@comchatx/cx-sponsored-ads-playbook 0.0.0-stage → 0.1.0

package/research/beyond/metrics-and-attribution.md

@@ -0,0 +1,837 @@
+# Metrics and attribution
+
+Read 2026-11-09 by the research seat. Tags: [official] Amazon's own words (Amazon Ads help pages, whose body is served inside the page's own HTML and was decoded from it; Amazon Ads API guides read through their markdown sources at `https://d3a0d0y2hgofx6.cloudfront.net/en-us/<path>.md`; the OpenAPI models on Amazon's CloudFront hosts; release notes; Amazon Ads library and "what's new" pages; SP-API models on `amzn/selling-partner-api-models`), plus the Media Rating Council's letters to Amazon and its accreditation list, tagged [third party: MRC] because they are not Amazon's own words, and the IAB/MRC and IAB/IAB Europe industry guidelines, tagged [third party: IAB/MRC] (Amazon Ads staff are listed among their contributors, but the words are the bodies'); [opinion] a named practitioner, agency or paid tool, with who says it; [research] published research, with what was measured and on what data; [seat] the research seat's own reasoning, allowed here, never a position.
+
+Figures marked **(human readers only)** are dated snapshots. A skill does not carry them as rules; it reads the current value from the page or the API at run time.
+
+## What this covers
+
+**The part of a seller's ad work.** Every number a seller reads about their ads, and the rules that make it. On the traffic side: impressions, gross and invalid impressions, viewable impressions, clicks, CTR, CPC, CPM and vCPM, cost. On the conversion side: purchases (formerly "orders"), units, sales, ROAS and ACOS, the click-based conversion rate and Amazon's new impression-based "purchase rate", the promoted / halo split (formerly "advertised SKU" / "other SKU" / "brand halo"), new-to-brand metrics, detail page views, add-to-cart, branded searches, Brand Store page views, video metrics, long-term sales, and the KDP metrics (KENP read, estimated KENP royalties, Kindle Unlimited qualified borrows). Around them: the attribution rule set (lookback window by ad product and by account type, eligible products, click against view, the last-touch ranking, one credit per conversion), how a conversion is dated (to the day of the ad interaction), when figures settle (restatement, invalid-traffic filtering, cancellations), the time zone and currency of a report's day, the change of every name between Reporting v3 and the Reporting API v1 (Amazon's "unified reporting"), the multi-touch attribution metrics that sit beside last-touch, Amazon Attribution for off-Amazon traffic, Amazon Marketing Cloud as far as a seller's own account reaches, the differences between console, API and bulksheet figures, and the places where Amazon's own pages define a metric backwards. Since the gap fill of 2026-11-09 it also holds the yardsticks a senior lead reads these rules against: the IAB/MRC retail-media measurement guidelines and the MRC's accreditation scope for Amazon (§R, D7a), what makes a long-term-sales forecast a valid long-term effect (F8–F11), the evidence on whether view-attributed purchases are incremental (C8–C9), and the attribution-method literature on why credit is not marginal value (Published research 9–17).
+
+**Why a senior Amazon advertising lead cares.** Every judgement in the other files of this set is a reading of these numbers. Four facts change the reading more than any optimisation does:
+
+1. **The day is the click's day, not the sale's.** A sale reported for 1 October can have happened on any day up to the end of the lookback window. The latest days of any report are incomplete by construction, and they fill in for weeks.
+2. **The same word means different things by surface and by date.** "Sales" for a Sponsored Products seller is a 7-day click-attributed figure that includes other products the seller sells; for a vendor it is a 14-day figure that includes the brand halo; for Sponsored Brands it is 14 days and, under vCPM, includes views. In unified reporting "Purchase rate" is purchases per *impression*, while the console search-term page still calls the per-*click* rate "Purchase rate". Amazon Attribution, alone, dates conversions to the day they happened.
+3. **The rules moved in 2025–2026.** View attribution for vCPM Sponsored Brands and Sponsored Display changed on 2026-01-01; Reporting API v1 went generally available on 2026-09-30 with new names, no window suffixes and halo folded into the base metrics; the console's old report centres shut on 2026-12-31; Reporting v3 stops on 2027-06-30. A year-over-year comparison that crosses these dates compares two measurements.
+4. **Attributed is not caused.** Amazon's metrics are last-touch credit, not incrementality. That question belongs to `areas/ads-organic-and-total.md` and `method/change-measurement.md`; this file states what the numbers are.
+
+**Lane.** Report mechanics (create, poll, retention, rate limits, file formats) are `method/reporting.md`'s; this file names the report types and fields only where a metric's meaning depends on them. Ads against total sales (TACoS, the SP-API Sales and Traffic join, SQP, incrementality evidence) is `areas/ads-organic-and-total.md`'s. Break-even ACOS, margins and targets are `beyond/goals-and-economics.md`'s. Why a number moved is `areas/diagnosis.md`'s; this file supplies the definitional and settling causes it checks first. The Ads API surface itself is `method/ads-api-surface.md`'s. Placement and impression-share use in bidding is `areas/bids-and-placements.md`'s.
+
+## Amazon's own rules and guidance
+
+RULE marks something Amazon enforces or a shape its API, report or attribution system fixes (a definition, a field, a window, a limit, a date). GUIDANCE marks a recommendation. Help pages carry a `lastModified` stamp in their own data; it is given where it dates a rule.
+
+### A. What an attributed conversion is
+
+A1. RULE. Amazon's frame, verbatim: "Attribution measures conversion events that occur after shoppers interact with ads." "Conversions may take up to 12 hours to appear in reports and are reflected on the date of the shopper's ad interaction, which may be different from the conversion date. Attribution metrics for a given report date remain incomplete until the lookback window has ended." The page names the conversion types: purchases ("formerly known as Orders on sponsored ads"; "A purchases can include multiple sold units of product"), new-to-brand purchases, detail page views, branded searches and add to cart. [official] (Ad campaign attribution, https://advertising.amazon.com/help/GX7KDKHMWQYMJ385, lastModified 2026-10-20, read 2026-11-09)
+
+A2. RULE. How a conversion is assigned, in Amazon's three terms: "Product relevance: Campaigns can receive attribution for products beyond those advertised. Advertised products are always eligible, and campaign type determines which additional relevant products may qualify." "Ad interaction type: Some campaign types will only attribute to ad clicks. Others will attribute to clicks and views, depending on the selected cost type." "Lookback window: Maximum number of days that a conversion is considered relevant to an ad click or view." Amazon's worked example: "A 14-day lookback means that a purchase on October 17 can be attributed to ad interactions between October 3 and October 17. An ad interaction on October 2 would be ineligible for attribution." [official] (Ad campaign attribution, above)
+
+A3. RULE. The table by ad product and advertiser type, verbatim in the load-bearing cells:
+
+| Ad product, advertiser | Lookback | Eligible products | Eligible interactions | Model |
+|---|---|---|---|---|
+| Sponsored Products, Seller | 7 days | "Products featured in the ads as well as any other products sold by the Seller Central merchant." | "Clicks only" | "Last click" |
+| Sponsored Products, Vendors and Authors | 14 days | "Products featured in ads as well as any other products from the same brands ("brand halo"). Only products sold by Amazon are attributed. Book products don't have a brand halo." | "Clicks only" | "'Last touch' attribution that prioritizes product relevance and clicks over views. Competes with other Amazon Ads products." |
+| Sponsored Brands and Display, all advertisers ("must be registered brand owners") | 14 days | "Products on the landing page as well as any other products from the same brands ("brand halo"). Attributed purchases may include sales by Amazon or third-party sellers, in New or Used condition." | "Clicks (CPC) or clicks and views (vCPM)" | as the vendor row |
+| Streaming TV, all advertisers | 14 days | products advertised plus the brand halo; Amazon or third-party sellers, New or Used | "Clicks (including QR scans) and views" | as the vendor row |
+| Amazon DSP, all advertisers | 14 days | "Products tracked by the campaign as well as any other products from the same brands" | "Clicks and views" | as the vendor row |
+
+The 7 and 14 are **(human readers only)** as numbers; a skill reads the page. [official] (Ad campaign attribution, above)
+
+A4. RULE. One sale, one credit: "Amazon will attribute a conversion to no more than 1 ad interaction across all campaigns and all advertisers. A sale won't, for example, be attributed to 2 campaigns in both a promoted and brand halo context, or as both a view-through and click-through conversion. It's attributed once to the most relevant ad interaction." The ranking: "Promoted ASIN ad clicks", then "Brand halo ad clicks", then "Promoted ASIN ad views", then "Brand halo ad views", where brand halo here means "Products within the same brand and subcategory as a promoted ASIN" (example: "Category = Electronics", "Subcategory = Headphones, Home Audio, Video Game Consoles & Accessories, etc."). "If the customer interacted with multiple ads, attribution considers each ad's relevance to the product, then uses a last-touch model that prioritizes the last click over the last view. Clicks are prioritized over views because they're a stronger engagement signal." [official] (Ad campaign attribution, above)
+
+A5. RULE. Who competes with whom: "All Amazon Ads products that promote brands (Display, Video, and Audio campaigns, Display, Streaming TV, Sponsored Brands, and Sponsored Products for vendors) compete for attribution." "Sponsored Products for sellers is a merchant-based ad program that doesn't normally compete for attribution with brands. However, if the same advertiser wins attribution as both a merchant and a brand, attribution will choose the more relevant ad, ensuring the same conversion is attributed only once per advertiser." "Seller Central merchants who also use Display, Video, and Audio campaigns should ensure that their merchant token is entered in their Display, Video, and Audio campaigns account configuration." [official] (Ad campaign attribution, above)
+
+A6. RULE. Why attributed sales differ from retail sales, Amazon's own list (verbatim): "In campaign reports, sales are reported on the ad interaction date, which may differ from the actual sale date. Sales reported on a given date may have occurred at any time between that date and the end of the lookback window." "Attributed sales don't reflect any discounts applied in the cart." "Attributed sales don't include tax. Inclusive tax such as VAT is subtracted from sales totals." "Depending on the campaign type, attributed sales may include items sold by you, Amazon, or another seller. They can also be items sold in Used condition at a different price." "Depending on the campaign type, attributed sales may include products that weren't promoted in the campaign but that are from the same brands as products that were promoted." "Payment failures and purchases that are cancelled within 72 hours of order creation will be deducted from attribution metrics. These deductions may take up to 7 days to process after the cancellation." "Ads that feature a specific product may have conversions attributed to them for different products. In reports grouped by "Advertised ASIN," the ASIN is the product featured in the ad. Attributed sales for the ad may include other ASINs." [official] (Ad campaign attribution, above). The 72-hour and 7-day figures are **(human readers only)**.
+
+A7. RULE. Returns are not among the deductions Amazon lists for sponsored ads: the page names payment failures and cancellations within 72 hours, nothing else. Amazon Attribution, by contrast, says its restatements take "into account returns or refunds" (§Q). [official] (Ad campaign attribution; Amazon Attribution API how-to, both read 2026-11-09). Whether a sponsored-ads sale returned on day 20 is ever removed is not stated on any page read (Open questions). [seat]
+
+A8. RULE. The help page's own metric frame for brand advertisers, which is the reason halo is in base metrics: "The totals metrics that you see in campaign management data tables, downloadable reports, and reporting APIs contain conversions for promoted products (tracked ASINs) and conversions for brand halo products (additional ASINs within your promoted brands). We recommend using totals conversions metrics as the best way to measure the full value delivered by your brand's advertising." (GUIDANCE in its second sentence.) [official] (Campaign success metrics, https://advertising.amazon.com/help/G9EMRNUYZPP58HLX, lastModified 2026-08-14, read 2026-11-09)
+
+### B. How a conversion is dated
+
+B1. RULE. Sponsored ads date conversions to the day of the ad interaction (A1, A6). The Reporting API v1 makes this universal: "All conversion metrics are now reported on "traffic date," meaning conversions are reported on the date of the ad interaction rather than the date the conversion occurred." Example: "If a shopper clicked an ad on Oct 1 and purchased on Oct 3, the purchase conversion is now reported on Oct 1". "Previously, DSP reported conversions on the conversion date." [official] (Overview of the Reporting v1 API, https://d3a0d0y2hgofx6.cloudfront.net/en-us/guides/reporting/ads-v1/overview.md, read 2026-11-09)
+
+B2. RULE. The console's version of the same change names a sponsored-ads exception that ended: "Previously, Display, Video, and Audio campaigns and legacy Sponsored Brands campaigns reported conversions on the conversion date." "Tip: You may observe higher conversion volumes earlier in the campaign flight for Display, Video, and Audio campaigns and legacy Sponsored Brands campaigns that previously reported on conversion date." "Both types of reporting reflect the same number of conversions overall, they are just distributed differently across dates." [official] (Understand unified reporting, https://advertising.amazon.com/help/GMH8A8AJSH4ATV6T, lastModified 2026-10-20, read 2026-11-09)
+
+B3. RULE. Two Amazon measures do *not* date to the click:
+- KENP read: "We track clicks associated with your advertised titles, and attribute them to the reading events for the same titles up to 14 days after customers click the ad. The estimated number of pages read are reported to the specific date when readers actually read the pages." "Pages read from offline devices can take 1 week to be reported." [official] (Kindle Edition Normalized Pages (KENP) metrics, https://advertising.amazon.com/help/GSJCXUYD4UGQP2MX, lastModified 2023-03-30, read 2026-11-09)
+- Amazon Attribution (off-Amazon traffic): "In the report, each metric's value reflects the events which took place on the row's date; in other words, events are reported on the date they occurred, not on the date of the attributed click. Attribution is calculated using a 14-day last-click model." [official] (Amazon Attribution OpenAPI model, `ReportRequestBody.metrics`, https://dtrnk0o2zy01c.cloudfront.net/openapi/en-us/dest/AmazonAttribution_prod_3p.json, read 2026-11-09)
+
+The unified-reporting page's statement that "All conversion metrics are now reported on "traffic date"" was read beside these two. Whether KENP in unified reporting has moved to traffic date is not stated (Open questions). [seat]
+
+B4. RULE. AMC exposes both datings as separate tables. "In the `amazon_attributed_events_by_conversion_time` table, the conversion events all happened within the time window the query is run over, while the traffic events the conversions were attributed to may happen up to 14 days before"; "The `amazon_attributed_events_by_traffic_time` table has the opposite behavior … the conversion events may have occurred up to 30 days after the time window", and "the output of a workflow that uses this table may change over time." "Note that these windows are not the same thing as an attribution window, as the attribution window is decided based on the campaign and cannot be changed in AMC." [official] (Amazon attributed events by conversion time table, https://d3a0d0y2hgofx6.cloudfront.net/en-us/guides/amazon-marketing-cloud/datasources/amazon_attributed_events_by_conversion_time.md, read 2026-11-09)
+
+B5. [seat] Consequence a skill can reason from: for any traffic-dated metric, the last `lookback` days of a report are structurally incomplete before any restatement; ROAS, ACOS and conversion rate for those days are biased toward worse (cost is nearly complete, sales are not). The settling schedule is §L.
+
+### C. Clicks, views, viewability and the 2026 view model
+
+C1. RULE. Which campaigns count views: "View attribution is enabled for: All Display, Video, and Audio campaigns … All Streaming TV campaigns … Sponsored Brands and Display campaigns using vCPM billing." "Amazon Ads records view conversions for impressions that meet viewability standards (50% of pixels visible for 1+ second for display ads, or 2+ seconds for video ads). When direct viewability measurement isn't available, viewability is inferred based on historical data and placement characteristics." "Standard conversion metrics such as Purchases and ROAS include the ad interaction types supported by that campaign. … Click-only breakouts are labeled clearly, such as ROAS from clicks or Purchases from clicks. The standard lookback window for views is 14 days." [official] (Ad campaign attribution, above)
+
+C2. RULE. "Campaigns that do not measure view-through attribution (e.g., Sponsored Products or Sponsored Brands CPC) will continue to report 0 for conversions from views." [official] (Overview of the Reporting v1 API, above; the console page says the same for "Sponsored Brands campaigns using cost per click, or CPC, billing", Understand unified reporting, above)
+
+C3. RULE. The view model changed on 2026-01-01. Amazon's launch page: a "shopping-signal enhanced last-touch attribution model" whose metrics (Purchases, Sales, ROAS) become standard reporting; it applies to "Sponsored Brands and Sponsored Display campaigns billed on a viewable impression (vCPM) basis, including branded keyword campaigns" and "Amazon DSP campaigns that serve ads in the Store"; the model gives "credit to ad views that reach shoppers during early discovery moments such as exploratory browsing or general category searches" and "applies a shorter attribution window"; "Conversion reporting for all other campaign types and inventory types remains unchanged. Click-based attribution remains unchanged."; legacy measurement stays available as "conversion metrics that include all ad views for eligible campaigns within a 14-day window" under the name "all views". The page does not give the new window's length. [official] (View attribution updates for Amazon Store ads, https://advertising.amazon.com/resources/whats-new/view-attribution-updates-for-amazon-store-ads, read 2026-11-09). The attribution help page's sentence: "Beginning January 1, 2026, view attribution for some placements will incorporate additional signals, such as shopping behavior patterns, when determining which ad views are eligible for attribution. This may result in changes to attribution trends for campaigns that attribute to views." [official] (Ad campaign attribution, above)
+
+C4. RULE. The "all views" family in v1 has 54 fields, e.g. `metric.allViewsPurchases`, `metric.allViewsSales` (requires `budgetCurrency.value`), each described as "calculated using the original attribution model, which includes all ad views for eligible campaigns. Metrics labeled "all views" are provided for comparison with the updated attribution model. For ongoing performance assessment, we recommend using the standard purchase metrics." Program eligibility lists Sponsored Products among others. [official] (Purchases (all views), https://d3a0d0y2hgofx6.cloudfront.net/en-us/guides/reporting/ads-v1/metrics/amazon-retail-conversions/purchases-all-views.md, read 2026-11-09)
+
+C5. RULE. Viewability fields and who has them. `metric.viewableImpressions`: "The number of times the ad was viewed. This excludes impressions that were deemed invalid by traffic quality filters. An ad is counted as viewed when at least 50% of the ad shows on screen for at least 1 second for display ads and 2 seconds or longer for video ads." Program eligibility: Sponsored Brands, Sponsored Display, Sponsored TV, Amazon DSP, Full-Funnel, **not Sponsored Products**. Same eligibility for `metric.vcpm` ("(cost / viewable impressions) * 1,000"; v3 name "cost per 1,000 viewable impressions (VCPM)"), `metric.viewableDistribution` ("viewable impressions / impressions"; v3 `viewabilityRate`) and `metric.viewableRate` ("vieweable impressions / measurable impressions", Amazon's spelling). `metric.measurableImpressions` and `metric.measurableRate` do list Sponsored Products. [official] (Viewability (Delivery), https://d3a0d0y2hgofx6.cloudfront.net/en-us/guides/reporting/ads-v1/metrics/delivery/viewability.md, read 2026-11-09)
+
+C6. RULE. vCPM in the console: "This metric is available when you use the bidding option to optimize for viewable impressions. … This means you only pay when shoppers actually viewed your ad. In Display, Video, and Audio campaigns, this metric is available when you choose Optimize for reach as your bid optimization strategy. In Sponsored Brands, this metric is available when you choose Grow brand impression share as your campaign goal." "An image is counted as viewable when at least 50% is visible for at least 1 second. A video is counted as viewable when at least 50% is visible while playing for at least 2 seconds. This is the Media Rating Council's (MRC) definition of viewability." [official] (Performance metrics, https://advertising.amazon.com/help/GG44RFW942U9F6F5, lastModified 2026-11-05, read 2026-11-09). Amazon's library guide adds that larger desktop ads need only 30% of pixels in view. [official] (Viewability guide, https://advertising.amazon.com/library/guides/viewability, read 2026-11-09)
+
+C7. RULE. v3's naming collides with v1's: v3 `viewabilityRate` is "View-through rate (vtr). Views divided by impressions"; v3 `viewClickThroughRate` is "Click-through rate for views (clicks divided by views)"; v3 `impressionsViews` and `viewableImpressions` are "Number of impressions that met the Media Ratings Council (MRC) viewability standard." The console page uses "View-through rate (VTR)" for video as "dividing the views by the impressions". [official] (Reporting columns (version 3), https://d3a0d0y2hgofx6.cloudfront.net/en-us/guides/reporting/v3/columns.md; Performance metrics, above; both read 2026-11-09). In v1 the same concept is `metric.viewableDistribution`; "VTR" is not a v1 name. [seat]
+
+C8. RULE (what Amazon says, and does not say, about view credit; gap fill 2026-11-09). Amazon's stated basis for the 2026 view model is its own research, not a published test: "We are introducing a shopping-signal enhanced last-touch attribution model on January 1, 2026, based on our continuous research on shopping journeys in the Amazon store." The page's list of affected campaigns reads "Sponsored Brands , Sponsored Display campaigns billed on a viewable impression (vCPM) basis, including branded keyword campaigns" and "Amazon DSP campaigns that serve ads in the Store", and "Click-based attribution remains unchanged." [official] (View Attribution Updates for Amazon Store ads, datePublished 2026-01-01, dateModified 2026-10-09, read 2026-11-09). Where viewability cannot be measured Amazon infers it: "When direct viewability measurement isn't available, viewability is inferred based on historical data and placement characteristics." [official] (Ad campaign attribution, lastModified 2026-10-20, read 2026-11-09). No Amazon page read publishes an incrementality figure for view-attributed purchases, under the old model ("all views") or the new one; both are attribution rules. [seat, as searched]
+
+C9. [seat] The evidence on whether views cause purchases sits in Published research (items 16–17, with Gordon et al. in items 3 and 5): randomized display experiments outside Amazon found that most of a retailer's measured lift came from shoppers who never clicked (Lewis & Reiley 2014) and that display retargeting raised purchases among exposed users (Johnson, Lewis & Nubbemeyer 2017), so click-only credit can miss real effects; observational exposed-versus-unexposed comparisons missed the experimental lift, mostly by overstating it on Facebook (Gordon et al. 2019, 2023) and once by understating it (Lewis & Reiley's cross-sectional estimator), so view-attributed credit can also count purchases that would have happened anyway. For the choice between vCPM and CPC on Sponsored Brands or Sponsored Display, the reporting consequence is fixed by C1–C3 (vCPM adds view-attributed purchases to the base metrics, CPC reports 0 from views); whether those purchases are incremental is not answered by any Amazon metric a seller can read, and the IAB/IAB Europe guidelines class "baseline vs. exposed analysis" and "platform-reported incrementality" as "Weak" in causal strength (§R7).
