Telegram export taken on the laptop 2026-08-08 05:17 to 05:24 PT: 16,268 registrations, 1,745 payments, 10 unmatched, payment data cut 2026-08-08 12:24 UTC. Registrations rose against the previous cut's 15,234 and unmatched sits at 0.6%, which are the tells that the export is not truncated.

11Appendix — definitions

The report clock is UTC. Adjust reports in UTC, so a UTC-bucketed report lines up with the attribution dashboard. Meta's ad account is fixed at UTC-7, so a UTC day is Meta day D-1 17:00–23:59 plus Meta day D 00:00–16:59, which is why every daily needs two hourly exports. Table C's per-asset country mix cannot follow it: the rake is per ACCOUNT, against that account's own hourly margin, and an asset has no hourly margin of its own.

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

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.

A regime is a stretch over which the ad set's daily budget did not change. It matters 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.