Daily campaign report — 2026-08-04, against 2026-08-03

1Overview

08-03's delivery columns are not a fact about the account. A single hour of one arm carrying a broken build took 181,927 of that day's 213,053 impressions, 85.4% of them, at $0.52 CPM. Every pooled impression-side contrast above is mostly that hour leaving, which is why CPM appears to rise sixfold and CTR to more than triple. The install-side columns are less affected and the revenue columns not at all. §2 shows the same two days hour by hour, where that single hour is visible rather than buried, and the two readings do not agree.

Cohort at 24 hours is 35% covered on 08-04 and cannot be read against 08-03's 100%. The payment cut is 10:59 UTC on 08-05, so only installs from the first ten hours of the day have lived a full 24 hours. The figure is still rising. Within-day is sealed on both days and is the honest day-over-day measure, and on it the account went 0.89 → 1.32.

The day gained on basket size, not on how many people paid. Both days closed with exactly 39 distinct within-day payers, on 805 registrations then 824, so the payer rate went 4.8% to 4.7%. Within-day ARPPU went $10.20 to $15.09, and that alone is the whole 48% revenue gain. The gain is also more concentrated: the largest single payer is 15.1% of the day against 10.4% the day before, and the top ten payers hold 67.2% against 56.0%.

Booked revenue on 08-04 splits by the day its payer installed: 63.9% from 08-04's own installs, 10.4% from 08-03, 12.6% from 08-02, 11.1% from 08-01 and 2.1% from 07-31. Inherited cohort is 36.1% of the day's booked money, which is the whole of the gap between booked 2.07 and within-day 1.32.

2Hourly, 2026-08-04

Delivery columns are Meta's on both sides of each ratio. Payer counts and revenue are ours. Booked sits on the payment hour; within-day and cohort sit on the install hour.

Spend is close to flat across the clock, between $10.59 and $28.73, with no hour holding more than 6.4% of the day. CPI runs $0.348 to $1.036 with no clean diurnal shape on this clock: the four cheapest hours are 20, 21, 03 and 05. The revenue columns are not readable at this grain. Within-day payers run 0 to 3 per hour, so a single person moves an hour outright: hour 10 shows within-day ROAS 7.07 on two payers and hour 07 shows 5.11 on one.

The cohort column decays as the day goes on rather than being uniformly partial. Its coverage is 100% through hour 09, 96% at hour 10 and 0% from hour 11 onward, so every cohort figure after hour 10 counts only the payments already in hand and will rise.

3Country

⚠ The spend and install columns are a META-day cut. 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. The revenue columns are the UTC day split by country. The two sides are seven hours apart, so the ROAS columns here straddle two windows: read this table for mix, never for level against §1 or §4. Ordered by spend. Meta's booked value is bracketed.

The United States carries 23.3% of the day's within-day revenue on five people. Its $136.89 comes from five distinct payers, of whom the largest is 65.0% and the top two are 86.9%. Against the 16-payer bar this account uses (#37) the honest statement is that the level is not readable at this sample size, and what the row supports is a direction on where the money sits, not a magnitude.

India is the mirror image and holds the largest payer count of any country on the day: 13 distinct payers behind $136.79, the same share of the day's within-day revenue as the United States, bought at $0.621 an install against the United States at $1.432 on the same cut. Its largest payer is 32.8% of it, so even India sits under the bar. Germany at $2.418 an install is the dearest cell on this cut, 3.9 times India's, and is discussed in §8.

4Aggregate, both days

Pooled over the 19 hours both days delivered in, on the UTC clock. Meta's instrument on both sides of every ratio.

• click → install 13.41% → 23.95% (+78.6%, p = 0.000) — rose.
• CTR 2.65% → 9.19% (+246.9%, p = 0.000) — rose.

Pooling across hours is a mix comparison, and on these two days it is a misleading one. It weights each hour by how much it delivered, so a day that bought differently across the clock can move these rows without any hour changing. §2 counts the same two days hour by hour and finds neither ratio separating at all. Both p-values above are carried by hour 14 of 08-03 and should not be read as a change in how the account converts.

Held to a single creative that ran on both days, on the ad account's own clock, only two tests clear p < 0.05:

• bodysuit click → install fell 36.33% to 25.55%, a 29.7% decline, p = 0.001. This is the one genuine negative in the comparison. Its CPI is flat across the two days, $0.511 to $0.506, so what moved is the share of clicks that became installs and not the price of an install. Coverage at H=6 is 72% on 08-04. bodysuit also went from $100 to $150 per day inside the compared window, so the decline and a budget change share it and cannot be separated here.
• spill click → install rose 2.61% to 26.42%, p = 0.000. This is the rejected build being replaced by its copy, already recorded as DECISION_LOG.md #50. It is not new information and it is the source of the pooled improvement above.

bikini (−10.2%, p = 0.224), pool5 (−11.9%, p = 0.240) and mugshot (−31.4%, p = 0.052) do not separate. No install → pay test on any single creative clears 0.05 on either day. These per-creative tests key on cells the estimator builds on Meta's clock, so their day boundaries are the account's and their totals do not reconcile against the tables above.