+
+### D. Traffic and cost metrics
+
+D1. RULE. Impressions: "The number of times the ad was displayed. This excludes impressions that were deemed invalid by traffic quality filters. Impressions = gross impressions - invalid impressions" (`metric.impressions`, v3 `impressions`). Gross: "The total number of times the ad was displayed. This can include valid as well as invalid impressions." Invalid impression rate: "invalid impressions / gross impressions" (`metric.invalidImpressionRate`). [official] (Impressions (Delivery), https://d3a0d0y2hgofx6.cloudfront.net/en-us/guides/reporting/ads-v1/metrics/delivery/impressions.md; Traffic metrics, https://advertising.amazon.com/help/GMCTGJGY2M438E2R, lastModified 2026-11-02; both read 2026-11-09)
+
+D2. RULE. Clicks: "The number of times the ad was clicked. This excludes clicks that were deemed invalid by traffic quality filters. Clicks = gross clicks - invalid clicks" (`metric.clicks`; v3 `clicks`). `metric.grossClicks` (v3 `grossClickThroughs`), `metric.invalidClicks` (v3 `invalidClickThroughs`), `metric.invalidClickRate` ("invalid clicks / gross clicks"; v3 `invalidClickThroughRate`). `metric.ctr`: "CTR = clicks / impressions" (v3 `clickThroughRate`, "Clicks divided by impressions."). `metric.vctr`: "vCTR = clicks / viewable impressions" (v3 `viewClickThroughRate`). `metric.cpc`: "Cost per click (CPC) = cost / clicks", requires `budgetCurrency.value` (v3 `costPerClick`, "Total cost divided by total number of clicks."). [official] (Clicks (Delivery), https://d3a0d0y2hgofx6.cloudfront.net/en-us/guides/reporting/ads-v1/metrics/delivery/clicks.md; Reporting columns (version 3); both read 2026-11-09)
+
+D3. RULE. Clicks and every traffic metric are incompatible with the converted-product and product-relevance dimensions: "Impressions, clicks, and other traffic metrics are not specific to any 1 conversion. Because a single ad interaction can lead to multiple product conversions, traffic metrics cannot be meaningfully reported at the converted product level and are not available when this dimension is selected." [official] (Data availability in unified reporting, https://advertising.amazon.com/help/G8A5Z6UD9ME5W3GZ, lastModified 2026-10-12, read 2026-11-09; the v1 Clicks page lists "Converted product" and "Product relevance" among its incompatible dimensions)
+
+D4. RULE. Cost. `metric.totalCost`: "The total amount spent on running the campaign. This excludes the third-party fees paid by the agency." v3 Sponsored Ads name `spend` (also `cost`, "Total cost of ad clicks."). "Cost is assessed per click or per 1,000 views depending on cost type." The console renamed the column: "Total Cost (formerly known as Spend on sponsored ads)". `metric.totalCostReconciled`: "The total amount spent on media, fees, and technology costs after reconciliation", with `metric.roasReconciled` ("sales / total cost reconciled"); unified reporting lists "Total cost (reconciled), ROAS (reconciled)" as available only from 8/4/2024. [official] (Costs (Costs and fees), https://d3a0d0y2hgofx6.cloudfront.net/en-us/guides/reporting/ads-v1/metrics/costs-and-fees/costs.md; Performance metrics; Data availability in unified reporting; all read 2026-11-09)
+
+D5. RULE. Cost is provisional for days: "The metrics in the campaign manager are updated every hour. Sales data takes up to 48 hours to populate and might be incomplete in the "Yesterday" view in campaign manager. In addition, consider the Cost (formerly known as Spend on sponsored ads) column to be an estimate as it can be updated until 3 days after the click or viewable impression occurs, while we validate your click traffic." [official] (Performance metrics, above). Figures **(human readers only)**.
+
+D6. RULE. CPM and its siblings: `metric.cpm` "(cost / impressions) * 1000", listed for Sponsored Products too; benchmark quartiles `metric.cpcP25/P50/P75`, `metric.ctrP25/…`, `metric.cpmP25/…` describe "peer brands" and require benchmark dimensions (`benchmarkBrand.id`, `benchmarkBrowseCategory.id`, …). [official] (Impressions (Delivery); Clicks (Delivery); both above). Amazon's help page defines CPC (p50, converted) as "the value where 25% of peer brands fall at or below" — a copy error beside the correct p50 wording for the unconverted field. [official, the page's own inconsistency] (Traffic metrics, above)
+
+D7. RULE. Measurement accreditation is narrow. The Media Rating Council's letter of 2026-07-08 continues accreditation of "Amazon Advertising's first-party measurement and reporting of Sponsored Products ("the Service") Display and Search Clicks inclusive of Sophisticated Invalid Traffic (SIVT) filtration for Amazon placements. This applies to total reporting and does not include any environment (Desktop, Mobile Web, In-app) segmentations." Accredited statistics: "Gross Clicks" and "Clicks (Total Net of SIVT)". Impressions and conversions are not on the list. [third party: MRC] (MRC accreditation letter to Amazon Ads, July 8, 2026, https://www.mediaratingcouncil.org/sites/default/files/Accreditation%20Letters/Amazon/GI070826_Amazon%20Sponsored%20Products%20Accreditation%20Letter.pdf, read 2026-11-09). The console can filter to it: "The MRC accreditation table setting provides the ability to filter sponsored ads user interfaces and reports to only show MRC accredited data." [official] (MRC accredited data for Sponsored Products campaigns, https://advertising.amazon.com/help/GVLUHLCD2Y6C364M, lastModified 2026-02-25, read 2026-11-09)
+
+D7a. RULE. Where Amazon's accreditation stops, across ad products (gap fill 2026-11-09). The MRC's list of accredited digital services names two first-party Amazon services and no others: "Amazon Ads Sponsored Products" ("Clicks: Display Total only", "Clicks: Search Total only", "Sophisticated Invalid Traffic Detection/Filtration Total only") and "Amazon DSP Retail and IMDb, Amazon Publisher Services (APS), Fire TV and Third-party Exchanges (3PX), Alexa Connected Devices" (rendered impressions, clicks and viewable impressions by supply source and environment, "Not all metrics and environments are accredited for every supply source"); a third-party entry, "Integral Ad Science Server to Server (S2S) Integrated Third-party Amazon DSP Reporting", is IAS's service, not Amazon's. Neither "Sponsored Brands" nor "Sponsored Display" appears anywhere on the list. [third party: MRC] (MRC Accredited Services — Digital, https://www.mediaratingcouncil.org/accreditation/digital, read 2026-11-09). The DSP letter of 2026-05-29 continues accreditation of Display "BTR Impressions and Click-throughs (reported Gross and Net of General Invalid Traffic [GIVT])" on Amazon Retail, IMDb, APS and 3PX, and grants initial accreditation, among others, of the "Amazon Retail supply source, for: Video BTR Impressions … and Click-throughs … and Display and Video Viewable impressions (reported Net of General Invalid Traffic [GIVT])"; its invalid-traffic basis is GIVT, where the Sponsored Products letter's is SIVT. [third party: MRC] (MRC accreditation letter to Amazon Ads, Amazon DSP, May 29, 2026, https://www.mediaratingcouncil.org/sites/default/files/Accreditation%20Letters/Amazon/GI052926_Amazon%20DSP%20Accreditation%20Letter.pdf, read 2026-11-09). The Sponsored Products letter's TAG addendum adds that the SIVT accreditation includes "Domain list filtering" and "Data Center IP list filtering". [third party: MRC] (Sponsored Products letter, D7). [seat] Read against the metric families: for Sponsored Products the accredited figures are `metric.grossClicks` and `metric.clicks` in total; `metric.impressions`, `metric.invalidImpressions`, every conversion metric, and every Sponsored Brands and Sponsored Display traffic figure (impressions, viewable impressions, the vCPM denominator, clicks) are outside any MRC accreditation on the list read, so Amazon's invalid-traffic filtering of them (§L) rests on Amazon's own description.
+
+D8. RULE. Impression share. `metric.impressionShare` ("The percentage share of all your impressions compared to all other advertisers"; v3 "Search Term Impression Share"), `metric.impressionShareRank` ("The numeric rank of your account-wide impression share compared to all other advertisers"), `metric.topOfSearchImpressionShare` ("The percentage of impressions that appeared in top search results compared to the number of times they were eligible to appear in top search results"); programs Sponsored Products, Sponsored Brands, Full-Funnel. [official] (Search (Impression share), https://d3a0d0y2hgofx6.cloudfront.net/en-us/guides/reporting/ads-v1/metrics/impression-share/search.md, read 2026-11-09). Console: top-of-search IS "is available for the trailing 90-days"; "For Sponsored Products, this metric gives the top-of-search IS for the first page only." [official] (Performance metrics, above). Unified reporting carries `topOfSearchImpressionShare` at campaign level only, where the legacy reports had it "at the keyword, targeting, and campaign level". [official] (Migrate your scheduled reports to unified reporting, https://advertising.amazon.com/help/GK3JZZE5QLTF8U8A, lastModified 2026-10-19, read 2026-11-09). Bidding use is `areas/bids-and-placements.md`'s.
+
+### E. Conversion metrics: purchases, sales, units, ROAS, ACOS, rates
+
+E1. RULE. The base conversion metrics in unified reporting and v1 carry no window suffix and include halo. The console's statement: "Sponsored Products: Base metrics automatically align with campaign defaults, 7-day lookback for sellers and 14-day lookback for vendors, consistent with the advertising console and downloadable reports. ROAS now corresponds to roas7d (sellers) or roas14d (vendors). Sales now corresponds to 7 Day Total Sales (sellers) or 14 Day Total Sales (vendors)." "Other campaign types: The standard lookback is 14 days, except for Display, Video, and Audio campaigns campaigns using flexible lookback windows". [official] (Understand unified reporting, above). The API's statement, under "Removal of interval suffixes (1d, 7d, 14d, 30d)": "Conversion metrics no longer include a look-back interval in their names. The base metric (e.g., `metric.purchases`, `metric.sales`, `metric.unitsSold`, `metric.roas`) always reflects the correct look-back for the campaign type", with Sponsored Products at "7-day look-back for sellers and 14-day look-back for vendors" and "Other campaign types: The standard look-back is 14 days, except for DSP campaigns using flexible look-back windows". [official] (Overview of the Reporting v1 API, above). The release note of 2026-09-30 lists "Conversion metrics drop interval suffixes (1d, 7d, 14d, 30d)" for the Reporting API v1 and says Stream v2 and the Reporting API v1 "share the same standardized metric names, definitions, and entity terminology". [official] (Amazon Ads advanced tools release notes, https://d3a0d0y2hgofx6.cloudfront.net/en-us/release-notes/index.md, read 2026-11-09)
+
+E2. RULE. No 1-day, 14-day-for-sellers or 30-day Sponsored Products conversion figure exists in v1. Of the 148 fields on v1's Purchases page none carries a day count. [official] (Purchases, https://d3a0d0y2hgofx6.cloudfront.net/en-us/guides/reporting/ads-v1/metrics/amazon-retail-conversions/purchases.md, read 2026-11-09). Reporting v3 still offers them for Sponsored Products: `purchases1d`, `purchases7d`, `purchases14d`, `purchases30d` ("Number of attributed conversion events occurring within N days of an ad click"), `sales1d…30d`, `unitsSoldClicks1d…30d`, and the same-SKU family `purchasesSameSku1d…30d`, `attributedSalesSameSku1d…30d`, `unitsSoldSameSku1d…30d`. [official] (Reporting columns (version 3), above). [seat] A skill that needs a seller's 14-day or 30-day Sponsored Products figure has it only through v3 until v3 stops on 2027-06-30.
+
+E3. RULE. Purchases. v1 `metric.purchases`: "The number of purchases attributed to an ad interaction." Its "Version 3 Reporting Name (Sponsored Ads)": "purchases, orders14d, purchases7d, purchases14d [SP: 7d sellers / 14d vendors]". The console's conversion-metrics table gives "Purchases" the former sponsored-ads names "Orders, 14 day total orders, 7 day total orders". A purchase is an order event, not a unit: "A purchases can include multiple sold units of product" (Ad campaign attribution). Units: `metric.unitsSold` ("The number of product units in purchases attributed to an ad interaction. A single purchase event can include multiple units of the same product."). [official] (Purchases (v1); Conversion metrics, https://advertising.amazon.com/help/G7FMJZAJLTR3DKK2, lastModified 2026-10-12; Ad campaign attribution; all read 2026-11-09)
+
+E4. RULE. Sales. v1 `metric.sales`: "The sales from purchases attributed to an ad interaction.", requires `budgetCurrency.value`; its v3 names are "sales, sales7d, sales14d [SP: 7d sellers / 14d vendors]". The console's conversion-metrics table gives "Sales" the former sponsored-ads names "Sales, 14 day total sales, 7 day total sales". Sales exclude tax and cart discounts and may include items sold by others, used items and halo items (A6). [official] (Purchases (v1); Conversion metrics; Ad campaign attribution; all read 2026-11-09)
+
+E5. RULE. Promoted and halo. Unified reporting and v1 split every conversion family into a promoted part (the advertised or tracked product) and a halo part, and the halo part takes over the old names "other SKU" (seller Sponsored Products) and "brand halo" (vendor Sponsored Products, Sponsored Brands, Sponsored Display, DSP): the console's table lists "Purchases (halo)" ("The number of purchases for highly relevant products that were not directly promoted in the campaign, attributed to an ad interaction") with former names "Purchases - brand halo" and "14 day brand halo ASIN orders, 7 day other SKU orders". v1: `metric.purchasesPromoted` (v3 names "purchasesPromoted, purchasesSameSku7d, purchasesSameSku14d"), `metric.salesPromoted`, `metric.unitsSoldPromoted`; `metric.purchasesHalo` (v3 `purchasesOtherSku7d/14d`), `metric.salesHalo` (`salesOtherSku7d/14d`), `metric.unitsSoldHalo`. v1's overview: "terms such as "Brand Halo" and "Other SKU" are now standardized as "Halo"". [official] (Conversion metrics; Purchases (v1); Overview of the Reporting v1 API; all read 2026-11-09). The meaning of "halo" therefore differs by advertiser: for a seller's Sponsored Products it is any other product the same merchant sells, whatever the brand; for everyone else it is other products of the same brand (A3). [seat, from A3]
+
+E6. RULE. The "From clicks" and "From views" breakouts. `metric.purchasesFromClicks` (v3 `purchasesClicks`, `ordersClicks14d`), `metric.salesFromClicks` (`salesClicks`, `salesClicks14d`), `metric.unitsSoldFromClicks`, and `metric.purchasesFromViews`, `metric.salesFromViews`. For click-only campaigns the base and the "from clicks" figure are equal and "from views" is 0 (C2). [official] (Purchases (v1), above)
+
+E7. RULE. Rates over clicks and over impressions. In unified reporting and v1, "Purchase rate" is per **impression**: `metric.purchaseRate`, "purchases / impressions"; the per-click rate is `metric.purchaseRateOverClicks`, "purchases from clicks / clicks", whose v3 name is "14 Day Conversion Rate". The console's conversion-metrics page: "Purchase rate | The rate of purchases relative to ad impressions. Purchase rate = purchases / impressions" and "Purchase rate over clicks | The rate of click-attributed purchases relative to ad clicks. Purchase rate over clicks = purchases from clicks / clicks", former sponsored-ads name "14 day conversion rate". The same rename runs through the other families: "Detail page view rate" and "Add to cart rate" are per impression too (F3, F4). [official] (Purchases (v1); Conversion metrics; both read 2026-11-09)
+
+E8. RULE (and an Amazon inconsistency). The Sponsored Products search-term report help page still uses the old meaning: "Purchase rate (formerly known as 7 day conversion rate)": "The number of ad-attributed purchases over ad attributed clicks occurring within 7 days of your ad being clicked." The same page defines the vendor column "Purchases (formerly known as 14 day total orders on sponsored ads)" with that rate's wording ("The number of ad-attributed purchases over ad attributed clicks occurring within 14 days…"), a second label error. [official] (Search term report for Sponsored Products, https://advertising.amazon.com/help/G3HEFZYWZF84NPS9, lastModified 2026-05-18, read 2026-11-09). Two Amazon pages give "Purchase rate" two denominators. [seat] The page dated 2026-05-18 predates unified reporting's 2026-10 pages; a column called "Purchase rate" in a seller's file must be read by its source and date.
