Skip to main content
A buyer, or an agent working for one, should be able to answer “why did my cost per result go up?” from Apostra’s data alone. That takes three views of the same campaign:
  1. Cost per result against the buyer’s target, day by day. For example the cost per purchase, as one-, seven- and thirty-day lines.
  2. What moved. Cost per thousand impressions, click-through rate, conversion rate and frequency, each compared with the 28 days before the change.
  3. Which ads drove it. Each ad’s cost per result and click-through rate against that ad’s own earlier results.
This page lists what a seller’s reporting has to carry for those views. It is for the developer of a sales agent, for the people who build an adapter to an ad platform, and for the agents that read the result. Each requirement has an id, WNR-1 to WNR-13, so a finding, a conversation and a skill can name the same thing.

What this page does and does not do

  • Everything here is asked, not enforced. Apostra judges each requirement as a report-only finding. A finding shows on the agent’s page and stops nothing: it does not refuse a buy and does not affect certification. No requirement on this page has a blocking date. If one gets a date, Apostra tells sellers first, and the date appears on this page and on What is certification?.
  • It builds on the package, creative and country rows Apostra already asks for. Apostra asks every seller to report delivery by package, by creative and by country, each set adding up to the package figure for the day. Region and metro rows are highly recommended, and a campaign can require them. Country, region and metro are not repeated below. This page adds what a cost-per-result diagnosis also needs: the day grain, conversions by event type, daily reach and frequency, and revisions.
  • It uses AdCP’s shapes. Every field named below is AdCP’s own, from get_media_buy_delivery and reporting_capabilities. Where AdCP has no field for something yet, the requirement says so and says what to send meanwhile.
  • Buyers do not see all of this yet. Per-ad delivery for Meta and TikTok campaigns is live today (Per-ad delivery). For other sellers, buyers cannot yet read creative rows, conversions by event type or daily reach and frequency. Apostra asks for them now so that reporting can show them when it does, and so a missing piece is visible to you before it is visible to a buyer.

The requirements at a glance

“No” means no finding judges it yet. Each one becomes a report-only finding, and this table changes when it does.

Granularity

WNR-1: package rows by day

Report one row for every day a media buy ran, for every package, in by_package[].daily_breakdown, with the date in the timezone you declare (see WNR-9). Why buyers care. Cost per result over time is the first view a buyer opens. Totals for the whole flight cannot show a change in the last week. AdCP shape. by_package[].daily_breakdown, one item per day. How it is checked. Not yet. Report-only finding to come.

WNR-2: creative or ad rows by day

Report one row for each creative, or each ad where the platform has ads, in each package, for each day. Use the creative id that Apostra’s creative assignment used. Where the platform made the ad and Apostra does not know it, use a stable id of your own, the same one on every report. Why buyers care. Which ad drove a change is the third view, and it is the one a package row cannot answer. AdCP shape. by_package[].by_creative[], each row with creative_id and the delivery metrics of WNR-5 to WNR-8. A buyer asks an agent for creative rows with reporting_dimensions.creative on get_media_buy_delivery. Apostra’s own reads of your agent do not send it yet. Where AdCP cannot carry it yet. AdCP has no daily row on a creative: a by_creative row covers the whole reporting period, and daily_breakdown exists only on a media buy and a package. Until it does, a report for a single day carries creative rows that each cover exactly that day. Apostra’s own per-ad delivery returns a daily_breakdown on each creative row as an extension and says so (Per-ad delivery). Apostra’s own adapters. Google, LinkedIn, Meta, OpenAI Ads, Pinterest, Reddit, Snap, Spotify and TikTok send these rows today. Each sends one row for each Apostra creative that ran in the window, and one for each ad Apostra did not make, under its ad set, ad group or line item, with an explicit unknown row for delivery no ad row accounts for. Amazon does not send them yet. Meta rate limits Graph reads, so its rows are an extra read that is skipped while an account is near its limit. When a platform’s per-ad figures add up to more than the package, or cannot be read, the creative rows are left out and the package totals stand. A buyer reads them with get_delivery and report: "live_campaign_delivery" for a campaign connected to one of these platforms; the stored campaign_delivery report does not carry them yet. How it is checked. In part. A report that carries creative rows is judged for WNR-3 and WNR-4, by the delivery by creative finding. That a seller sends creative rows at all, and for each day, is not checked yet.

WNR-3: a creative’s row is its own delivery

Each creative row reports what your platform counted for the ads that ran that creative. Never fill the rows by dividing a package or campaign total among its creatives. If your platform cannot report delivery by creative, leave by_creative out. AdCP makes it optional. Why buyers care. Rows made by dividing a total make every creative in the package look the same, and nothing in the rows says they were divided. How it is checked. Yes, as a report-only finding, when a report carries creative rows. This is the practice Report each creative’s own delivery on the agent requirements page. The delivery by creative finding flags a package whose creative rows are equal shares of its total.

WNR-4: rows add up

For each package and day, the creative rows add up to the package figure. Delivery you cannot attribute to a creative, or hold back for privacy, goes in an explicit unknown row so the set still adds up. The same holds for conversion event types (WNR-6). Why buyers care. A view that does not add up to the package figure cannot be trusted for either number. How it is checked. Yes, as a report-only finding, when a report carries creative rows: the delivery by creative finding flags a package whose creative rows differ from it by more than rounding in impressions or spend. The same split of conversion event types is not checked yet.

