Asset Coverage Check Ad Creative

Asset Coverage Check

Shared quality bar: ../references/output-standard.md. All numbers cited here live in ../references/thresholds.md.

A coverage gap is a supply problem, not a performance problem. An ad set that never had enough distinct material cannot produce a readable result, so every verdict downstream inherits the gap.

Use this skill when

Somebody asks how much creative is live, or the pipeline has stalled and new assets are not arriving, or a losing week needs to be explained before any ad is blamed. Also run it before commissioning production, so the brief fills a counted gap rather than adding another version of what already runs.

Its window is now and its subject is the live set: is there enough variety serving this week for the money going out this week to teach anybody anything. The historical question, what arguments this account has actually put in front of buyers across the whole served library including everything paused, belongs to angle-inventory-ad-creative. Run that one when the complaint is that the account has tested a lot and learned nothing; run this one when the complaint is about the spend that is running today.

Required input

• A list of live ads per ad set: ad ID, ad name, launch date, last edit date.
• For each ad, the angle it argues, stated as a sentence.
• For each ad, the format, static or video, and the aspect ratios uploaded.
• Which placements the ad set is eligible for, and whether placements are automatic or manual.
• Which Advantage+ creative enhancements are on for the ad set, image expansion in particular, and the date any of them changed (test.creative_enhancements_rule).
• For search assets, the headline and description counts per ad group.

Analysis workflow

1. Count creatives using the denominator in thresholds.md: one ad ID with its own creative and its own delivery. Two ads sharing an image with different copy count as two. Reject any count supplied as concepts or file names.
2. Group the live creatives by angle, written as a sentence naming the argument for acting. Two entries collapse into one angle under coverage.angle_distinctness when they share their main claim AND their reason for it. This is a test on the argument and not on the words: "save time" and "stop wasting hours" have no word in common and are still one angle. The noun and verb test is a different rule, it belongs to coverage.rsa_headline_floor, and it is about headline slots rather than angles.
3. Compare the angle count per ad set against coverage.angles_per_ad_set, and apply its recalibration rule where the ad set sits under test.volume_floor. Say in the output if you moved the value.
4. Within each angle, count formats and compare against coverage.formats_per_angle. An angle running in one format only cannot separate a format effect from an angle effect, so record that as a limit on every later read of that angle.
5. Check uploaded aspect ratios against coverage.ratio_set, which is two ratios, 4:5 and 9:16. A missing ratio does not remove the placement: with placement asset customisation off, the platform crops, letterboxes or pads whatever you gave it, so the ad still serves to that surface in a frame nobody chose. Name that mechanism by its product name, because the account can switch it off: image expansion is the Advantage+ creative enhancement that generates the vertical crop and the padded background this rule has been describing, and it is covered by test.creative_enhancements_rule. Record whether it is on, and report the ratio gap as a production job either way.
6. Read supply against coverage.stale_spend_share: take each ad set's spend in the window and work out what share of it rode creatives older than the age that key names. The measure is money, not the launch date of the newest file, because an account shipping every week always has a young newest asset while most of its budget still sits on three files from the spring. This is a supply reading only. Do not translate it into a fatigue verdict; that comparison lives under fatigue.* and requires the signals in fatigue.verdict_rule.
7. For search ad groups, count distinct headlines against coverage.rsa_headline_floor, applying the distinctness test in that entry.
8. Write the single production instruction the account most needs: which angle, which format, which ratio, and which ad set it fills.

Decision rules

• Angle count below coverage.angles_per_ad_set and the ad set is spending: approval_needed on production, because the fix costs money outside this pack.
• Format coverage below coverage.formats_per_angle: test on the missing format against the existing one as control.
• A ratio in coverage.ratio_set absent while placements are automatic: rewrite, naming the surfaces now being served from a transformed asset. Not investigate on delivery, because the ad is being delivered; it is being delivered in a frame nobody chose.
• Image expansion on while 9:16 is missing: rewrite, and say the vertical frame currently running was generated rather than designed. Where the account cannot say when it was switched on, test.creative_enhancements_rule also puts every later comparison across that date at low, and the coverage output carries that forward so a fatigue read does not inherit it silently.
• Share of spend on old creatives above coverage.stale_spend_share: monitor, with the date the next asset is due, plus a note that every performance verdict in this ad set is now read on ageing supply.
• Coverage clear on all four counts: say so plainly and name what made it clear. Other skills in this pack read that as a green light.

Output format

No opening verdict line. This skill returns a count, and a verdict line over a list is theatre.

What high means in this skill, since it counts supply rather than reading a test or a time series: the ad list came from an export or from the interface listing every live ad in the ad set, every angle sentence was written from the creative itself rather than from the ad name, the aspect ratios were read off the uploaded assets, and the spend share was computed from the same export. The comparison family's test.min_runtime and test.volume_floor conditions are conditions of a test and are not borrowed to dress a count as a measurement.

Then one sentence naming the single production instruction, then What this could not see, Missing data, Approval gates.

Practical example

Illustrative made-up account data, not a real advertiser.

Input: ad set "Prospecting UK", 6 live ads. Angles, as sentences, reduce to two: "cheaper than hiring an agency" and "set it up in an afternoon". Ads 1, 2 and 3 argue the first; ads 4, 5 and 6 argue the second. All six are static. Ad set spend in the window 6,400, of which 4,350 ran on ads launched before the age named in coverage.stale_spend_share. Headline counts not supplied. Aspect ratios not supplied.

Reading: 6 creatives, 2 angles, which sits below coverage.angles_per_ad_set, and the ad set is above test.volume_floor so the recalibration rule inside that entry does not fire. Formats per angle is 1 against coverage.formats_per_angle, so any weak result on either angle is unverified as an angle effect. Old creatives carried 4,350 of 6,400 spend, which is 68 per cent and above coverage.stale_spend_share, so the supply reading fails on money even though nobody in the account would have called this library old.

Declined: no aspect ratio verdict. The upload list was not supplied, so ratio coverage on both entries in coverage.ratio_set is needs_data and the row is left empty rather than inferred from the placement setting. Saying the account is missing 9:16 because it is a feed account would be an assumption presented as a count. The enhancement state was not supplied either, so whether the vertical frame now serving was generated by image expansion is needs_data under test.creative_enhancements_rule rather than assumed off.

Output: three findings, all export except the ratio row, which is needs_data. Production instruction: one video for "set it up in an afternoon", 9:16, into Prospecting UK.

Guardrails

• Never count concepts, files or ad names as creatives.
• Never convert an ageing library into a fatigue verdict.
• Never infer aspect ratio coverage from placement settings.
• Never state a per-placement cost per creative as fact. On campaigns using dynamic creative or Advantage+ creative asset variations the served asset differs per impression, so reporting.no_creative_level_placement_cost applies and any such figure is an estimate that must be labelled one.
• Do not upload, pause or edit anything. Recommend, and let a human approve.