+
+E9. RULE. ROAS. v1 `metric.roas`: "Return on ad spend (ROAS) = sales / total cost"; v1's page gives its v3 names as "roas, roas7d, roas14d", while the v3 column list carries `roas` (on Sponsored TV and DSP report types only: `stCampaigns`, `stTargeting`, `dspCampaign`, `dspInventory`), `roasClicks7d` and `roasClicks14d`, and no `roas7d` or `roas14d`. Amazon's ROAS help page treats it as "an index (multiplier) rather than a percent (%)". `metric.returnPerThousandImpressions` ("Return per thousand impressions = sales / (impressions / 1,000)"; formerly eRPM) is not offered for Sponsored Products. `metric.costPerPurchase` = "total cost / purchases" (former Display, Video, and Audio name "Total effective cost per purchase (total eCPP)"). [official] (Purchases (v1); Return on ad spend (ROAS), https://advertising.amazon.com/help/G78DVL9PAZZ83KQX; Understand unified reporting; all read 2026-11-09)
+
+E10. RULE. ACOS is absent from unified reporting and from the v1 field library. No field, column or definition named ACOS, ACoS or "advertising cost of sales" appears on the conversion-metrics, traffic-metrics, performance-metrics, unified-reporting or attribution help pages, or on any v1 metric page read (Purchases, including its multi-touch fields; Purchases (all views); Purchases (new to brand)). It survives in three places: (a) Reporting v3, as `acosClicks7d` ("Advertising cost of sales based on purchases made within 7 days of an ad click") and `acosClicks14d`, offered on `spTargeting`, `spSearchTerm`, `spAdvertisedProduct` and the ad-extension report types `spPromptAdExtension`, `sbPromptAdExtension` and `spVideoAdExtension` (not `spCampaigns`), with siblings `roasClicks7d`, `roasClicks14d` on the same report types; (b) bulksheets, where the search-term sheet defines ACOS as "Total spend divided by your total sales during the campaign. It is expressed as a percentage."; (c) Amazon's library guide: "ACOS = (ad spend ÷ ad revenue) x 100", "ROAS is the inverse of ACOS". [official] (Reporting columns (version 3); Bulksheets search term report, https://d3a0d0y2hgofx6.cloudfront.net/en-us/no-code-tools/bulksheets/bulksheets-search-term-report.md; What is ACOS, https://advertising.amazon.com/library/guides/acos-advertising-cost-of-sales; all read 2026-11-09). [seat] ACOS from v1 data is a derived figure, `metric.totalCost / metric.sales`, and is exactly 1/ROAS for the same rows.
+
+E11. RULE. Reporting v3's ACOS and ROAS columns for Sponsored Products are click-window-specific; v3's `roas` is not. A seller's v3 `acosClicks14d` is a 14-day figure even though the seller's base Sponsored Products window is 7 days. [official] (Reporting columns (version 3), above). [seat] A 14-day seller ACOS and the console's seller ROAS (7-day) are two windows; comparing them as one figure is an error.
+
+E12. RULE. Sponsored Brands and Sponsored Display in v3 carry their own unsuffixed columns: `purchases`, `sales`, `unitsSold` ("within 14 days of an ad click or view"), `purchasesClicks`, `salesClicks`, `unitsSoldClicks`; Sponsored Brands report types also carry `purchasesPromoted` ("Same as purchasesSameSku14d") and `salesPromoted` ("Same as attributedSales14dSameSKU"), which the v3 list does not offer on any Sponsored Display report type. v3 `sales14d`: "Includes sales attributed to both clicks and views for vCPM campaigns, but only sales attributed to clicks for CPC campaigns." [official] (Reporting columns (version 3), above)
+
+### F. New-to-brand, detail page views, add to cart, branded searches, long-term sales
+
+F1. RULE. New-to-brand. The NTB help page: "We review the last 12 months of a customer's purchase and browsing history to determine if a purchase or detail page view (DPV) is new to your brand. If the customer hasn't purchased from your brand within the 12-month look back window, the purchase is considered new-to-brand." The attribution page says "hadn't purchased from the brand in the previous 365 days". Performance metrics: "Purchase history is based on all shopping activity within a given brand on Amazon and isn't limited to attributed purchases." The NTB help page: "New-to-brand data isn't available for Sponsored Products." Start dates: "For Sponsored Brands campaigns, data are available from November 1, 2018"; "For Display, data are available from May 1, 2021"; for legacy Sponsored Brands campaigns (DPV from views (NTB), add to cart and branded-search metrics): "While you still may see "0" or "XX" values in these columns for legacy Sponsored Brands campaigns, this data isn't accurate and shouldn't be used". NTB detail page views and NTB add to cart use the same 12-month lookback. [official] (Sponsored ads new-to-brand metrics, https://advertising.amazon.com/help/G5EWB37XK4RS4J4T, lastModified 2026-08-19; Performance metrics; Ad campaign attribution; all read 2026-11-09)
+
+F2. RULE (and an Amazon inconsistency). v1's "Purchases (new to brand)" family (102 fields) lists Sponsored Products in the program eligibility of `metric.newToBrandPurchases` (v3 `newToBrandPurchases`, `newToBrandPurchases14d`), `metric.newToBrandSales` and `metric.newToBrandUnitsSold`, while `metric.newToBrandRoas` omits it. v3 offers `newToBrandPurchases` only on Sponsored Brands, Sponsored Display, Sponsored TV, DSP, `conversionPath` and `sbPromptAdExtension` report types (no `sp*` type) and says "Not available for book vendors". `metric.percentOfPurchasesNewToBrand` lists `budgetCurrency.value` as a required field. [official] (Purchases (new to brand) (Amazon retail conversions), https://d3a0d0y2hgofx6.cloudfront.net/en-us/guides/reporting/ads-v1/metrics/amazon-retail-conversions/purchases-new-to-brand.md; Reporting columns (version 3); both read 2026-11-09). [seat] An empty v1 value means "not available", not zero: "A missing value does not imply zero" (I3; diagnosis.md §L).
+
+F2a. RULE (settled 2026-11-09, re-opened for a cross-file contradiction). Three Amazon sources were re-read. (1) The NTB help page's opening paragraph, verbatim: "New-to-brand metrics are available for Sponsored Brands and Display . … New-to-brand data isn't available for Sponsored Products." (lastModified 2026-08-19). (2) The v3 column list offers `newToBrandPurchases` on no `sp*` report type. (3) The v1 page lists Sponsored Products in the program eligibility of 90 of its 102 NTB fields, including view-attributed fields such as `metric.newToBrandPurchasesFromViews` and `metric.newToBrandSalesFromViews`, while Amazon states that Sponsored Products reports 0 conversions from views (C2). [official] (Sponsored ads new-to-brand metrics, https://advertising.amazon.com/help/G5EWB37XK4RS4J4T; Reporting columns (version 3); Purchases (new to brand) (v1); Overview of the Reporting v1 API; all read 2026-11-09). [seat] The v1 eligibility list names Sponsored Products even on fields Sponsored Products cannot populate, so it is not evidence that SP has NTB data; the help page states Amazon's rule in words and the v3 column list agrees with it. On Amazon's own words, Sponsored Products has no new-to-brand data. The attribution help page's NTB definition ("a subset of attributed purchases") does not name ad products, so it neither includes nor excludes SP.
+
+F3. RULE. Detail page views. `metric.detailPageViews`: detail page views attributed to an ad click or view; `metric.detailPageViewRate` is per impression; breakouts FromClicks, FromViews, Promoted, Halo; NTB DPV is `metric.newToBrandDetailPageViews` (on the separate "Detail page views (new to brand)" page). Program eligibility excludes Sponsored Products. The console says DPV "isn't unique per customer" and that "Detail page views now includes promoted and halo conversions" in unified reporting. Sponsored Products's own metric set does not carry DPV in any surface read. [official] (Detail page views (v1), https://d3a0d0y2hgofx6.cloudfront.net/en-us/guides/reporting/ads-v1/metrics/amazon-retail-conversions/detail-page-views.md; Performance metrics; Understand unified reporting; all read 2026-11-09)
+
+F4. RULE. Add to cart. `metric.addToCart`, `metric.addToCartRate` (per impression), `metric.addToCartFromClicks` (v3 "14-day ATC clicks"), `metric.addToCartPromoted`, `metric.addToCartHalo`; no Sponsored Products in program eligibility. Unified reporting renamed eCPATC to "Cost per add to cart" and eCPATL to "Cost per add to list". Sponsored Products v1 migrations do carry `metric.addToList` on its search-term, targeting and advertised-product examples. [official] (Add to cart (v1); Understand unified reporting; Sponsored Products search term report migration example, https://d3a0d0y2hgofx6.cloudfront.net/en-us/guides/reporting/ads-v1/report-types/sp-search-term.md; all read 2026-11-09)
+
+F5. RULE. Branded searches. v3 `brandedSearches` and `brandedSearchesClicks` ("within 14 days"; also legacy `brandSearches`). Release note of 2025-10-01 ("Branded searches metrics enhancements"): "We have improved how branded searches are identified in attribution" and "Advertisers may notice new trends in their branded search metrics after this update." Release note of 2026-08-26: NTB branded searches and Brand Store page views added to the v1 API; retention of the `date.value`, `week.value`, `day.value` and `dayOfWeek.value` dimensions ""extended from 15 months to 24 months. Maximum report date ranges are unchanged." [official] (Reporting columns (version 3); Amazon Ads advanced tools release notes; both read 2026-11-09). [seat] A branded-search trend crossing 2025-10-01 crosses a definition change.
+
+F6. RULE. Long-term sales. A modelled figure, not a count: Amazon's help page defines it as "a metric that captures both your immediate sales and a 12-month estimate of future sales from new-to-brand (NTB) shopping events" (Long-term Sales = "[Change in Return on Engagement] + [14 day Total Product Sales]"), with requirements (12 months of data, about 30 orders a month and 4,500 detail page views a month **(human readers only)**), a list of marketplaces, and not Sponsored Products. v1 fields `metric.longTermSales`, `…30d/60d/90d/180d`, `metric.longTermRoas`; v3 `longTermSales`, `longTermROAS`; program eligibility Sponsored Brands, Sponsored Display, Sponsored TV, DSP. [official] (Long-term sales, https://advertising.amazon.com/help/GG8E72AW6ASSR6SA; Long-term sales (v1); both read 2026-11-09). Relevant Audience, summarising Digiday, says the metric was "unveiled in closed beta in October 2024"; the help page and the release note of 2025-06-05 ("Long term sales metrics are now available in campaign reporting") show it in reporting since. [opinion, press report; official] (Relevant Audience, Antonio Fernandez, 2026-10-02, citing Digiday 2026-10-01)
+
+F7. RULE. Conversion path reporting (console only, brand owners): a 30-day path of ad touches before a purchase, simplified to first touch. Not yet in unified reporting ("Conversion path report" is on the list of report types the migration tool cannot carry). [official] (Conversion path reporting, https://advertising.amazon.com/help/GV7Y3HZ5XUGZKWUK; Migrate your scheduled reports to unified reporting; both read 2026-11-09)
+
+F8. RULE. Long-term sales after 2026-09-25: a machine-learning forecast with an actuals check (gap fill 2026-11-09). The release note "Validate long-term sales estimates with accumulated sales reporting" (September 25, 2026): "LTS estimates now use machine learning models in place of the previous heuristic-based calculation"; it adds horizons ("`metric.longTermSales` remains the 365-day estimate"), cohort estimates `metric.cohortLongTermSales30d`, `…60d`, `…90d`, `…180d`, `metric.cohortLongTermSales` (365 days), and "Actual sales from a cohort": `metric.cohortAccumulatedSales30d`, `…60d`, `…90d`, `…180d`, `…365d`, grouped by the new dimension `longTermSalesCohort.value` ("groups shoppers by the calendar month in which their new-to-brand conversion was attributed to the campaign, for example `2026M05`"). "Cohort metrics require the `campaign.id` and `longTermSalesCohort.value` dimensions and aren't summable across cohorts." Purpose, in Amazon's words: "Compare LTS estimates against actual accumulated sales from new-to-brand customers to validate estimate accuracy." Availability: "All advertisers in the United States, in the advertising console and the Reporting API v1." [official] (Amazon Ads advanced tools release notes, entry of September 25, 2026, read 2026-11-09)
+
+F9. RULE. The v1 field definitions, verbatim. `metric.longTermSales`: "Attributed sales plus the estimated 365-day value of new-to-brand shopper engagements attributed to the campaign, based on the brand's historical shopping patterns." `metric.cohortAccumulatedSales365d`: "Sales over 365 days from the cohort of shoppers whose shopping events were used as inputs to long-term sales calculations." Each cohort field requires `budgetCurrency.value`, `campaign.id` and `longTermSalesCohort.value`; `metric.longTermSales` lists "Long-term sales cohort" among its incompatible dimensions; all 24 fields list Sponsored Brands, Sponsored Display, Sponsored TV, Amazon DSP and Full-Funnel (benchmarks omit Full-Funnel), not Sponsored Products; the v3 name of every new horizon and cohort field is "N/A". [official] (Long-term sales (Amazon retail conversions), https://d3a0d0y2hgofx6.cloudfront.net/en-us/guides/reporting/ads-v1/metrics/amazon-retail-conversions/long-term-sales.md, read 2026-11-09)
+
+F10. RULE (dated page). The console help page for long-term sales (lastModified 2026-08-26, before the September switch) still describes the heuristic: "A nominal value is assigned to each brand engagement in the shopper funnel based on customers who have taken that same shopping event, and their subsequent sales over the following 12-months. This is called return on engagement (ROE)."; "an incremental return on engagement value is assigned to the responsible campaign that drove the new customer engagement"; Long-term Sales = "[Change in Return on Engagement] + [14 day Total Product Sales]". Its worked example subtracts the value of the shopper's previous funnel stage ("$20 (add to cart value) - $3 (LTS from product detail page view in week 1) = $17 LTS for Campaign B"). Amazon calls the metric both "ad-attributed" and "incremental": the June 5, 2025 release note says LTS and LTS ROAS "are ad-attributed measures that estimate the incremental sales value your brand can expect to generate over the next 12 months". [official] (Long-term sales, https://advertising.amazon.com/help/GG8E72AW6ASSR6SA; Amazon Ads advanced tools release notes, June 5, 2025; both read 2026-11-09). [seat] In the help page's own mechanics "incremental" means the step in return-on-engagement value between a shopper's funnel stages, credited to the campaign the engagement is attributed to; no page read describes a holdout or a comparison with unexposed shoppers in computing LTS. LTS is therefore an attribution-based forecast: the campaign's attributed new-to-brand engagements, valued by a model of what such shoppers later spend.
+
+F11. [seat] What makes such a forecast a valid long-term *effect*, stated in the terms of the surrogate literature (Published research, item 14): LTS uses short-term engagements (detail page view, branded search, add to cart, purchase) as surrogates for 12-month sales. Reading LTS as the campaign's causal 12-month value needs (a) the short-term engagements credited to the campaign to be caused by it (attribution does not establish this; §A, M3), (b) the Prentice surrogacy condition, that 12-month sales are independent of the ad given the surrogates (an ad that also changes later behaviour by another path breaks it), and (c) comparability, that the engagement-to-sales relation learned from the brand's historical shoppers holds for the shoppers this campaign reached. `metric.cohortAccumulatedSales*` checks the prediction step for the attributed cohort (did these shoppers go on to spend what LTS forecast?); it does not check (a), because the cohort has no unexposed comparison. The IAB/MRC guidelines require of any 12-month forecast that "The basis for such forecasts must be disclosed and the methodology for extending shorter period results to 12 months or longer periods should be re-evaluated periodically based on empirical evidence" (§R4); the accumulated-sales fields are the means of that re-evaluation on a seller's own data, available in the US.
+
+### G. Video and engagement metrics
+
+G1. RULE. Video quartiles count unduplicated progress: "First quartile (video ad)": "The number of times at least 25% of a video ad was played. If the ad was rewound, we use the total percentage of unduplicated video that was viewed to calculate the completion rate.", and the same for midpoint, third quartile and complete views. "5-second views (video ad)": "This is only available for Sponsored Brands video ads." (console); v3's `video5SecondViews` says "Sponsored Brands and Sponsored Products video-only metric" and is offered on `spVideoAdExtension` too. VTR is views over impressions; CPV is cost per view. [official] (Engagement metrics, https://advertising.amazon.com/help/GH4WXV3ZBQE37R6M, lastModified 2026-10-01; Performance metrics; both read 2026-11-09)
+
+G2. RULE. v1 names: `metric.startsVideoAd`, `metric.firstQuartileVideoAd`, `metric.midpointVideoAd`, `metric.thirdQuartileVideoAd`, `metric.completeViewsVideoAd`, `metric.completionRateVideoAd`, `metric.costPerCompletedViewVideoAd`, `metric.unmutesVideoAd`, `metric.5secondViewsVideoAd`, `metric.5secondViewRateVideoAd`. v3: `videoFirstQuartileViews`, `videoMidpointViews`, `videoThirdQuartileViews`, `videoCompleteViews`, `videoUnmutes`, `video5SecondViews`, `video5SecondViewRate`. [official] (Attention (video ad) (Engagement) (v1); Reporting columns (version 3); both read 2026-11-09)
+
+G3. RULE. Sponsored Products video is not in unified reporting yet ("Sponsored Products Video report" cannot be migrated). Interactive impressions and notification opens or clicks are separate engagement metrics; legacy Display, Video, and Audio pixel conversion metrics (Accept, Lead, Signup) "are deprecated" and "don't return values in reports you download" (Migrate your scheduled reports to unified reporting). [official] (Migrate your scheduled reports to unified reporting; Engagement metrics; both read 2026-11-09)
+
+### H. KDP metrics
+
+H1. RULE. KENP read and estimated royalties. KENP Read and KENP Royalties are visible only under this condition: "You'll be able to see KENP metrics in your campaign manager if, as a minimum, 1 of your Sponsored Products campaigns promotes a book that has at least 1 KENP Read attributed to it." Reading events are attributed up to 14 days after the click and dated to the reading day (B3); offline reads can take a week; royalties are an estimate: the metric "is calculated by multiplying KENP read by the KENP rate at the time the pages were read. If the KENP rate of the day isn't yet published, we use the latest published rate to estimate the royalties." [official] (KENP metrics, above; Ad campaign attribution, above). The attribution page adds for authors: "Attribution reflects orders when they're created", "KDP sales reports for paperback and hardcover books reflect sales after printing and shipping", and "The attributed sales for audiobooks include credit redemption, borrows, and cash purchases." [official] (Ad campaign attribution, above)
+
+H2. RULE. v1 names: `metric.kindleEditionNormalizedPagesRead` (v3 `kindleEditionNormalizedPagesRead`, `kindleEditionNormalizedPagesRead14d`), `…FromClicks`, `…FromViews`; `metric.kindleUnlimitedQualifiedBorrows` (v3 `qualifiedBorrows`); `metric.estimatedKindleUnlimitedRoyalties` (v3 `royaltyQualifiedBorrows`). v3 also has `kindleEditionNormalizedPagesRoyalties14d`. The release note of 2025-09-30 added Sponsored Display for authors and further Kindle metrics. [official] (KENP / Kindle Unlimited (v1); Reporting columns (version 3); Amazon Ads advanced tools release notes; all read 2026-11-09)
+
+H3. RULE (documentation error). v1's page maps `metric.kindleUnlimitedQualifiedBorrowsFromClicks` to the v3 name `qualifiedBorrowsFromViews` and `metric.kindleUnlimitedQualifiedBorrowsFromViews` to `qualifiedBorrowsFromClicks`: the two are swapped. [official, the page's own error] (KENP / Kindle Unlimited (v1), above). [seat] A skill maps by the v1 field ID, not by the v3 column the page names for it.
+
+H4. [seat] An author's ACOS or ROAS from attributed sales excludes KENP royalties unless the two are added; whether to add them is an economics question (`beyond/goals-and-economics.md`).