Metrics, on package rows and creative rows

WNR-5: spend, impressions and clicks

Report spend, impressions and clicks for every row. spend is in the account’s real currency, one currency for each package and day. State which clicks you count: every click on the ad, or only clicks that leave the platform for the advertiser’s site. Why buyers care. Click-through rate and cost per click are only comparable between sellers that count the same thing. The same ad has a different click-through rate when it counts every click than when it counts only clicks that leave the platform. AdCP shape. spend, impressions, clicks, and the currency of the package. Where AdCP cannot carry it yet. AdCP has no field that names the click definition. Say it in your product description and in reporting_capabilities documentation for your agent until it does. How it is checked. Not yet. A report that fails AdCP’s schema is judged by the valid responses finding. The click definition, and one currency for each package and day, are not checked yet.

WNR-6: conversions by event type

Report conversions and conversion_value, and split them by event type in by_event_type: each entry has an event_type from AdCP’s list, its count, and its value where you have one. When you give both, the entries’ counts add up to conversions. Why buyers care. A cost per purchase that counts sign-ups and page views is wrong by a different amount every day. With one conversions figure and no event type, Apostra withholds its verdict on a purchase goal. AdCP shape. conversions, conversion_value and by_event_type[] with event_type, event_source_id, count and value. Where AdCP cannot carry it yet. AdCP puts by_event_type on totals and on packages and creative rows, but not on a daily_breakdown item. Send a day’s split on a report for that single day (see WNR-2). How it is checked. Not yet. Report-only finding to come: where conversions is above zero, by_event_type is present and adds up.

WNR-7: daily reach

Report reach for each day, with reach_unit (what is counted: individuals, households, devices, accounts, cookies or custom) and a one-day reach_window. Reach over a longer period is its own figure. Never send it as the sum of days. Why buyers care. Reach cannot be added up. A day’s reach next to the period’s reach is how a buyer sees whether new people are still being found. AdCP shape. reach, reach_unit and reach_window ({ "kind": "period", "period": { "interval": 1, "unit": "days" } } for a day). Where AdCP cannot carry it yet. AdCP defines reach on totals and on packages, not on a daily_breakdown item. Send it on a one-day report. How it is checked. Not yet. Report-only finding to come, for a seller that declares reach.

WNR-8: daily frequency

Report frequency for the same day and window as the reach: impressions for the window divided by its reach. Do not average frequency across days or rows. Why buyers care. Frequency climbing while click-through falls is how a buyer sees an ad tiring. AdCP shape. frequency, with the reach_window of WNR-7. Where AdCP cannot carry it yet. As for WNR-7. How it is checked. Not yet. It is part of the same finding as WNR-7.

Timing and revisions

WNR-9: the timezone

Declare the source’s timezone in reporting_capabilities.timezone. A day is a calendar day in that timezone, and the dates in daily_breakdown are those days. Why buyers care. A day that is cut at the wrong hour moves delivery from one day to the next, which is exactly the kind of change this diagnosis looks for. How it is checked. Yes. See the reporting timezone matches declaration and reporting configuration echo findings.

WNR-10: freshness

Declare how late a day’s report may arrive in reporting_capabilities.expected_delay_minutes, and how often you report in reporting_capabilities.available_reporting_frequencies, and deliver each day within that. Why buyers care. A buyer reading yesterday’s results at nine in the morning needs to know whether yesterday is final or still arriving. How it is checked. Yes, for the use cases a certification run buys. See “Reports arrive on time” in What is certification?.

WNR-11: revisions

When a figure for a day that has already been reported changes, for example because conversions arrived inside the platform’s attribution window, send the change as a revision. State the attribution window, and set is_final when it has closed and the day will not change again. Why buyers care. Conversions are the most revised number in advertising. A cost per result that quietly changes after a buyer acted on it is a bad experience, and a revision that is marked as one is not. AdCP shape. Reliable Reporting’s immutable revisions and reporting_capabilities.measurement_windows. See Reliable Reporting. Where AdCP cannot carry it yet. A plain get_media_buy_delivery answer has no revision marker, so this needs Reliable Reporting. How it is checked. In part. The Reliable Reporting capability checks that a source can deliver revisions. A finding for a change to a closed day is not built yet.

WNR-12: honest gaps

A report that is pending, delayed or failed says so: answer with an error, or with reporting_delayed. Never answer with zero spend and zero impressions that mean “I could not read it”. A creative or a day with no delivery is a row with zero, and a day you do not have is a missing row. Why buyers care. A zero reads as “the campaign stopped delivering”, and a buyer pauses or rewrites the campaign. The cost of a wrong zero is real money. How it is checked. Not yet. Report-only finding to come. It judges a seller that answers a pending or failed read with zeros.

WNR-13: the declaration matches the data

Declare in reporting_capabilities what you report: supports_creative_breakdown when you send creative rows, and available_metrics with conversions, reach and frequency when you send them. Do not declare what you do not send. Why buyers care. Buyers and their agents choose sellers on what the declaration says. A seller that declares less than it reports is passed over, and one that declares more is found out on the first campaign. Where AdCP cannot carry it yet. reporting_capabilities does not declare which conversion event types you report, whether reach is reported daily or only for the period, or what a click is. Put these in your agent’s description and in the product descriptions until it does. How it is checked. In part. See declarations match behaviour. It does not yet judge the fields on this page.

Where AdCP cannot carry a requirement yet

These are gaps in the protocol, not in your agent. A finding for one of these requirements accepts what the right-hand column names. When AdCP gains a field, this table changes.