Meta's bracketed figures on both rows are quoted on its own conversion clock and are shown so the day is traceable to what Meta will report, not as a second instrument.

Booked revenue on each day, split by the day its payer installed:

The top row of each column is the within-day figure. That identity is why the gap between booked and within-day is exactly the inherited cohort and nothing else.

5Per ad set, hourly

Exactly one ad delivered per ad set on this day, so the ad-set row is the ad row here and the rows are labelled by asset. That identity is a fact about the day and not about the campaign: the ..._d100_spill ad set holds two ads, vid_ww_spill_8s_9x16_v1 and its copy, and they are a replacement pair rather than a pair of arms — the first delivered on the ad account's 08-03 and the second from 08-04, never both at once. The German buy in §8 is the object on this account where the identity genuinely fails, at fifteen creatives in one ad set.

A single-clock delivery cut in six-hour UTC blocks. All columns are Meta's instrument. No revenue column appears here because per-asset per-hour payer counts are 0 to 2 and carry nothing.

spill2 holds the account's highest CTR in the 18–23 block at 15.37%, and the cheapest installs in both blocks where it ran with any volume, $0.341 and $0.370. mugshot runs 12 to 21 CPM against a pack of 9 to 15 and converts 11.6% to 20.0% of clicks against 19% to 34%, which keeps its CPI above every other arm in every block it ran. bodysuit, pool5 and mugshot each buy their cheapest installs in the 18–23 block, the ad account's late morning and afternoon; bikini is cheapest in 00–05 and spill2 in 12–17, so the diurnal pattern is not shared across the pack.

6Per ad set × country

⚠ Spend and installs are a META-day cut, as in §3; revenue is the UTC day. Cells above $5 of spend, ordered by spend. Revenue is within-day.

pool5 still finishes first on every revenue measure, and the cell carrying it holds two people. $93.94 of its $228.50 within-day revenue, 41.1%, is one United States cell, and that cell holds exactly two payers, at $88.95 and $4.99. The larger of the two is 94.7% of the cell and 38.9% of everything the asset earned all day. The same creative bought $15.17 of India on the same window and returned within-day 0.76 there.

That single payer is also 15.1% of the whole account's within-day revenue. Remove him and pool5 falls from within-day 2.21 to 1.35, below bodysuit's untouched 1.57. The ranking between the day's top two arms therefore turns on one person, and that is true whichever country he is in.

7Per ad set, aggregate, all three measures

The three tables below are the standing per-asset block, emitted by report_tables.py and pasted unedited. They carry the same columns in the same order in every report in this series, so any two days can be read against each other. Rows are every ad object with spend or installs on the day, including ones that took no spend and still attracted late-attributed installs.

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. 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, 35% covered on this day. Report clock UTC.

The 16-payer bar (#37): any row above whose payer count is under 16 is a direction and never a magnitude.

C. Per asset — country mix

⚠ This table is a META-day cut, not a UTC one. 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.

Why a row in B is readable or not. 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 asset bought a geography rather than earning a result.

No arm reaches the 16-payer bar on this day. The highest count in table B is pool5 at 12, then bodysuit at 11. Every revenue row above is therefore a direction, and the gap between an arm twelve payers deep and one eleven payers deep is not a result.

Read the three tables together and the day splits cleanly. Table A says spill2 bought the cheapest installs on the account at $0.369 and the highest CTR at 13.34%, and that mugshot converted 14.99% of clicks against a 24–28% pack at roughly twice the price. Table B says pool5 leads the five worldwide arms on all three revenue measures and on both per-person measures, at a 6.2% payer rate and $19.04 ARPPU. Table C says 41.1% of pool5's revenue came from a country holding 13.5% of its spend, a gap of 28 points; bodysuit is the only other arm on the positive side of that gap, at eight points.

A further caution on table B's level: $57.72 of the day's $588.70 within-day revenue, 9.8%, came from two arms that spent nothing at all, on installs attributed late to builds that were switched off. The five worldwide arms that actually bought media returned $505.58 on $434.33 of spend, a within-day ROAS of 1.16 rather than the day's headline 1.32.