+
+### I. Reporting v3 against Reporting API v1: the name map
+
+I1. RULE. Shape. v3: `POST /reporting/reports` with `reportTypeId` (e.g. `spCampaigns`, `spTargeting`, `spSearchTerm`, `spAdvertisedProduct`, `spPurchasedProduct`, `sbPurchasedProduct`), `groupBy`, `columns`, `timeUnit` ("DAILY" or "SUMMARY") and `filters`. v1: `POST /adsApi/v1/create/reports` with `query.fields` (`metric.*` plus dimension fields such as `campaign.id`, `date.value`), `periods[].datePeriod`, `format`, `currencyOfView` and `accessRequestedAccounts`. In v1, time grain is a dimension: `date.value` for daily, `week.value`, `month.value`, `hour.value`. [official] (OfflineReport OpenAPI model, https://d1y2lf8k3vrkfu.cloudfront.net/openapi/en-us/dest/OfflineReport_prod_3p.json; Helpful concepts (v1), https://d3a0d0y2hgofx6.cloudfront.net/en-us/guides/reporting/ads-v1/helpful-concepts.md; both read 2026-11-09). Report mechanics are `method/reporting.md`'s.
+
+I2. RULE. The map, from v1's own "Version 3 Reporting Name" lines (v1 field → v3 Sponsored Ads column):
+
+| Concept | v1 field | v3 column(s) |
+|---|---|---|
+| Impressions | `metric.impressions` | `impressions` |
+| Gross / invalid impressions | `metric.invalidImpressions`, `metric.invalidImpressionRate` (no gross-impressions field on v1's Impressions page) | `grossImpressions`, `invalidImpressions`, `invalidImpressionRate` |
+| Clicks | `metric.clicks` | `clicks` |
+| Gross / invalid clicks | `metric.grossClicks`, `metric.invalidClicks`, `metric.invalidClickRate` | `grossClickThroughs`, `invalidClickThroughs`, `invalidClickThroughRate` |
+| CTR | `metric.ctr` | `clickThroughRate` |
+| CPC | `metric.cpc` | `costPerClick` |
+| Cost | `metric.totalCost` | `spend`, `cost` |
+| Viewable impressions | `metric.viewableImpressions` | `viewableImpressions`, `impressionsViews` |
+| Viewable share of impressions | `metric.viewableDistribution` | `viewabilityRate` |
+| vCTR | `metric.vctr` | `viewClickThroughRate` |
+| Purchases | `metric.purchases` | `purchases`, `purchases7d`, `purchases14d`, `orders14d` |
+| Sales | `metric.sales` | `sales`, `sales7d`, `sales14d` |
+| Promoted purchases / sales | `metric.purchasesPromoted`, `metric.salesPromoted` | `purchasesSameSku7d/14d`, `attributedSalesSameSku7d/14d`, `purchasesPromoted`, `salesPromoted` (v1's page names `salesSameSku7d/14d`, which the v3 column list does not carry) |
+| Halo purchases / sales | `metric.purchasesHalo`, `metric.salesHalo` | `purchasesOtherSku7d/14d`, `salesOtherSku7d/14d`; v3 also has `salesBrandHalo` (Sponsored Display vCPM and `dspProduct` only) |
+| Click-attributed purchases | `metric.purchasesFromClicks` | `purchasesClicks`, `ordersClicks14d` |
+| Per-click conversion rate | `metric.purchaseRateOverClicks` | "14 Day Conversion Rate" |
+| ROAS | `metric.roas` | `roas` (Sponsored TV and DSP report types only), `roasClicks7d`, `roasClicks14d` (SP); v1's page also names `roas7d`, `roas14d`, which the v3 column list does not carry |
+| ACOS | none | `acosClicks7d`, `acosClicks14d` |
+| NTB purchases | `metric.newToBrandPurchases` | `newToBrandPurchases`, `newToBrandPurchases14d` |
+| KENP read | `metric.kindleEditionNormalizedPagesRead` | `kindleEditionNormalizedPagesRead`, `kindleEditionNormalizedPagesRead14d` |
+| Top-of-search IS | `metric.topOfSearchImpressionShare` | `topOfSearchImpressionShare` |
+| Budget currency | `budgetCurrency.value` | `campaignBudgetCurrencyCode` |
+| Search term | `searchTerm.value` | `searchTerm` |
+| Placement | `placementClassification.value` | `placementClassification` |
+
+[official] (v1 metric and dimension pages under https://d3a0d0y2hgofx6.cloudfront.net/en-us/guides/reporting/ads-v1/; Reporting columns (version 3); all read 2026-11-09). The table is a convenience; the run-time source is the v1 field page.
+
+I3. RULE. v1's required companions: "If you want to include a dimension attribute in a report (e.g. `campaign.name`), you must also include the primary key (e.g. `campaign.id`)"; money metrics such as `metric.sales` and `metric.cpc` require `budgetCurrency.value`; converted-currency metrics require `convertedCurrency.value`. Incompatible combinations return "400004: field {FIELD_1} cannot be used along with fields: {FIELD_2}" and a missing required field "400006: field {FIELD_1} cannot be used without also including fields: {FIELD_2}". Nulls: in CSV "Fields without a value are represented as empty strings", in JSON as `null`; "A missing value does not imply zero." [official] (Helpful concepts (v1), above; the `ReportCreate` schema in the merged OpenAPI model, https://d1y2lf8k3vrkfu.cloudfront.net/openapi/en-us/dest/AmazonAdsAPIALLMerged_prod_3p.json, read 2026-11-09, which does not carry the error codes)
+
+I4. RULE. Lineage of names in the console, to read old exports: "Total cost" was "Spend"; "Purchases" was "Orders"; "Purchase rate over clicks" was "14 day conversion rate"; "Purchases (promoted)" was "7 day advertised SKU orders"; "Purchases (halo)" was "7 day other SKU orders" / "14 day brand halo ASIN orders"; "Impression share rank" was "Search term impression rank"; "Viewable distribution" was "VTR" / "View-through rate"; "Cost per add to list" was eCPATL; "Return per thousand impressions" was eRPM; "Percent of purchases new to brand" was "14 day % of orders new-to-brand". [official] (Conversion metrics; Traffic metrics; Understand unified reporting; Migrate your scheduled reports to unified reporting; all read 2026-11-09)
+
+### J. Time zone and currency of a report's day
+
+J1. RULE. v1 aggregates by the campaign's time zone: dates are "aggregated based on the time zone of each campaign" (`timeZone.value`), with Amazon's worked table showing the same click landing on different calendar days for campaigns set to EST, GMT, CET and JST; Amazon states that normalisation to UTC is planned. `periods` dates are inclusive. [official] (Dates and historical data retention (v1), https://d3a0d0y2hgofx6.cloudfront.net/en-us/guides/reporting/ads-v1/dates.md, read 2026-11-09). `timeZone.value` was not found on the v1 dimension library list read the same day (Open questions).
+
+J2. RULE. The profile carries the account's zone and currency: `GET /v2/profiles` returns `countryCode`, `currencyCode` and `timezone` per profile, and Amazon's own example shows `America/Los_Angeles` for US, CA and MX profiles. v3 exposes `advertiserTimezone` on DSP report types only (`dspCampaign`, `dspInventory`, `dspAudience`). Console downloads show "the time zone under which the account is registered". AMC defaults to UTC, with a custom time zone available in the query. [official] (Profiles, https://d3a0d0y2hgofx6.cloudfront.net/en-us/guides/account-management/authorization/profiles.md; Reporting columns (version 3); Downloadable reports in sponsored ads, https://advertising.amazon.com/help/GBYSPTSLR337JMLH; AMC data sources guides; all read 2026-11-09). [seat] The SP-API order and Sales and Traffic data a skill might join to are dated in their own zones; the join rule is `areas/ads-organic-and-total.md`'s and diagnosis.md §M's.
+
+J3. RULE. Currency. Money metrics are in the campaign budget currency (`budgetCurrency.value`; v3 `campaignBudgetCurrencyCode`). v1's `currencyOfView` and the `*Converted` fields (`metric.totalCostConverted`, `metric.cpcConverted`, `metric.salesConverted` …) give a converted view, which "requires `convertedCurrency.value`"; Helpful concepts' example names `metric.convertedSales`, while the Purchases field page carries `metric.salesConverted`. The Sponsored Products campaign report converts with a daily exchange rate. In the console, the old "Currency of Value" fields are replaced by the column setting "Show Converted Currency". [official] (Helpful concepts (v1); Campaign report for Sponsored Products, https://advertising.amazon.com/help/G4KJE82E7LDKAHZX; Migrate your scheduled reports to unified reporting; all read 2026-11-09). [seat] A multi-marketplace total in converted currency moves with exchange rates even when nothing in the account changed.
+
+J4. RULE. Amazon's own ad charges to a seller sit in SP-API Finances as `ProductAdsPaymentEvent` (`postedDate`, `transactionType` charge or refund, `invoiceId`, `baseValue`, `taxValue`, `transactionValue` = base + tax): an invoice-level money flow, not a per-day metric, and it includes tax where the ad metrics do not. [official] (Selling Partner API Finances model, https://raw.githubusercontent.com/amzn/selling-partner-api-models/main/models/finances-api-model/financesV0.json, read 2026-11-09). Data Kiosk's economics schema has an `AdSummary` ("advertisements spend data summary … aggregated by transaction event type: charge"). [official] (Data Kiosk analytics economics schema 2024-03-15, https://raw.githubusercontent.com/amzn/selling-partner-api-models/main/schemas/data-kiosk/analytics_economics_2024_03_15.graphql, read 2026-11-09). [seat] Reported cost (`metric.totalCost`, after invalid-traffic removal, before tax) and invoiced charge (posted on invoice dates, with tax, after adjustments) are two numbers for one spend.
+
+### K. Console, API and bulksheet: where the same metric differs
+
+K1. RULE. Bulksheets define two metrics backwards. Amazon's bulksheet search-term page, column R "Click-through Rate": "This is calculated as impressions divided by clicks."; column W "Conversion Rate": "The total number of orders divided by sales within 7 days from the date that your ad was clicked." [official] (Bulksheets search term report, above; the cx-listing-playbook research recorded the same text earlier, so the error has stood for months). [seat] The computed column in a downloaded bulksheet is not checked here; the definitional text is plainly inverted against every other Amazon page (D2, E7). A skill computes CTR and conversion rate from the raw columns.
+
+K2. RULE. The bulksheet search-term sheet has its own window: "The lookback window in bulksheets search term reports is always 7 days for both vendors and sellers", and the date range "does not affect the search term report metrics"; it "includes only the terms that resulted in at least one ad click". [official] (Bulksheets search term report, above). [seat] A vendor's bulksheet search-term sales (7-day) and the same vendor's console search-term sales (14-day) differ by construction.
+
+K3. RULE (and an Amazon inconsistency). Search-term coverage. The v3 FAQ says the console search-term report "contains all search terms that had an impression, regardless of if that term resulted in a click" while "The API search term report only contains impressions that also had at least one click". Unified reporting's template says search terms "that resulted in at least 1 click", and the SP search-term help page says the report shows "only the search terms that resulted in at least 1 ad click". [official] (Reporting v3 FAQ, https://d3a0d0y2hgofx6.cloudfront.net/en-us/guides/reporting/v3/faq.md; Understand unified reporting; Search term report for Sponsored Products; all read 2026-11-09). [seat] The newer pages say click-only; the FAQ may predate them. Either way, impression-only search terms are absent from the API.
+
+K4. RULE. The search-term report covers inferred terms too: search terms used "or inferred as best match … (such as off Amazon placements on social sites)"; ASIN-like strings appear as terms; the Sponsored Products search-term report lookback is 65 days. [official] (Search term report for Sponsored Products, above). Use of the terms is `areas/search-terms-and-negatives.md`'s.
+
+K5. RULE. Report rows that do not add up, by Amazon's own word: the purchased-product report's totals "won't match" campaign totals (it is the converted-product side, available to sellers only, 60-day lookback); traffic metrics are absent when the converted-product dimension is selected (D3). [official] (Purchased product report for Sponsored Products, https://advertising.amazon.com/help/G9JR57B7AEWBQ3P2; Data availability in unified reporting; both read 2026-11-09). The purchased-product page labels "Sales (halo) (formerly 7 day other SKU sales) (#)" as "The number of orders" — a label error. [official, the page's own error]
+
+K6. RULE. Sponsored Brands report previews in v3 omit campaigns with `isMultiAdGroupsEnabled=False` (legacy Sponsored Brands); legacy Sponsored Brands targeting and search-term reports are not yet in unified reporting. [official] (Reporting v3 FAQ; Understand unified reporting; both read 2026-11-09)
+
+K7. RULE. History limits differ by surface. Unified reporting keeps up to 24 months at daily or weekly grain and 6 years at monthly or yearly grain. v1's table gives, per time dimension, "Data Retention" and "Max Report Date Range": `hour.value` 14 days / 14 days; `date.value` 24 months / 120 days; `week.value`, `day.value`, `dayOfWeek.value` 24 months / 15 months; `month.value` 6 years / 25 months; `year.value` and `dateRange.value` 6 years / 6 years. The help page's table for the console and the Reporting API reads the same (72 months for month and year). Limited-history fields start at fixed dates: advertised and converted product from 2025-02-13, reconciled cost and ROAS from 2024-08-04, reach from 2024-07-01, off-Amazon from 2025-06-27, benchmarks from 2025-06-01. The SP campaign help page gives a 5-year lookback (hourly 30 days). **(human readers only)** [official] (Understand unified reporting; Data availability in unified reporting; Dates and historical data retention (v1); Campaign report for Sponsored Products; all read 2026-11-09). Mechanics are `method/reporting.md`'s.
+
+K8. RULE. Not yet in unified reporting, by Amazon's lists: MRC viewability, Benchmarks (added 2026-09-25 per the release notes), Conversion Path, SP video, SP prompts, SP across retailers, legacy SB targeting and search-term reports; the migration tool cannot carry the pricing-transparency, conversion-path, Amazon Attribution, SP retailer-dimension, SP/SB prompt, SP video, industry-standard, benchmark, SP budget and Amazon Business SP placement reports. [official] (Understand unified reporting; Migrate your scheduled reports to unified reporting; Amazon Ads advanced tools release notes; all read 2026-11-09). These lists are dated and shrink; a skill reads them.
+
+K9. RULE. Console cadence: "The metrics in the campaign manager are updated every hour"; sales "up to 48 hours"; "Yesterday" may be incomplete (D5). The ROAS help page gives Sponsored Products sales up to 48 hours and Sponsored Brands up to 12 hours. Unified reporting lags 1 day; some DSP reports 2 days. [official] (Performance metrics; Return on ad spend (ROAS); Data availability and freshness, https://advertising.amazon.com/help/GLN6KZCS23ESMTUS; all read 2026-11-09). **(human readers only)**
+
+K10. [opinion] Openbridge's support article (2024-09-27) says Amazon treats a console-to-API gap above 5% as a possible data issue for Amazon to investigate, smaller gaps as normal, and that "Amazon recommends retrieving reports after 72 hours from the Advertising API"; it lists console-only reports (SP Performance Over Time, SP and SB Search Term Impression Share, SP Budget, SB Category Benchmark). No Amazon page read states the 5% figure. (Openbridge, "Amazon Advertising Console vs Advertising API", https://docs.openbridge.com/en/articles/5575078-amazon-advertising-console-vs-advertising-api, read 2026-11-09)
+
+### L. When figures settle: restatement, invalid traffic, cancellations
+
+L1. RULE. Invalid traffic is filtered and removed from what a seller pays and sees. Amazon's definition covers non-human and fraudulent traffic and also "duplicate clicks" and accidental clicks; filtering runs pre-bid and post-impression, with monitoring "for up to 30 days"; invalid traffic is removed from invoices and reports and reports may be restated. [official] (Traffic validation, https://advertising.amazon.com/help/GBYPYH79NGJJ5JPS, lastModified 2026-11-02, read 2026-11-09). "Most" detection completes within 6 hours and invalid clicks are usually removed the same day; Amazon recommends "waiting at least 5 days" before judging; detection continues up to 30 days and invoice adjustments follow the billing cycle. [official] (Adjustments to invalid spend, https://advertising.amazon.com/help/GWQ7F2EL4E2Y7H4B, read 2026-11-09). Figures **(human readers only)**.
+
+L2. RULE. Amazon on reading invalid-traffic rates: "IVR level on its own doesn't indicate undetected IVT"; "if you notice a significant drop in IVR, with an anomalous increase in select campaign metrics like click-through rate (CTR) or Impressions or Clicks, it could suggest undetected IVT was billed to your campaign. For Sponsored Ads campaigns, a drop in Conversion Rate (CVR) along with the above changes in metrics, is a higher confidence indicator for undetected IVT." For Display, Video, and Audio campaigns, "a small discrepancy (up to 5%) between Amazon and third-party reported IVR is expected". [official] (Traffic validation, above). Diagnosis use is `areas/diagnosis.md` §K.
+
+L3. RULE. API restatement schedule as Amazon states it: data available within 12 hours; clicks can change for up to 3 days; conversions appear within 24 hours and are restated at 1, 7 and 28 days, up to 42 days for a 14-day window; reports take up to 3 hours. [official] (Reporting v3 FAQ, above). Amazon Ads reporting overview adds: invalid-traffic adjustment for up to 30 days, "waiting at least 5 days", and conversions can change for up to 60 days, so re-retrieve monthly; DSP invalid traffic is monitored for 72 hours. [official] (Amazon Ads reporting overview, https://d3a0d0y2hgofx6.cloudfront.net/en-us/guides/reporting/overview.md, read 2026-11-09). The DSP freshness page: invalid-traffic decreases appear within 72 hours; ASIN changes trigger restatement within 28 days; reach and frequency are recomputed within 96 hours. [official] (Data availability and freshness, above). All figures **(human readers only)**.
+
+L4. RULE. Marketing Stream restates by re-delivery: restated records arrive again with a new total, and the consumer keeps the record with the highest `streamBatch.version`. [official] (Amazon Ads advanced tools release notes, 2026-09-30 entry, above). [seat] cx does not consume Stream (the door is request and response); the fact matters only for a seller comparing their own Stream-fed tool with cx's pull.
+
+L5. RULE. Cancellations within 72 hours and payment failures are deducted, taking up to 7 days (A6). Downloadable reports "exclude canceled purchases". [official] (Ad campaign attribution; Downloadable reports in sponsored ads; both above)
+
+L6. [seat] Combining A1, B1, L1 and L3: a day's figures are near-final for cost after about 3 days, and for conversions only after the lookback window plus the restatement schedule, which for a 14-day product runs past four weeks. `method/change-measurement.md` owns how this enters a before-and-after read; the numbers here are read from the pages at run time.
+
+### M. Multi-touch attribution beside last touch
+
+M1. RULE. Amazon's multi-touch attribution (MTA), "Announced at unBoxed 2024", is a beta for US advertisers that divides credit across ad touches "in proportion to their value", using experiments, machine learning and shopping signals; its metrics, "Orders (multi-touch)", "Sales (multi-touch)" and "ROAS (multi-touch)", sit beside the last-touch metrics, and they "can differ from last-touch attribution at all reporting grains". The page also names a multi-touch ACoS and breakouts for promoted and brand halo, same and other SKU, and click and view. Its two surface sentences, verbatim (re-read 2026-11-09; page metadata datePublished 2024-10-15, dateModified 2026-10-09): "Multi-touch metrics are available alongside familiar last-touch metrics in Ad console, Amazon DSP campaign manager, sponsored ads downloadable reports (Campaign, Targeting, Keyword, Placement, Search term, Advertised product, and Purchased product), Amazon DSP downloadable custom reports (Campaign, Inventory, Geography, Technology, Product) for all US advertisers using Amazon DSP, Sponsored Display , Sponsored Brands , or Sponsored TV ." and "Multi-touch metrics can also be accessed in most report types for version 3 of the Amazon Ads API, as well as Amazon Marketing Stream conversion datasets for Amazon DSP, Sponsored Products , Sponsored Brands, and Sponsored Display." [official] (Uncover ad value across the funnel with multi-touch attribution (beta), https://advertising.amazon.com/resources/whats-new/uncover-ad-value-across-funnel-with-multi-touch-attribution, read 2026-11-09). [seat] The first sentence's advertiser list omits Sponsored Products; the second names Sponsored Products in a list that may govern the Stream datasets only or the v3 report types too. The page alone does not settle whether a seller's Sponsored Products rows carry multi-touch values.
+
+M1a. RULE (settled 2026-11-09, re-opened for a cross-file contradiction). The v3 claim is not borne out by Amazon's own v3 column list: Reporting columns (version 3) contains no column whose name or description includes "multi-touch", "multitouch" or "MTA" (0 matches in 341,340 characters), and every one of v1's 56 `metric.multiTouch*` fields gives "Version 3 Reporting Name (Sponsored Ads): N/A" and "Version 3 Reporting Name (DSP): N/A". The Amazon Marketing Stream data guide and Stream v2 overview read the same day contain no multi-touch field either. The fields that do exist are v1's: `metric.multiTouchPurchases` ("The number of purchases attributed to an ad interaction. This is measured using multi-touch attribution.", required field `metric.purchases`), `metric.multiTouchRoas` ("ROAS (multi-touch) = sales (multi-touch) / total cost", required field `metric.roas`), and the rest of the family in M2. [official] (Reporting columns (version 3), https://d3a0d0y2hgofx6.cloudfront.net/en-us/guides/reporting/v3/columns.md; Purchases (Amazon retail conversions) (v1); Amazon Marketing Stream data guide, https://d3a0d0y2hgofx6.cloudfront.net/en-us/guides/amazon-marketing-stream/data-guide.md; Amazon Marketing Stream v2 overview, https://d3a0d0y2hgofx6.cloudfront.net/en-us/guides/amazon-marketing-stream/v2/overview.md; all read 2026-11-09). So both earlier readings in this research set hold in part: "no multi-touch column on the v3 columns page" is true; "no API column name was found" is not, because v1 names 56. The release notes contain no multi-touch entry (searched 2026-11-09).
+
+M2. RULE. v1 fields: `metric.multiTouchPurchases`, `metric.multiTouchSales`, `metric.multiTouchRoas`, with FromClicks, Promoted and Halo variants; plus units, cost-per-purchase, rate, return-per-thousand and reconciled-ROAS variants (56 `metric.multiTouch*` fields), each with program eligibility "N/A" and v3 names "N/A" on the page. They sit on the Purchases page; v1 has no separate multi-touch page. [official] (Purchases (Amazon retail conversions), https://d3a0d0y2hgofx6.cloudfront.net/en-us/guides/reporting/ads-v1/metrics/amazon-retail-conversions/purchases.md, read 2026-11-09). The method is published by Amazon researchers (§Published research).
+
+M3. [seat] Last touch credits the single last interaction (A4); multi-touch splits it. Neither is an incrementality measure in itself; the MTA paper calibrates to experiments, which brings it closer, but the seller-facing number remains a modelled credit.
+
+### N. Reporting surfaces and their dates
+
+N1. RULE. Dates that change what a seller's numbers mean **(dates are rules; read the page for any move)**:
+- 2025-10-01: branded-search matching enhanced (F5).
+- 2026-01-01: shopping-signal view model for vCPM Sponsored Brands and Sponsored Display and DSP Store ads (C3).
+- 2026-06-08: console migration to unified reporting opens; 2026-11-16 a 45-day warning; 2026-12-17 new legacy reports and edits disabled, read-only; 2026-12-31 the old report centres "are permanently decommissioned. All remaining saved reports, schedules, and historical data are deleted." (Migrate your scheduled reports to unified reporting, above; "These dates are subject to change.")
+- 2026-09-30: Reporting API v1 and Stream v2 generally available (E1).
+- 2026-11-02: AMC NTB columns include modeled conversions (release notes).
+- After 2027-06-30: "To be shutoff" — v3 reporting, Amazon Marketing Stream version 1, Version 2 reporting (sponsored ads) and Amazon DSP reporting (Amazon Ads reporting overview, above).
+[official] (pages as cited, read 2026-11-09)
+
+N2. RULE. v1 report limits as stated by Amazon conflict: the v1 overview says "one concurrent report per client ID, not per advertiser account", while the reporting overview says "one concurrent report per advertiser. Only one report can be running at a time for a given advertiser account", and the release note of 2026-09-30 also states the limit per advertiser; both overviews put the result at "approximately 100 reports per day". [official] (Overview of the Reporting v1 API; Amazon Ads reporting overview; Amazon Ads advanced tools release notes; all above). Ownership of limits is `method/reporting.md`'s.
+
+N3. RULE. Reporting warnings in v1 include DATE_LIMITED, NO_PERMISSION and PENDING_DEPRECATION (returned on completed reports since the release note of 2026-10-05); formats CSV, GZIP_JSON, PARTITIONED_CSV, PARTITIONED_GZIP_JSON. Account count per request: Helpful concepts says "up to 5 accounts at the same time within the `accessRequestedAccounts` configuration", while the merged model's create-report request sets `accessRequestedAccounts` `maxItems` 2000; the two Amazon sources disagree. [official] (merged OpenAPI model, above; Helpful concepts (v1); Amazon Ads advanced tools release notes)
+
+### O. Amazon Marketing Cloud, as far as a seller reaches
+
+O1. RULE. AMC is a clean room over pseudonymized event-level data; raw events cannot be exported; outputs pass aggregation thresholds. A sponsored-ads advertiser can get a self-service instance "within minutes", scoped to one advertiser; queries are AMC SQL. Instance-level APIs retired 2024-08-01; instance listing is `GET /amc/instances` with the header `Amazon-Advertising-API-AdvertiserId`, and AMC terms "must be accepted once in each region (NA, EU, and FE)". [official] (How does Amazon Marketing Cloud work, …/how-amc-works.md; Overview - Amazon Marketing Cloud, …/overview.md; Get access to AMC using self-service APIs, …/get-started/instant-access-to-amc.md; all under https://d3a0d0y2hgofx6.cloudfront.net/en-us/guides/amazon-marketing-cloud/, read 2026-11-09)
+
+O2. RULE. Tables a seller meets: `sponsored_ads_traffic` (one record per impression and per click, with `ad_product_type`); `conversions` ("ad-attributed if a user was served a traffic event within the 28-day period prior"); `amazon_attributed_events_by_conversion_time` and `…_by_traffic_time` (B4). Modeled conversions show NULL below ad-line level. Sponsored Products and Sponsored Brands search terms are queryable. Retention 13 months to 5 years by data view (traffic and conversion views up to 5 years), with "most tables" backfilled to June 2023. Aggregation thresholds run none / low (2 users) / medium (100) / high (100 plus filter limits) / very high (`user_id`). [official] (AMC data sources guides, above). **(human readers only)** for the counts.
+
+O3. RULE. AMC cannot change a campaign's attribution window ("the attribution window is decided based on the campaign and cannot be changed in AMC", B4), but its `conversions` table relates conversions to traffic over 28 days, which is how practitioners look past the 7- or 14-day window. Amazon's own launch page of 2022-03-25 says advertisers "can now set lookback window to be up to 28 days in AMC" and choose first-touch, last-touch, equal-weight or position-based models in query analysis, beside "The standard "14-day last-touch attribution" … practiced in Amazon Ads standard campaign reporting". [official] (AMC data sources, above; Amazon Marketing Cloud now supports custom attribution analysis, https://advertising.amazon.com/resources/whats-new/amazon-marketing-cloud-custom-attribution, read 2026-11-09). [opinion] Intentwise's Sreenath Reddy (2023-04-13) describes reaching 28 days through AMC SQL for this purpose.
+
+O4. [seat] AMC sits on the Ads API host; whether cx's door reaches the AMC paths is `method/ads-api-surface.md`'s question. This file records only what the data is.
+
+### P. Amazon Attribution (off-Amazon traffic)
+
+P1. RULE. Amazon Attribution measures Amazon conversions from non-Amazon media (search, social, email) through tags. The API overview opens it to "professional seller brand owners enrolled in Amazon Brand Registry and vendors" in US, CA, MX, UK, DE, FR, IT, ES and NL; the product page adds KDP authors and agencies and calls it free; it is reached through the advertiser's ad console account, and a profile with `subType` `AMAZON_ATTRIBUTION` is a legacy profile that "New integrators should filter out". [official] (Amazon Attribution overview and get-started guides, https://d3a0d0y2hgofx6.cloudfront.net/en-us/guides/amazon-attribution/; Amazon Attribution product page, https://advertising.amazon.com/solutions/products/amazon-attribution; both read 2026-11-09)
+
+P2. RULE. "the last-clicked tag in the 14-day window gets full credit"; promoted against total and brand halo are reported; clicks take up to 24 hours and conversions up to 48 hours to appear; restatements at 1, 7 and 28 days, and Amazon recommends pulling the last 30 days daily; rows with fewer than 10 lifetime clicks are masked as zero. [official] (Amazon Attribution how-to, above)
+
+P3. RULE. Dated to the event, not the click (B3) — the opposite of sponsored ads. [official] (Amazon Attribution OpenAPI model, above)
+
+P4. RULE. Report types: PERFORMANCE (groupBy CAMPAIGN, ADGROUP or CREATIVE; last 13 months) and PRODUCTS (always ad-group level, no click-throughs, last 90 days). Performance metrics: `Click-throughs`, `attributedDetailPageViewsClicks14d`, `attributedAddToCartClicks14d`, `attributedPurchases14d`, `unitsSold14d`, `attributedSales14d`, the `attributedTotal…14d` / `totalUnitsSold14d` / `totalAttributedSales14d` family, `brb_bonus_amount` (Brand Referral Bonus; needs an explicit metric list and groupBy CAMPAIGN or ADGROUP), `kindleEditionNormalizedPagesRead14d`, `kindleEditionNormalizedPagesRoyalties14d`. Products metrics add `brandHalo…14d` and `attributedNewToBrand…14d`. Dates are "YYYYMMDD"; up to 5000 rows a page with `cursorId`. [official] (Amazon Attribution OpenAPI model; Amazon Attribution how-to; both above)
+
+### Q. Amazon Attribution against sponsored ads: the same name, other rules
+
+Q1. RULE. Restatement in Amazon Attribution takes "into account returns or refunds" (P2's page); the sponsored-ads attribution page lists only cancellations within 72 hours and payment failures (A6, A7). [official] (Amazon Attribution how-to; Ad campaign attribution; both above)
+
+Q2. [seat] An Amazon Attribution `attributedSales14d` and a Sponsored Brands `sales14d` share a name and a window but differ in dating (event against click), in the click model (last tag against Amazon's cross-advertiser last-touch ranking), in the masking of low-click rows, and in the treatment of returns. A seller's off-Amazon ROAS and on-Amazon ROAS are two measurements.
+
+### R. The industry standard Amazon's figures are read against (gap fill 2026-11-09)
+
+These are the IAB's and the MRC's words, not Amazon's. They bind an accredited measurer only through an MRC audit; for Amazon's sponsored ads they are the benchmark a buyer applies, and Amazon Ads staff are named among the contributors ("Aly Anzalone, Senior Program Manager, Measurement Standards Compliance, Amazon Ads"; "Celeste Donahoe, Director, Retail Media Measurement, Amazon Ads"; "Amazon" among the company participants). [third party: IAB/MRC] (IAB/MRC Retail Media Measurement Guidelines, January 2024, https://www.iab.com/wp-content/uploads/2024/01/IAB_Retail_Media_Measurement_Guidelines_January2024.pdf, pp. 4–6, read 2026-11-09). Page numbers below are the document's printed page numbers.
+
+R1. GUIDANCE (industry). Attribution windows must be supported, disclosed in advance and consistent: "such windows must be empirically supported, but they must also be logical with regard to campaign objectives and length as well as category sales cycles"; "These windows must be disclosed before campaign execution and measurement and as part of disclosures corresponding to reported results"; "typical attribution windows for digital measurement are often 3, 7, 14, 28, or 30 days"; "a reasonable maximum attribution window should be established with support and applied consistently, as well as disclosed to measurement users"; and, for reconciliation, "consistent granularity of component days (data providing day-level granularity on the basis of when an ad was viewed or clicked relative to when a conversion occurred should be made available to a measurement user by a retail measurement organization) is encouraged". [third party: IAB/MRC] (Guidelines, §3.6 "Attribution windows", p. 27). [seat] Against Amazon: the windows are disclosed per ad product and advertiser type (A3); the day-lag granularity exists for Sponsored Products in v3 (`purchases1d/7d/14d/30d`, E2) and at event level in AMC (O2), and v1 removes the suffixes (E1).
+
+R2. GUIDANCE (industry). Weights and halo must be disclosed: "Attribution approaches should assign value or weight to different exposures or touchpoints, with the process being suited to the campaign objective, supported by evidence, and disclosed transparently"; "At minimum, SKU-level reporting is recommended where relevant …; reporting of any halo products being counted for is encouraged to be transparent"; and, in the metric table, "Attributable/assisted halo sales/revenue/ROAS must provide transparency on methodology, SKUs that are being attributed, online and offline distinction, and deterministic vs. probabilistic outcomes as well as the attribution window and session period." The guideline also warns that "Many attribution approaches may not account for elements such as price, promotion, influencers, PR, competition, weather, economy, and other factors that can influence consumer actions. Therefore, the weights assigned to exposures may not reflect the broader context." [third party: IAB/MRC] (Guidelines, §3.6, pp. 26–27; §5.3 metric table, p. 48)
+
+R3. GUIDANCE (industry). Viewability and attribution: "The MRC requires viewable impressions for attribution of outcomes to ad exposures"; attribution "based in part or in whole on impressions that are non-viewable (including cases in which viewability cannot be measured due to technical limitations such as may occur for CTV) may still be reported. However, they must be properly labeled with a clear disclaimer (these would not be considered fully compliant) alongside fully compliant attribution metrics"; "'strong user interaction' such as a click or tap can serve as a proxy for viewability". The metric table's "Attributable Direct Sales/ ROAS/Conversions" are "Outcomes that can be attributed to a viewable ad based on MRC viewability standards", with "Need to disclose if attribution is based on viewability or extrapolation". [third party: IAB/MRC] (Guidelines, §3.5, p. 25; §5.3, p. 47). [seat] Against Amazon: Sponsored Products and CPC Sponsored Brands credit clicks only (the click proxy); vCPM view credit uses the MRC thresholds but "viewability is inferred based on historical data and placement characteristics" where it cannot be measured (C1, C8), and no Amazon page read labels the inferred share.
+
+R4. GUIDANCE (industry). Forecasts beyond the window: "Marketers will sometimes need to make decisions based on a shorter time period of measured sales. In these cases, it is common for the measurement provider to forecast the full 12-month effect of the campaign. The basis for such forecasts must be disclosed and the methodology for extending shorter period results to 12 months or longer periods should be re-evaluated periodically based on empirical evidence." Time-decay curves are to be "revisited and validated regularly (at least annually)", and post-campaign adstock modelled so a later campaign's sales are not credited to an earlier one. [third party: IAB/MRC] (Guidelines, §3.6, pp. 27–28). [seat] This is the standard Amazon's long-term sales (F6, F8–F11) meets or misses; the accumulated-sales fields are the periodic check.
+
+R5. GUIDANCE (industry). Modelled and extrapolated conversions: measurers are "recommended to disclose … The percentage of attributed outcome metrics (sales, units, visits, etc.) that are extrapolated within each measurement deliverable (e.g., 20% of attributed sales in ROAS calculation are extrapolated) as required by MRC Minimum Standards … and MRC Outcomes and Data Quality Standards". [third party: IAB/MRC] (Guidelines, §3.6 "Modeled attribution for unknown data", p. 29). [seat] Against Amazon: AMC's NTB columns include modeled conversions from 2026-11-02 (N1) and modeled conversions show NULL below ad-line level (O2); no sponsored-ads page read states a modelled share for the standard metrics.
+
+R6. GUIDANCE (industry). Invalid traffic: "All digital measurement for retail media organizations should at least meet MRC GIVT requirements … and are strongly encouraged to also meet SIVT requirements. The MRC requires SIVT filtration for measuring outcomes"; "IVT decision rates should be disclosed as required by MRC and the inability to collect sufficient signals to make IVT determinations … should result in disclosed unknown IVT (not valid or invalid by default)"; crawling and scraping "should be filtered from reported/monetized measurement and disclosed". [third party: IAB/MRC] (Guidelines, §3.3, pp. 23–24). [seat] Against Amazon: SIVT filtration is MRC-accredited for Sponsored Products clicks only (D7, D7a).
+
+R7. GUIDANCE (industry). Incrementality. "Incrementality measures the true value created by any business strategy … In other words, incrementality is the potential causal impact of marketing"; "MRC Standards highly recommends using RCTs, where individuals or units are randomly assigned to a test group or a control group"; an RCT is to separate the treated from the intention-to-treat group ("it may be necessary to utilize intent-to-treat (ITT) analysis methods … or leverage approaches centered around other implementations such as Ghost Ads or PSA methods"); synthetic controls and matched markets are accepted where "the matching variables account for all confounding variables". [third party: IAB/MRC] (Guidelines, Chapter 4, pp. 31–34). The Amazon MTA paper quotes this sentence: the guidelines "'highly recommends using RCTs' to measure advertising incrementality". [research, Amazon-authored] (Lewis, Zettelmeyer, Gordon et al. 2025, arXiv 2508.08209 HTML, read 2026-11-09). The later IAB and IAB Europe guidelines (November 2025) define incrementality against attribution in so many words: "Incrementality differs from attribution and ROAS: those methods show what happened, not whether marketing caused the result", and grade four method families: "Experiment-based" (RCTs, "Hold out / Ghost Ads", matched markets) "Strong"; "Model-based Counterfactual" (synthetic control, ML propensity models) "Strong to Moderate"; "Econometric" (MMM, time-series regression) "Moderate to Weak"; "Hybrid Proxies" ("New-to-brand percentage, baseline vs. exposed analysis, platform-reported incrementality, Simple Multi-Touch Attribution") "Weak", "relying on platform-specific attribution / modeling rather than an independent, true control group". [third party: IAB/IAB Europe] (Guidelines for Incremental Measurement in Commerce Media, November 2025, https://iabeurope.eu/wp-content/uploads/IAB_and_IAB_Europe_Guidelines_Incremental_Measurement_Commerce_Media_November_2025.pdf, pp. 2–3, read 2026-11-09)
+
+R8. GUIDANCE (industry). Disclosure and definitions: new-to-brand "defines a shopper who has not purchased any products from the specified brand within a defined time frame that needs to be disclosed by the retailer"; the incrementality chapter's limitation is that it measures "the direct incremental impact within a retailer's ecosystem, rather than considering the broader 'halo effect' it may have on the rest of the market"; and a measurer's methodology "should be fully described and accessible to users of the data", including "data editing procedures or other types of data adjustment or projection". [third party: IAB/MRC] (Guidelines, §5.3, p. 47; §4.3 "Limitations", p. 40; "Disclosure Guidance", p. 55). [seat] Amazon discloses its NTB look-back (12 months / 365 days, F1); "halo" in the guideline's incrementality chapter means off-retailer effects, a different use of the word from Amazon's same-brand or same-merchant halo (E5).
+
+## What practitioners add
+
+Each item is the named source's claim, stated as theirs. Where a claim contradicts an Amazon page, the contradiction is noted; Amazon's page governs what a skill holds.
+
+1. **Openbridge, "Amazon Advertising Attribution, Definitions, and Timing"** (Openbridge docs, https://docs.openbridge.com/en/articles/4208366-amazon-advertising-attribution-definitions-and-timing, read 2026-11-09). [opinion] Sponsored Products 7 days for sellers and 14 for vendors, Sponsored Brands and Sponsored Display 14 days; sales take up to 48 hours to populate; cancellations within 72 hours are removed; for "Ad types, like lock screen Ads", spend can appear for up to 14 days after a campaign is paused; Openbridge pulls rolling 1, 3, 7, 14, 30 and 59-day windows to catch restatements, and gives "3-5 days" as its settling guide. Agrees with A3, A6, D5.
+
+2. **Openbridge, "Amazon Advertising Console vs Advertising API"** (Openbridge Support, 2024-09-27). [opinion] Gaps under 5% between console and API are normal and larger ones may need Amazon to investigate; retrieve API reports after 72 hours; invalid clicks are mostly removed within a day, some validation takes up to 3 days, and later findings restate click and spend data in the console and future API reports; ROAS was a console-calculated metric not in the API at the time and could be derived. (K10)
+
+3. **Optmyzr, Lakshmi Padmanaban, 2026-04-08** (Optmyzr blog, https://www.optmyzr.com/blog/amazon-ads-reporting-delays/, read 2026-11-09). [opinion] A study of 14,991 Amazon campaigns found that in the top 5% attributed sales "grew by at least 18.75% between Day 1 and Day 17", and impressions shifted "by at least 36.67%"; "one in every 20 campaigns showed discrepancies large enough to entirely change a budget or optimization decision"; Optmyzr treats the last 28 days as provisional because Amazon "is still processing cancellations, returns, payment failures, and traffic validation" — a claim about returns no Amazon sponsored-ads page read makes (A7). The method is published in a companion post by Frederick Vallaeys, "Amazon Ads Reporting Delay Is Killing Your Conversions…" (https://www.optmyzr.com/blog/amazon-ads-reporting-data-delay-study/): 302 accounts across 14 marketplaces, a 14-day window, day 17 taken as final. **(human readers only)**
+
+4. **Ad Badger, "Attribution Window Blind Spots: How 14-Day Windows Quietly Wreck Your Decisions", 2026-08-13** (Ad Badger blog, read 2026-11-09). [opinion] Argues for "maturity gates" before judging recent days. Two of its factual claims contradict Amazon: it describes Sponsored Products and Sponsored Brands as using 14-day click attribution (Amazon: SP sellers 7 days, A3) and reports as "built by order date" (Amazon: interaction date, A1, B1). Its example is hypothetical.
+
+5. **Ad Badger, The PPC Den podcast, Michael Erickson Facchin, "Why Your Amazon Ad Revenue Is Misleading (Same SKU vs Other SKU)", 2026-05-01** (adbadger.com, read 2026-11-09). [opinion] The ad revenue column "lumps in the sales of your other products if a buyer clicks on your ad"; a shopper who "ultimately chooses a different size, color, or a completely different item" moves the sale to other SKU; without the split, product-level ACoS can "look way worse" and lead to "bad optimization decisions". Matches A3 and E5 for sellers.
+
+6. **Helium 10, Vince Montero, 2022-03-24, updated 2022-08-26** (Helium 10 blog, https://www.helium10.com/blog/amazon-attribution-windows/, read 2026-11-09). [opinion] SP 7 days, SB and SD 14; 12-hour appearance; 72-hour cancellations; evaluate SP after 8 days and SB and SD after 15. The evaluation days are Helium 10's practice, not Amazon's.
+
+7. **Intentwise, Sreenath Reddy, 2023-04-13** (Intentwise blog, https://www.intentwise.com/blog/explained-how-does-amazon-attribute-ad-sales, read 2026-11-09). [opinion] Amazon uses last-touch attribution; vendor SP is 14 days; AMC lets an advertiser reach a 28-day view through SQL (O3). Intentwise's glossary (no date, read 2026-11-09) defines ACoS as "Total ad spend ÷ Total sales from ads", CVR as "Total conversions ÷ Total number of clicks" and "Same SKU Ad Revenue" as revenue "only when the product (SKU) that was clicked in an ad is the same as the one purchased" — the click-based conversion rate that unified reporting now calls "Purchase rate over clicks" (E7).
+
+8. **Code3, on the January 2026 view-attribution change** (https://code3.com/resources/amazon-quietly-tightened-attribution-and-its-changing-how-dsp-performance-is-measured/, no author or date, read 2026-11-09). [opinion] Reports "double-digit declines in attributed DSP revenue" after the change, advises using Purchases (all views) as a baseline across the change and comparing against MTA. Not measured by Amazon on any page read.
+
+9. **PPC Land, Luís Rijo, 2026-01-14** (https://ppc.land/amazon-tightens-view-attribution-as-roas-reporting-splits/, read 2026-11-09). [opinion, press] Reports that Amazon did not specify the new view window's length (C3 agrees). It describes the scope as Sponsored Brands, Sponsored Display and DSP without Amazon's vCPM qualifier, which is broader than Amazon's launch page (C3). PPC Land's 2026-11-03 article reports an undated Amazon notice, posted on LinkedIn by a third party, that from 2027-01-01 display impressions for Display, Video, and Audio display ads on the Amazon Store count only once on screen, about 63% fewer; no Amazon page read states it, and it names no sponsored-ads product. (https://ppc.land/amazon-drops-off-screen-ads-from-impression-counts-expects-63-fewer/) **(human readers only)**
+
+10. **Relevant Audience, Antonio Fernandez, 2026-10-02, citing Digiday 2026-10-01** (relevantaudience.com, read 2026-11-09). [opinion, press report of an Amazon statement] Lily Tong, "a director of measurement at Amazon Ads", says "roughly 20% of the sales value for a brand shows up over time" after the attribution window, from Amazon's analysis of more than 2 million campaigns in June; Digiday reported no independent audit. **(human readers only)**
+
+10a. **Digiday, Seb Joseph, "Amazon wants to own the long game in ad measurement", 2026-10-01** (https://digiday.com/media-buying/amazon-wants-to-own-the-long-game-in-ad-measurement/, read directly 2026-11-09; earlier read only through Relevant Audience). [opinion, press; the quoted words are an Amazon staff member's] Verbatim from Lily Tong: "What we actually find is that roughly 20% of the sales value for a brand shows up over time and isn't captured in that immediate attribution window because that's just a cutoff date for measurement"; Digiday's frame: "around a fifth of a brand's value on Amazon arrives after the attribution window, typically seven to 30 days, has closed. It reached that figure by studying more than 2 million campaigns that ran on its ad products in June." The figure is across Amazon's ad products, not broken out for Sponsored Products, and its method is not published. On long-term sales, Digiday reports that agencies "are split": "Some say long-term sales helps brands look past last-touch attribution"; "Others treat it as a forecast that shifts whenever the campaign mix changes, and say the uplift Amazon credits to streaming TV doesn't show up when they check it against actual product sales week by week." Named agency voices: Bryce Hamlin (Power Digital), "If the right data exists we see full visibility but that's a huge 'if'"; Radoslav Mazganov (Hyperzon by Moburst), on off-Amazon spend, "So far, we haven't seen it move the needle." Digiday also reports that Amazon "gives agencies 90 days of historical sales data, while a marketing model needs closer to 18 months", with its MMM data feed offering "up to three years". **(human readers only)** for the figures.
+
+11. **MarketingCharts / Tinuiti figures** surfaced by search (79% of SP conversions within 24 hours; 77–78% in Q2 2020) could not be read; the pages answered 403. Not used. [seat]
+
+## Published research
+
+1. **Qin, "Lengthen your attribution window: which digital ads have the most long-term impact?"**, Applied Marketing Analytics 9(1):73–83, 2023, by an Amazon Ads scientist (https://cdn.amazon.science/79/6e/915822bb4663a28c97c869a75931/lengthen-your-attribution-window-which-digital-ads-have-most-long-term-impact.pdf, read 2026-11-09). [research] Fifteen US brands, weekly data from January 2021 to June 2022, a VARMAX time-series model of ad spend and outcomes. Upper- and mid-funnel products realise 30–50% of their effect within 2 weeks, lower-funnel 60–90%; Sponsored Products shows no carry-over beyond 3 weeks; DSP display about 40% of effect in week 5 and beyond; Sponsored Display about 30% after 5 weeks. Detail page views within 14 days: CPG 70%, hardlines 45%, softlines 30%. Limitation stated: only brands that requested the study were included. **(human readers only)**
+
+2. **Lewis, Zettelmeyer, Gordon et al., "Amazon Ads Multi-Touch Attribution"**, arXiv 2508.08209, 2025-08-11 (read 2026-11-09). [research] Describes the MTA behind M1: randomized experiments calibrate a machine-learning model that uses shopping signals to split credit across touches.
+
+3. **Gordon, Moakler and Zettelmeyer, "Close Enough? A Large-Scale Exploration of Non-Experimental Approaches to Advertising Measurement"**, Marketing Science 42(4):768–793, 2023 (arXiv 2201.07055, read 2026-11-09). [research] 663 Facebook experiments; observational methods (double/debiased machine learning, stratified propensity score matching) miss experimental lift badly: in the arXiv v2 abstract (2022-10-04) the median estimated lift by DML (SPSM) is 83% (173%) upper-funnel, 58% (176%) mid-funnel and 24% (64%) lower-funnel, against median RCT lifts of 29%, 18% and 5%. The journal version's figures were not read (the publisher page answered 403). **(human readers only)**
+
+4. **Gordon, Moakler and Zettelmeyer, "Predicted Incrementality by Experimentation (PIE) for Ad Measurement"**, arXiv 2304.06828, v2 2026-04-01 (read 2026-11-09). [research] 2,226 Meta experiments; a model trained on experiments predicts campaign-level incrementality with R² 0.88 against 0.19 for 7-day last-click attribution; decision disagreement with RCTs 8–12% against 12–20% for last-click. **(human readers only)**
+
+5. **Gordon, Zettelmeyer, Bhargava and Chapsky, "A Comparison of Approaches to Advertising Measurement: Evidence from Big Field Experiments at Facebook"**, Marketing Science 38(2):193–225, 2019 (read 2026-11-09 via abstract). [research] 15 experiments, about 500 million user-experiment observations; common observational methods fail to recover experimental lift.
+
+6. **Berman, "Beyond the Last Touch: Attribution in Online Advertising"**, Marketing Science 37(5):771–792, 2018 (read 2026-11-09 via abstract). [research] Last-touch attribution over-incentivises advertisers' exposure decisions; a Shapley-value scheme raises the advertiser's profit in the model.
+
+7. **Lewis and Rao, "The Unfavorable Economics of Measuring the Returns to Advertising"**, Quarterly Journal of Economics 130(4):1941–1973, 2015 (abstract via search; the RePEc page returned 404). [research] 25 field experiments with about $2.8 million spend; the median confidence interval on ROI is over 100 percentage points wide; measuring ad returns precisely is statistically hard even with experiments.
+
+8. **Chapelle, "Modeling Delayed Feedback in Display Advertising"**, KDD 2014 (metadata only). [research] Conversion labels arrive late relative to clicks; models that treat recent unconverted clicks as negatives are biased. [seat] The same mechanism as B5.
+
+*Added in the gap fill of 2026-11-09. Items 9–13 are the attribution-method literature, 14–15 the long-term-effect literature, 16–17 the evidence on views. Abstracts were read through Crossref, OpenAlex and EconPapers, and the NBER and IDEAS pages; no full text was read unless stated.*
+
+9. **Danaher and van Heerde, "Delusion in Attribution: Caveats in Using Attribution for Multimedia Budget Allocation"**, Journal of Marketing Research 55(5):667–685, 2018 (doi 10.1177/0022243718802845; abstract via Crossref, read 2026-11-09). [research] They define attribution formally as "the relative incremental contribution that each medium makes to a purchase, taking into account advertising carryover and interaction effects", and show that "attribution is proportional to the marginal effectiveness of a medium times its number of exposures. This means that often-used media will have high attribution weights. However, the profit-maximizing allocation for a fixed budget is a function of advertising effectiveness, but not a function of past exposure levels." Their simulated and empirical data show "how attribution can offer misleading insights on how to allocate resources across media", and "substantial gains in purchase probability can be made using profit-maximizing allocation compared with attribution-based allocation". *Transfer:* even an attribution that is causal per touch credits an ad product in proportion to how much of it ran; a seller's budget split read from attributed sales by campaign, ad product or MTA credit inherits that scale effect. [seat]
+
+10. **Li and Kannan, "Attributing Conversions in a Multichannel Online Marketing Environment: An Empirical Model and a Field Experiment"**, Journal of Marketing Research 51(1):40–56, 2014 (doi 10.1509/jmr.13.0050; abstract via Crossref). [research] Individual-level touch data model "consideration of online channels", "visits through these channels over time", and "subsequent purchases", estimating "the carryover and spillover effects of prior touches at both the visit and purchase stages"; channels' relative contributions "are significantly different from those found by other currently used metrics"; "A field study validates the proposed model's ability to estimate the incremental impact of a channel on conversions", and the estimates "help identify cases in which retargeting strategies may actually decrease conversion probabilities". *Transfer:* last-touch credit and a channel's incremental contribution diverge, and the divergence can change sign for retargeting-like touches (Sponsored Display remarketing is the Amazon analogue). [seat]
+
+11. **Anderl, Becker, von Wangenheim and Schumann, "Mapping the customer journey: Lessons learned from graph-based online attribution modeling"**, International Journal of Research in Marketing 33(3):457–474, 2016 (doi 10.1016/j.ijresmar.2016.03.001; abstract via EconPapers). [research] "a novel attribution framework reflecting the sequential nature of customer paths as first- and higher-order Markov walks", applied "to four large customer-level data sets from various industries, each entailing at least seven distinct online channels"; "The results show substantial differences from currently applied heuristics such as last click attribution", with "idiosyncratic channel preferences (carryover) and interaction effects both within and across channel categories (spillover)". *Transfer:* a path model on conversion-path data (Amazon's conversion path report, F7; AMC's event tables, O2) credits touches differently from last touch, but a Markov removal effect is still observational credit, not a holdout. [seat]
+
+12. **Dalessandro, Perlich, Stitelman and Provost, "Causally motivated attribution for online advertising"**, Proceedings of the Sixth International Workshop on Data Mining for Online Advertising (ADKDD 2012), doi 10.1145/2351356.2351363 (abstract via OpenAlex). [research] "we position attribution as a causal estimation problem and then propose two approximation methods as alternatives for when the full causal estimation can not be done. These approximate methods derive from our causal approach and incorporate prior attribution work in cooperative game theory. We argue that in cases where causal assumptions are violated, these approximate methods can be interpreted as variable importance measures." *Transfer:* Shapley-style splits (Berman, item 6; Amazon's MTA, M1) are causal only under the assumptions the authors state; otherwise they rank touches by importance. [seat]
+
+13. **Zantedeschi, Feit and Bradlow, "Measuring Multichannel Advertising Response"**, Management Science 63(8):2706–2728, 2017 (doi 10.1287/mnsc.2016.2451; abstract via Crossref). [research] Observational exposure data are hard to read "because of concerns that advertisers target customers with a higher propensity to buy or increase advertising during periods of peak demand"; a firm "randomly held out customers from each campaign", but "the estimated effect for any individual campaign is imprecise, because of the small size of the holdout"; a hierarchical Bayesian ad-stock model pooled across campaigns estimates "separate decay rates for each advertising medium" and finds "catalogs have substantially longer-lasting impact on customer purchase than emails"; targeting the most responsive customers "increases the predicted returns on advertising by approximately 70% versus traditional recency, frequency, and monetary value–based targeting". *Transfer:* decay differs by medium, which is the same question Qin (item 1) asks across Amazon ad products; single small holdouts are noisy, pooled ones informative. [seat]
+
+14. **Athey, Chetty, Imbens and Kang, "The Surrogate Index: Combining Short-Term Proxies to Estimate Long-Term Treatment Effects More Rapidly and Precisely"**, NBER Working Paper 26463, November 2019, revised August 2024 (doi 10.3386/w26463; NBER page read 2026-11-09; arXiv 1603.09326 v5, 2024-08-22, abstract read). [research] Short-term outcomes are combined into a "surrogate index"; "Under the Prentice surrogacy assumption, which requires that the primary outcome is independent of the treatment conditional on the surrogates, we show that the average treatment effect on the surrogate index equals the treatment effect on the long-term outcome." They "characterize the bias that arises from violations of each of the key assumptions" and "provide simple methods to validate these assumptions using additional observed outcomes". The arXiv version's setting is "two samples, an experimental sample containing data about the treatment indicator and the surrogates and an observational sample containing information about the surrogates and the primary outcome", and it notes "The validity of the surrogacy condition is often controversial." In their job-training application, six quarters of outcomes estimated the nine-year effect with "a 35% reduction in standard errors". *Transfer:* this is the frame for Amazon's long-term sales (F10–F11): the engagements are surrogates, the brand's historical shoppers are the observational sample, and the treatment-to-surrogate step is attributed credit rather than an experimental sample. [seat]
+
+15. **Dekimpe and Hanssens, "The Persistence of Marketing Effects on Sales"**, Marketing Science 14(1):1–21, 1995 (doi 10.1287/mksc.14.1.1; abstract via Crossref). [research] "persistence modeling to derive long-term marketing effectiveness from time-series observations on sales and marketing expenditures": "we use unit-root tests to determine whether sales are stable or evolving (trending) over time. If they are evolving, we examine how strong this evolution is (univariate persistence) and to what extent it can be related to marketing activity (multivariate persistence)"; for a home-improvement chain "some, but not all, advertising has strong trend-setting effects on sales", which "traditional modeling approaches would not pick up". Qin (item 1) applies this persistence method to Amazon brands and reports that "All tests reject the null hypothesis on the presence of a unit root", so all KPIs are stationary, which in Qin's words "implies that no marketing effects last forever" (Qin's own reference is the companion paper, Dekimpe and Hanssens, "Empirical Generalizations about Market Evolution and Stationarity", Marketing Science 14(3):G109–G121, 1995). *Transfer:* on the stationary Amazon series Qin studied, ad effects decay to zero, so a long-term value is a finite sum of decaying effects, not a permanent shift; a 12-month figure such as LTS sums such effects for new-to-brand shoppers. [seat]
+
+16. **Lewis and Reiley, "Online ads and offline sales: measuring the effect of retail advertising via a controlled experiment on Yahoo!"**, Quantitative Marketing and Economics 12(3):235–266, 2014 (abstract at https://ideas.repec.org/a/kap/qmktec/v12y2014i3p235-266.html, read 2026-11-09). [research] "A randomized experiment with 1.6 million customers"; "The advertising profitably increases purchases by 5%. 93% of the increase occurs in brick-and-mortar stores; 78% of the increase derives from consumers who never click the ads." The experiment "provides a specification check for observational difference-in-differences and cross-sectional estimators; the latter exhibits a large negative bias three times the estimated experimental effect." *Transfer:* exposure without a click moved purchases, so click-only credit (Sponsored Products, CPC Sponsored Brands; C2) can miss effects; and observational bias can run in either direction. The setting is display ads for a retailer with mostly offline sales, not Amazon. [seat]
+
+17. **Johnson, Lewis and Nubbemeyer, "Ghost Ads: Improving the Economics of Measuring Online Ad Effectiveness"**, Journal of Marketing Research 54(6):867–884, 2017 (doi 10.1509/jmr.15.0297; abstract via OpenAlex, read 2026-11-09). [research] "To measure the effects of advertising, marketers must know how consumers would behave had they not seen the ads"; Ghost Ads identify "the control-group counterparts of the exposed consumers in a randomized experiment" and "work with modern ad platforms that optimize ad delivery in real-time"; in "an online retailer's display retargeting campaign", "the ads lifted website visits by 17.2% and purchases by 10.5%", measured "just as precisely while spending at least an order of magnitude less" than intent-to-treat or PSA designs. *Transfer:* the lift is measured on exposed users against their randomized ghost counterparts, clicks or not; it is the design the IAB/MRC guidelines name (§R7) and the kind of evidence that would settle whether Amazon's view-attributed purchases are incremental; no Amazon page read offers it for sponsored ads. [seat]
+
+**Carry-over: weighing the three Amazon-relevant sources (gap fill; the owner of the window question is `method/change-measurement.md`).** (a) Qin 2023 [research, Amazon-authored] is the only Amazon-data measurement broken out for Sponsored Products: "SP does not show significant carry-over effects beyond three weeks for any brand", but only two of the fifteen brands carry an SP row in the appendix table, and in those rows week 1 holds 37% and 69% of the DPV effect (48% and 62% of revenue), week 2 27% and 22% (23%, 26%), week 3 36% and 9% (29%, 12%), weeks 4 and beyond 0%. So within three weeks, but with 63% and 31% of the measured SP effect on DPV (52% and 38% on revenue) arriving after week 1, i.e. after a seller's 7-day window (A3); the paper does not say whether its revenue KPI is ad-attributed or total, and the brands are ones "that asked Amazon Ads to quantify ad carry-over effects". (b) Lily Tong's ~20% after the window [Amazon staff via Digiday] pools all Amazon ad products and windows "typically seven to 30 days", with no published method, so it neither confirms nor contradicts (a) for Sponsored Products. (c) Hershaff's daily adstock decay of ~0.35 [opinion, one seller, about a month of daily on/off] implies carry-over measured in days. The non-Amazon studies `method/change-measurement.md` cites (Ju, Zhao & Aral on app-store rank feedback; Hohnhold, O'Brien & Tang on Google user learning; not re-opened by this seat) describe mechanisms that build over weeks or months; Qin's two SP rows show no such build on Amazon data. [seat] Read together: on Amazon data a Sponsored Products effect has been measured to fade within about three weeks, with a material part arriving in weeks 2–3; a reading window or a pause shorter than about three weeks can miss part of it; no Amazon-data study read shows a Sponsored Products effect lasting months.
+
+## Stable enough for a skill
+
+Items 1–17 are Amazon's documented definitions or rules, stable across the pages read; item 18 is the MRC's, item 19 the IAB/MRC's, item 20 published research's. Numbers inside them are read at run time (next section).
+
+1. Sponsored ads date every attributed conversion to the day of the ad interaction, not the day of the sale; Amazon Attribution and KENP read date to the event. (Ad campaign attribution; Overview of the Reporting v1 API; Understand unified reporting; KENP metrics; Amazon Attribution OpenAPI model)
+2. Attribution is per conversion, once, across all advertisers, ranked promoted clicks, halo clicks, promoted views, halo views, then last touch with clicks over views. (Ad campaign attribution)
+3. Lookback, eligible products and interactions differ by ad product and by seller against vendor; seller Sponsored Products halo is "any other products sold by the Seller Central merchant", vendor and brand halo is same-brand products. (Ad campaign attribution)
+4. Sponsored Products and CPC Sponsored Brands count clicks only; vCPM Sponsored Brands, Sponsored Display, Streaming TV and DSP count views too, under the MRC viewability standard; from 2026-01-01 a shopping-signal model governs vCPM view credit and "all views" metrics keep the old model for comparison. (Ad campaign attribution; View attribution updates for Amazon Store ads; Purchases (all views))
+5. Impressions and clicks are net of invalid traffic; gross and invalid counts are separate metrics. (Traffic metrics; Impressions and Clicks (v1))
+6. Unified reporting and v1 have no window suffixes; base metrics use the product's single window and include halo; promoted and halo are breakouts. (Understand unified reporting; Overview of the Reporting v1 API; Conversion metrics)
+7. "Purchase rate" in unified reporting and v1 is per impression; the per-click rate is "Purchase rate over clicks"; older pages and bulksheets use other meanings. (Conversion metrics; Purchases (v1); Search term report for Sponsored Products; Bulksheets search term report)
+8. ACOS is not a unified-reporting or v1 field; it is total cost over sales, the inverse of ROAS; v3 offers click-window ACOS on SP targeting, search-term and advertised-product reports and on the prompt and video ad-extension report types, not on `spCampaigns`. (What is ACOS; Reporting columns (version 3); Bulksheets search term report)
+9. Attributed sales exclude tax and cart discounts and can include items sold by other sellers or used; cancellations within the stated hours are deducted. (Ad campaign attribution)
+10. Traffic metrics do not exist at the converted-product level. (Data availability in unified reporting)
+11. v1 money metrics require the budget-currency dimension and are in the campaign budget currency unless converted; v1 days follow the campaign time zone. (Helpful concepts (v1); Dates and historical data retention (v1))
+12. Figures restate: clicks and cost for days, conversions for weeks, invalid traffic for up to a month. (Reporting v3 FAQ; Amazon Ads reporting overview; Traffic validation; Adjustments to invalid spend)
+13. Bulksheet search-term metrics use a fixed 7-day window for sellers and vendors, and the bulksheet page defines CTR and conversion rate inverted. (Bulksheets search term report)
+14. MRC accreditation of Amazon's Sponsored Products measurement covers clicks (gross and net of SIVT) on Amazon placements, in total, not impressions or conversions. (MRC letter of 2026-07-08; MRC accredited data for Sponsored Products campaigns)
+15. Multi-touch metrics are a separate, modelled family beside last touch, in beta; their API names are v1's `metric.multiTouch*` fields, which map to no v3 column. (Uncover ad value … multi-touch attribution (beta); Purchases (v1); Reporting columns (version 3))
+16. Amazon's NTB help page states that new-to-brand data is not available for Sponsored Products; v3 carries NTB on no `sp*` report type. (Sponsored ads new-to-brand metrics; Reporting columns (version 3))
+17. Long-term sales is an ad-attributed forecast of 12-month value from new-to-brand engagements, machine-learning based in the US since 2026-09-25, with cohort accumulated-sales fields to check the forecast against what the cohort spent. (Long-term sales help page; Long-term sales (v1); release notes of 2025-06-05 and 2026-09-25)
+18. Of Amazon's first-party measurement, only Sponsored Products clicks (gross and net of SIVT, total) and parts of Amazon DSP's impressions, clicks and viewability are MRC-accredited; Sponsored Brands and Sponsored Display are not on the MRC list. (MRC Accredited Services — Digital; MRC letters of 2026-05-29 and 2026-07-08)
+19. The IAB/MRC retail media guidelines ask that attribution windows be empirically supported and disclosed, that view-based attribution use MRC-viewable impressions or carry a disclaimer, that 12-month forecasts disclose their basis and be re-evaluated, and that incrementality be measured preferably by RCTs. (IAB/MRC Retail Media Measurement Guidelines, January 2024)
+20. Attribution credit is not marginal value: published models show credit scaling with exposure volume while the profit-maximising budget depends on effectiveness alone, and observational credit diverging from experimental lift. (Danaher & van Heerde 2018; Li & Kannan 2014; Gordon et al. 2019, 2023)
+
+## What cx must read at run time
+
+- The attribution table (windows, eligible products, interactions) from Ad campaign attribution, GX7KDKHMWQYMJ385; the lookback for each campaign follows from the profile's `accountInfo.type` (seller, vendor, agency) and the ad product.
+- The profile's `timezone` and `currencyCode` from `GET /v2/profiles`, and, for v1, the `budgetCurrency.value` returned on the rows.
+- Each v1 field's page (program eligibility, required fields, incompatible dimensions) before composing a query; the I2 table is a starting point, not the source.
+- The freshness and restatement figures (12 hours, 48 hours, 3 days, 5 days, 28/30/42/60 days) from the Amazon Ads reporting overview, the v3 FAQ, Performance metrics, Traffic validation and Adjustments to invalid spend.
+- The history limits by grain from Dates and historical data retention (v1) and Data availability in unified reporting.
+- The list of report types not yet in unified reporting and the migration dates (Understand unified reporting; Migrate your scheduled reports).
+- The release notes, for new metrics or definition changes since the last read (for example branded-search matching, NTB modelled conversions in AMC).
+- For a KDP author: whether KENP metrics exist on the account (they appear only with a KENP read).
+- For long-term sales (US; Sponsored Brands, Sponsored Display, Sponsored TV, DSP): the v1 fields `metric.longTermSales`, `metric.longTermSales30d/60d/90d/180d`, the cohort estimates `metric.cohortLongTermSales30d/60d/90d/180d` and `metric.cohortLongTermSales`, and the actuals `metric.cohortAccumulatedSales30d/60d/90d/180d/365d`, the cohort fields with `campaign.id`, `longTermSalesCohort.value` and `budgetCurrency.value` and never summed across cohorts; each field page for eligibility; the release notes for later changes to the LTS model.
+- For multi-touch: the v1 `metric.multiTouch*` fields with their required base metric (`metric.purchases`, `metric.roas`, …); whether they return values for the account is read from the report (the launch page's v3 claim has no column on the v3 list).
+- For new-to-brand on a seller's Sponsored Products: the NTB help page (G5EWB37XK4RS4J4T) for the current rule, before any v1 NTB value on an SP row is read.
+- For accreditation: the MRC's digital accreditation list and Amazon's current letters (https://www.mediaratingcouncil.org/accreditation/digital), whose scopes are renewed and change; the console's MRC filter for Sponsored Products.
+- For Amazon Attribution: whether the advertiser's ad console account carries Amazon Attribution; a profile with `subType` `AMAZON_ATTRIBUTION` is a legacy profile Amazon tells new integrators to filter out (Amazon Attribution get started).
+
+## Pages read
+
+All read 2026-11-09 by plain GET. Help pages under `https://advertising.amazon.com/help/` serve their body inside the page's HTML; API guides were read as markdown on `https://d3a0d0y2hgofx6.cloudfront.net/en-us/`, abbreviated `CF/` below.
+
+| Page name | URL | Type | Date read | What it gave |
+|---|---|---|---|---|
+| Ad campaign attribution | https://advertising.amazon.com/help/GX7KDKHMWQYMJ385 | official help | 2026-11-09 | Windows by product and advertiser, ranking, one credit, view rules, retail differences, KDP notes |
+| Traffic validation | https://advertising.amazon.com/help/GBYPYH79NGJJ5JPS | official help | 2026-11-09 | IVT definition, filtering, 30-day monitoring, IVR reading |
+| Adjustments to invalid spend | https://advertising.amazon.com/help/GWQ7F2EL4E2Y7H4B | official help | 2026-11-09 | 6-hour detection, wait 5 days, 30 days, invoice adjustment |
+| Understand unified reporting | https://advertising.amazon.com/help/GMH8A8AJSH4ATV6T | official help | 2026-11-09 | Shutdown date, base windows, traffic date, renames, gaps |
+| Data availability in unified reporting | https://advertising.amazon.com/help/G8A5Z6UD9ME5W3GZ | official help | 2026-11-09 | Ranges by grain, limited-history start dates, traffic metrics not at converted product |
+| Migrate your scheduled reports to unified reporting | https://advertising.amazon.com/help/GK3JZZE5QLTF8U8A | official help | 2026-11-09 | Migration dates, unsupported reports, dropped fields, TOS IS campaign-only |
+| Traffic metrics | https://advertising.amazon.com/help/GMCTGJGY2M438E2R | official help | 2026-11-09 | Impressions, clicks, CTR, CPC, CPM, vCPM, viewable rate, cost, IS, benchmarks |
+| Conversion metrics | https://advertising.amazon.com/help/G7FMJZAJLTR3DKK2 | official help | 2026-11-09 | Purchases, sales, promoted/halo, purchase rate per impression, ROAS, NTB, DPV, ATC, KENP |
+| Engagement metrics | https://advertising.amazon.com/help/GH4WXV3ZBQE37R6M | official help | 2026-11-09 | Video quartiles, interactive, notification metrics |
+| Performance metrics | https://advertising.amazon.com/help/GG44RFW942U9F6F5 | official help | 2026-11-09 | Hourly updates, 48h sales, cost estimate 3 days, DPV not unique, TOS IS 90 days, vCPM, video |
+| Return on ad spend (ROAS) | https://advertising.amazon.com/help/G78DVL9PAZZ83KQX | official help | 2026-11-09 | ROAS as ratio; SP 48h, SB 12h |
+| Sponsored ads new-to-brand metrics | https://advertising.amazon.com/help/G5EWB37XK4RS4J4T | official help | 2026-11-09 | NTB definition, not for SP, start dates, legacy SB values |
+| KENP metrics | https://advertising.amazon.com/help/GSJCXUYD4UGQP2MX | official help | 2026-11-09 | 14-day attribution, read-date dating, offline week, royalty estimate |
+| Campaign success metrics | https://advertising.amazon.com/help/G9EMRNUYZPP58HLX | official help | 2026-11-09 | Totals include halo; recommendation to use totals |
+| Data availability and freshness (DSP) | https://advertising.amazon.com/help/GLN6KZCS23ESMTUS | official help | 2026-11-09 | Lags, IVT 72h, ASIN restatement 28 days, reach 96h |
+| Downloadable reports in sponsored ads | https://advertising.amazon.com/help/GBYSPTSLR337JMLH | official help | 2026-11-09 | Report list, canceled purchases excluded, account time zone |
+| Search term report for Sponsored Products | https://advertising.amazon.com/help/G3HEFZYWZF84NPS9 | official help | 2026-11-09 | Click-only terms, inferred terms, 65 days, purchase rate per click |
+| Campaign report for Sponsored Products | https://advertising.amazon.com/help/G4KJE82E7LDKAHZX | official help | 2026-11-09 | Daily FX conversion, 5-year lookback, seller 7d / vendor 14d rows |
+| Purchased product report for Sponsored Products | https://advertising.amazon.com/help/G9JR57B7AEWBQ3P2 | official help | 2026-11-09 | Sellers only, 60 days, totals differ, label error |
+| MRC accredited data for Sponsored Products campaigns | https://advertising.amazon.com/help/GVLUHLCD2Y6C364M | official help | 2026-11-09 | MRC filter setting |
+| Conversion path reporting | https://advertising.amazon.com/help/GV7Y3HZ5XUGZKWUK | official help | 2026-11-09 | 30-day path, brand owners |
+| Long-term sales | https://advertising.amazon.com/help/GG8E72AW6ASSR6SA | official help | 2026-11-09 | Requirements, marketplaces, not SP, ROE example |
+| Amazon's conversion attribution | https://advertising.amazon.com/help/GJKPFEUSF9236MMW | official help | 2026-11-09 | Empty body (see Unreachable) |
+| What is ACOS | https://advertising.amazon.com/library/guides/acos-advertising-cost-of-sales | official library | 2026-11-09 | ACOS formula, inverse of ROAS |
+| Viewability guide | https://advertising.amazon.com/library/guides/viewability | official library | 2026-11-09 | MRC display and video thresholds, 30% for large desktop |
+| View Attribution Updates for Amazon Store ads (dated January 01, 2026) | https://advertising.amazon.com/resources/whats-new/view-attribution-updates-for-amazon-store-ads | official what's new | 2026-11-09 | 2026-01-01 model, scope, all-views comparison |
+| Uncover ad value across the funnel with multi-touch attribution (beta) | https://advertising.amazon.com/resources/whats-new/uncover-ad-value-across-funnel-with-multi-touch-attribution | official what's new | 2026-11-09 | MTA beta, metrics, SP via API |
+| Amazon Attribution product page | https://advertising.amazon.com/solutions/products/amazon-attribution | official product | 2026-11-09 | Free, 14 days, eligibility |
+| Overview of the Reporting v1 API | CF/guides/reporting/ads-v1/overview.md | official API guide | 2026-11-09 | Traffic date, halo rename, no suffixes, views 0 for CPC, concurrency |
+| Amazon Ads reporting overview | CF/guides/reporting/overview.md | official API guide | 2026-11-09 | Surfaces, shutoff dates, freshness, 60-day changes, concurrency |
+| Reporting v3 FAQ | CF/guides/reporting/v3/faq.md | official API guide | 2026-11-09 | Restatement 1/7/28, 42 days, search-term coverage, SB preview |
+| Reporting columns (version 3) | CF/guides/reporting/v3/columns.md | official API guide | 2026-11-09 | 740 v3 columns and descriptions |
+| Dates and historical data retention (v1) | CF/guides/reporting/ads-v1/dates.md | official API guide | 2026-11-09 | Campaign time zone, inclusive periods, retention and max date range by grain |
+| Helpful concepts (v1) | CF/guides/reporting/ads-v1/helpful-concepts.md | official API guide | 2026-11-09 | Time dimensions, required fields, nulls, account keys |
+| v1 metric library | CF/guides/reporting/ads-v1/metrics/library.md | official API guide | 2026-11-09 | Index of v1 metric families |
+| v1 dimension library | CF/guides/reporting/ads-v1/dimensions/library.md | official API guide | 2026-11-09 | Dimensions; no `timeZone.value` found |
+| Impressions (Delivery) | CF/guides/reporting/ads-v1/metrics/delivery/impressions.md | official API guide | 2026-11-09 | Impression fields, invalid, CPM |
+| Clicks (Delivery) | CF/guides/reporting/ads-v1/metrics/delivery/clicks.md | official API guide | 2026-11-09 | Click fields, CTR, CPC, vCTR, incompatibilities |
+| Viewability (Delivery) | CF/guides/reporting/ads-v1/metrics/delivery/viewability.md | official API guide | 2026-11-09 | Viewable fields, eligibility excludes SP |
+| Costs (Costs and fees) | CF/guides/reporting/ads-v1/metrics/costs-and-fees/costs.md | official API guide | 2026-11-09 | totalCost, reconciled, supply, reward |
+| Purchases (Amazon retail conversions) (v1) | CF/guides/reporting/ads-v1/metrics/amazon-retail-conversions/purchases.md | official API guide | 2026-11-09 | 148 purchase fields including 56 multi-touch fields, rates, ROAS, v3 names |
+| Purchases (all views) | CF/guides/reporting/ads-v1/metrics/amazon-retail-conversions/purchases-all-views.md | official API guide | 2026-11-09 | 54 all-views fields |
+| Purchases (new to brand) (Amazon retail conversions) (v1) | CF/guides/reporting/ads-v1/metrics/amazon-retail-conversions/purchases-new-to-brand.md | official API guide | 2026-11-09 | 102 NTB fields, SP listed |
+| Detail page views (new to brand) (v1) | CF/guides/reporting/ads-v1/metrics/amazon-retail-conversions/detail-page-views-new-to-brand.md | official API guide | 2026-11-09 | NTB DPV fields |
+| Detail page views (v1) | CF/guides/reporting/ads-v1/metrics/amazon-retail-conversions/detail-page-views.md | official API guide | 2026-11-09 | 35 DPV fields, no SP |
+| Add to cart (v1) | CF/guides/reporting/ads-v1/metrics/amazon-retail-conversions/add-to-cart.md | official API guide | 2026-11-09 | ATC fields, no SP |
+| KENP / Kindle Unlimited (v1) | CF/guides/reporting/ads-v1/metrics/amazon-device-conversions/kindle-edition-normalized-pages-read.md and …/kindle-unlimited-qualified-borrows.md | official API guide | 2026-11-09 | KENP and KU fields, mapping error |
+| Attention (video ad) (Engagement) (v1) | CF/guides/reporting/ads-v1/metrics/engagement/attention-video-ad.md | official API guide | 2026-11-09 | Video fields |
+| Long-term sales (v1) | CF/guides/reporting/ads-v1/metrics/amazon-retail-conversions/long-term-sales.md | official API guide | 2026-11-09 | LTS fields and eligibility |
+| Search (Impression share) | CF/guides/reporting/ads-v1/metrics/impression-share/search.md | official API guide | 2026-11-09 | IS, IS rank, TOS IS |
+| Search term (dimension) | CF/guides/reporting/ads-v1/dimensions/targeting/search-term.md | official API guide | 2026-11-09 | `searchTerm.value` |
+| Currency, date, placement, ad product, product relevance dimensions | CF/guides/reporting/ads-v1/dimensions/ (family pages) | official API guide | 2026-11-09 | Dimension IDs and v3 names |
+| SP search term migration example | CF/guides/reporting/ads-v1/report-types/sp-search-term.md | official API guide | 2026-11-09 | v3-to-v1 field mapping by report |
+| SP targeting migration example | CF/guides/reporting/ads-v1/report-types/sp-targeting.md | official API guide | 2026-11-09 | As above |
+| SP advertised product migration example | CF/guides/reporting/ads-v1/report-types/sp-advertised-product.md | official API guide | 2026-11-09 | As above |
+| Amazon Ads advanced tools release notes | CF/release-notes/index.md | official release notes | 2026-11-09 | GA 2026-09-30, Stream re-delivery, NTB, branded searches, KDP, LTS 2025-06-05, v1 warnings 2026-10-05, retention 2026-08-26 |
+| Bulksheets search term report | CF/no-code-tools/bulksheets/bulksheets-search-term-report.md | official API guide | 2026-11-09 | Fixed 7 days, inverted CTR and CVR, ACOS |
+| Profiles | CF/guides/account-management/authorization/profiles.md | official API guide | 2026-11-09 | Profile timezone and currency |
+| Retrieve profiles | CF/guides/get-started/retrieve-profiles.md | official API guide | 2026-11-09 | Profile example fields |
+| API contracts index | CF/contracts.json | official API index | 2026-11-09 | OpenAPI model URLs |
+| OfflineReport OpenAPI model | https://d1y2lf8k3vrkfu.cloudfront.net/openapi/en-us/dest/OfflineReport_prod_3p.json | official OpenAPI | 2026-11-09 | v3 request shape, timeUnit |
+| Merged OpenAPI model | https://d1y2lf8k3vrkfu.cloudfront.net/openapi/en-us/dest/AmazonAdsAPIALLMerged_prod_3p.json | official OpenAPI | 2026-11-09 | v1 ReportCreate, warnings, formats, currency, `accessRequestedAccounts` maxItems 2000 |
+| Merged OpenAPI model (all) and AmazonAdsAPIALL | https://d1y2lf8k3vrkfu.cloudfront.net/openapi/en-us/dest/AmazonAdsAPIALLMerged_prod_3p_ALL.json | official OpenAPI | 2026-11-09 | TimeZone enums |
+| Amazon Attribution OpenAPI model | https://dtrnk0o2zy01c.cloudfront.net/openapi/en-us/dest/AmazonAttribution_prod_3p.json | official OpenAPI | 2026-11-09 | Event-date reporting, 14-day last click, metrics |
+| Amazon Attribution overview | CF/guides/amazon-attribution/overview.md | official API guide | 2026-11-09 | Eligibility, marketplaces |
+| Amazon Attribution get started | CF/guides/amazon-attribution/get-started.md | official API guide | 2026-11-09 | Legacy `AMAZON_ATTRIBUTION` profile subType, console-account access |
+| Amazon Attribution how-to | CF/guides/amazon-attribution/how-to.md | official API guide | 2026-11-09 | Last-clicked tag, delays, restatement, returns, masking, report types |
+| How does Amazon Marketing Cloud work | CF/guides/amazon-marketing-cloud/how-amc-works.md | official API guide | 2026-11-09 | Clean room, no raw export, thresholds |
+| Overview - Amazon Marketing Cloud | CF/guides/amazon-marketing-cloud/overview.md | official API guide | 2026-11-09 | Self-service instance, marketplaces |
+| Get access to AMC using self-service APIs | CF/guides/amazon-marketing-cloud/get-started/instant-access-to-amc.md | official API guide | 2026-11-09 | `GET /amc/instances`, header, terms per region, single-advertiser scope |
+| AMC sponsored ads traffic table | CF/guides/amazon-marketing-cloud/datasources/sponsored_ads_traffic.md | official API guide | 2026-11-09 | Record per impression and click |
+| AMC conversions table | CF/guides/amazon-marketing-cloud/datasources/conversions.md | official API guide | 2026-11-09 | 28-day ad-attributed flag |
+| Amazon attributed events by conversion time | CF/guides/amazon-marketing-cloud/datasources/amazon_attributed_events_by_conversion_time.md | official API guide | 2026-11-09 | Two datings, window not changeable |
+| AMC data retention; Aggregation threshold | CF/guides/amazon-marketing-cloud/datasources/data-retention.md; CF/guides/amazon-marketing-cloud/aggregation-threshold.md | official API guide | 2026-11-09 | 13 months to 5 years by data view, June 2023 backfill, threshold levels |
+| Amazon Marketing Cloud now supports custom attribution analysis | https://advertising.amazon.com/resources/whats-new/amazon-marketing-cloud-custom-attribution | official what's new | 2026-11-09 | 2022-03-25; lookback up to 28 days in AMC queries; first/last/equal/position models |
+| SP-API Finances model | https://raw.githubusercontent.com/amzn/selling-partner-api-models/main/models/finances-api-model/financesV0.json | official model | 2026-11-09 | ProductAdsPaymentEvent |
+| Data Kiosk economics schema | https://raw.githubusercontent.com/amzn/selling-partner-api-models/main/schemas/data-kiosk/analytics_economics_2024_03_15.graphql | official model | 2026-11-09 | AdSummary |
+| MRC accreditation letter, Amazon Sponsored Products | https://www.mediaratingcouncil.org/sites/default/files/Accreditation%20Letters/Amazon/GI070826_Amazon%20Sponsored%20Products%20Accreditation%20Letter.pdf | third party: MRC | 2026-11-09 | Accredited statistics and scope |
+| cx-amazon-ads SKILL.md and references/door.md | local copy of the open-door plugin | cx internal | 2026-11-09 | Door shape, origins, profiles call |
+| cx-listing-playbook research, paid-keywords | local copy | cx internal | 2026-11-09 | Earlier record of bulksheet inversion |
+| cx-listing-playbook research, help-page-access | local copy | cx internal | 2026-11-09 | Seller Central help is JS shell plus forbidden POST |
+| Openbridge, Amazon Advertising Attribution, Definitions, and Timing | https://docs.openbridge.com/en/articles/4208366-amazon-advertising-attribution-definitions-and-timing | practitioner | 2026-11-09 | Windows, settling, rolling pulls |
+| Openbridge, Amazon Advertising Console vs Advertising API | https://docs.openbridge.com/en/articles/5575078-amazon-advertising-console-vs-advertising-api | practitioner | 2026-11-09 | 5% gap, 72 hours, console-only reports |
+| Optmyzr, Lakshmi Padmanaban | https://www.optmyzr.com/blog/amazon-ads-reporting-delays/ | practitioner | 2026-11-09 | Late-data growth figures |
+| Optmyzr, Frederick Vallaeys, reporting-delay study | https://www.optmyzr.com/blog/amazon-ads-reporting-data-delay-study/ | practitioner | 2026-11-09 | Study method: 302 accounts, 14 marketplaces, day 17 final |
+| Ad Badger, Attribution Window Blind Spots | https://www.adbadger.com/blog/attribution-window-blind-spots-how-14-day-windows-quietly-wreck-your-decisions/ | practitioner | 2026-11-09 | Maturity gates; contradicted claims |
+| Ad Badger PPC Den, Same SKU vs Other SKU | https://www.adbadger.com/blog/amazon-ppc-education/why-your-amazon-ad-revenue-is-misleading-same-sku-vs-other-sku-the-ppc-den-podcast/ | practitioner | 2026-11-09 | Other-SKU reading |
+| Helium 10, Vince Montero | https://www.helium10.com/blog/amazon-attribution-windows/ | practitioner | 2026-11-09 | Windows, evaluation days |
+| Intentwise, Sreenath Reddy | https://www.intentwise.com/blog/explained-how-does-amazon-attribute-ad-sales | practitioner | 2026-11-09 | Last touch, AMC 28 days |
+| Intentwise Amazon ads glossary | https://help.intentwise.com/amazon-ads-glossary | practitioner | 2026-11-09 | ACoS, CVR, same-SKU definitions |
+| Code3, view-attribution change | https://code3.com/resources/amazon-quietly-tightened-attribution-and-its-changing-how-dsp-performance-is-measured/ | practitioner | 2026-11-09 | DSP revenue declines, all-views baseline |
+| PPC Land, Luís Rijo | https://ppc.land/amazon-tightens-view-attribution-as-roas-reporting-splits/ | press | 2026-11-09 | Window unspecified; scope stated more broadly than Amazon's |
+| PPC Land, Luís Rijo, impression counting | https://ppc.land/amazon-drops-off-screen-ads-from-impression-counts-expects-63-fewer/ | press | 2026-11-09 | Reported 2027-01-01 DVA Store display count-on-screen change |
+| Relevant Audience, Antonio Fernandez | https://www.relevantaudience.com/ecommerce-marketing/amazon-ads-20-percent-sales-after-attribution-window/ | press | 2026-11-09 | 20% after window, Lily Tong |
+| Qin 2023, Lengthen your attribution window | https://cdn.amazon.science/79/6e/915822bb4663a28c97c869a75931/lengthen-your-attribution-window-which-digital-ads-have-most-long-term-impact.pdf | research | 2026-11-09 | Carry-over by ad product and category |
+| Lewis, Zettelmeyer, Gordon et al., Amazon Ads MTA | https://arxiv.org/abs/2508.08209 | research | 2026-11-09 | MTA method |
+| Gordon, Moakler, Zettelmeyer 2023 | https://arxiv.org/abs/2201.07055 | research | 2026-11-09 | Observational lift estimates against RCT |
+| Gordon, Moakler, Zettelmeyer, PIE | https://arxiv.org/abs/2304.06828 | research | 2026-11-09 | PIE against last-click |
+| Gordon, Zettelmeyer, Bhargava, Chapsky 2019 | Marketing Science 38(2) abstract | research | 2026-11-09 | 15 Facebook RCTs |
+| Berman 2018 | Marketing Science 37(5) abstract | research | 2026-11-09 | Last-touch incentives, Shapley |
+| Lewis and Rao 2015 | QJE 130(4) abstract via search | research | 2026-11-09 | 25 experiments, ROI interval width |
+| Chapelle 2014 | KDD 2014 metadata via search | research | 2026-11-09 | Delayed conversion feedback |
+| *Gap fill 2026-11-09:* IAB/MRC Retail Media Measurement Guidelines (January 2024) | https://www.iab.com/wp-content/uploads/2024/01/IAB_Retail_Media_Measurement_Guidelines_January2024.pdf | third party: IAB/MRC | 2026-11-09 | Windows, weights, halo, viewability for outcomes, 12-month forecasts, extrapolation, IVT, RCTs; Amazon staff among contributors |
+| IAB and IAB Europe, Guidelines for Incremental Measurement in Commerce Media (November 2025) | https://iabeurope.eu/wp-content/uploads/IAB_and_IAB_Europe_Guidelines_Incremental_Measurement_Commerce_Media_November_2025.pdf (landing page https://iabeurope.eu/knowledge_hub/guidelines-for-incremental-measurement-in-commerce-media/) | third party: IAB/IAB Europe | 2026-11-09 | Incrementality against attribution; four method families graded by causal strength |
+| MRC Accredited Services — Digital | https://www.mediaratingcouncil.org/accreditation/digital | third party: MRC | 2026-11-09 | Amazon services on the list (SP clicks and SIVT; DSP); no SB or SD |
+| MRC accreditation letter, Amazon DSP (May 29, 2026) | https://www.mediaratingcouncil.org/sites/default/files/Accreditation%20Letters/Amazon/GI052926_Amazon%20DSP%20Accreditation%20Letter.pdf | third party: MRC | 2026-11-09 | Continuing and initial DSP accreditation by supply source; GIVT basis |
+| MRC accreditation letter, Amazon Sponsored Products (July 8, 2026), re-opened | as above (D7) | third party: MRC | 2026-11-09 | Re-checked scope; TAG addendum (domain and data-centre IP filtering) |
+| Sponsored ads new-to-brand metrics, re-opened | https://advertising.amazon.com/help/G5EWB37XK4RS4J4T | official help | 2026-11-09 | "available for Sponsored Brands and Display"; "isn't available for Sponsored Products"; lastModified 2026-08-19 |
+| Ad campaign attribution, re-opened | https://advertising.amazon.com/help/GX7KDKHMWQYMJ385 | official help | 2026-11-09 | NTB "subset of attributed purchases", 365 days; inferred viewability |
+| Long-term sales, re-opened | https://advertising.amazon.com/help/GG8E72AW6ASSR6SA | official help | 2026-11-09 | ROE heuristic text, "incremental return on engagement", worked example; lastModified 2026-08-26 |
+| Long-term sales (Amazon retail conversions) (v1), re-opened | CF/guides/reporting/ads-v1/metrics/amazon-retail-conversions/long-term-sales.md | official API guide | 2026-11-09 | All 24 fields with descriptions, required fields, eligibility; cohort fields |
+| Amazon Ads advanced tools release notes, re-opened | CF/release-notes/index.md | official release notes | 2026-11-09 | 2026-09-25 LTS machine-learning and accumulated-sales entry; 2025-06-05 LTS entry; no multi-touch entry |
+| Purchases (Amazon retail conversions) (v1), re-opened | CF/guides/reporting/ads-v1/metrics/amazon-retail-conversions/purchases.md | official API guide | 2026-11-09 | 56 multi-touch fields, all with eligibility and v3 names "N/A"; required base metrics |
+| Purchases (new to brand) (v1), re-opened | CF/guides/reporting/ads-v1/metrics/amazon-retail-conversions/purchases-new-to-brand.md | official API guide | 2026-11-09 | SP named on 90 of 102 NTB fields, including FromViews fields |
+| Reporting columns (version 3), re-opened | CF/guides/reporting/v3/columns.md | official API guide | 2026-11-09 | No multi-touch column (0 matches) |
+| Amazon Marketing Stream data guide; Stream v2 overview | CF/guides/amazon-marketing-stream/data-guide.md; CF/guides/amazon-marketing-stream/v2/overview.md | official API guide | 2026-11-09 | No multi-touch field found |
+| Uncover ad value across the funnel with multi-touch attribution (beta), re-opened | as above (M1) | official what's new | 2026-11-09 | The two surface sentences verbatim; dates in metadata |
+| View Attribution Updates for Amazon Store ads, re-opened | as above (C3) | official what's new | 2026-11-09 | "based on our continuous research on shopping journeys"; scope list |
+| Digiday, Seb Joseph, Amazon wants to own the long game in ad measurement | https://digiday.com/media-buying/amazon-wants-to-own-the-long-game-in-ad-measurement/ | press (Amazon staff quoted) | 2026-11-09 | Tong's 20% verbatim; agencies split on LTS; data access |
+| Hershaff, Estimating Amazon Ad Incrementality for Sellers, re-opened | https://whatstheimpact.substack.com/p/estimating-amazon-ad-incrementality | practitioner (seller-analyst) | 2026-11-09 | Adstock decay ~0.35 per day; ROAS figures |
+| Qin 2023, re-opened | as above | research (Amazon-authored) | 2026-11-09 | SP rows of Table 1 (Brands A and D) by week; persistence-model method; stationarity |
+| Amazon Ads Multi-Touch Attribution, HTML, re-opened | https://arxiv.org/html/2508.08209v1 | research (Amazon-authored) | 2026-11-09 | Quotation of the IAB/MRC RCT sentence |
+| Danaher & van Heerde 2018 | Crossref record, doi 10.1177/0022243718802845 (OpenAlex record also read) | research (abstract) | 2026-11-09 | Attribution proportional to effectiveness × exposures; allocation |
+| Li & Kannan 2014 | Crossref record, doi 10.1509/jmr.13.0050 | research (abstract) | 2026-11-09 | Carryover/spillover attribution, field validation, retargeting |
+| Anderl, Becker, von Wangenheim & Schumann 2016 | https://econpapers.repec.org/RePEc:eee:ijrema:v:33:y:2016:i:3:p:457-474 | research (abstract) | 2026-11-09 | Markov-walk attribution on four data sets |
+| Dalessandro, Perlich, Stitelman & Provost 2012 | OpenAlex record, doi 10.1145/2351356.2351363 | research (abstract) | 2026-11-09 | Attribution as causal estimation; approximations as variable importance |
+| Zantedeschi, Feit & Bradlow 2017 | Crossref record, doi 10.1287/mnsc.2016.2451 | research (abstract) | 2026-11-09 | Holdouts pooled by hierarchical Bayes; decay by medium |
+| Athey, Chetty, Imbens & Kang, The Surrogate Index (NBER w26463) | https://www.nber.org/papers/w26463 ; arXiv API record for 1603.09326 | research (abstract) | 2026-11-09 | Prentice surrogacy; bias; validation; 35% smaller standard errors |
+| Dekimpe & Hanssens 1995 | Crossref and OpenAlex records, doi 10.1287/mksc.14.1.1 | research (abstract) | 2026-11-09 | Persistence modelling; unit roots |
+| Lewis & Reiley 2014 | https://ideas.repec.org/a/kap/qmktec/v12y2014i3p235-266.html | research (abstract) | 2026-11-09 | 78% of lift from non-clickers; observational bias negative |
+| Johnson, Lewis & Nubbemeyer 2017, Ghost Ads | OpenAlex record, doi 10.1509/jmr.15.0297 | research (abstract) | 2026-11-09 | Ghost-ad RCT design; retargeting lift 17.2% visits, 10.5% purchases |
+| Gordon, Zettelmeyer, Bhargava & Chapsky 2019, re-opened | OpenAlex record, doi 10.1287/mksc.2018.1135 | research (abstract) | 2026-11-09 | "Observational methods often fail to accurately recover the treatment effects" |
+
+Every row now carries the exact path read; the paths the research seat left as directories were found again and filled in by the verifier.
+
+### Unreachable
+
+- Seller Central help pages (sellercentral.amazon.com/help/hub/…): a plain GET returns a JavaScript shell, and the content call is the POST `.../mons-api/GetHelpTopic`, which the brief forbids. Not read.
+- Amazon's conversion attribution, https://advertising.amazon.com/help/GJKPFEUSF9236MMW: the page's embedded content was empty.
+- Search Engine Land article on latent orders: 403 / Cloudflare challenge.
+- MarketingCharts (Tinuiti SP conversion-timing figures): 403.
+- INFORMS PubsOnline (Marketing Science full texts): 403; abstracts read elsewhere.
+- IDEAS/RePEc page for Lewis and Rao 2015: 404.
+- Intentwise blog URL with `?p=28588`: served the blog index, not the article.
+- Amazon Ads API hosts, Login with Amazon, the Amazon Ads MCP server and SP-API hosts: not called, by the brief's hard limits. No seller data was read.
+- Web search budget ran out late in the session; no further practitioner pages (for example Scale Insights on data delay) were searched.
+- Gap fill 2026-11-09: the Dstillery copy of Dalessandro et al. (https://dstillery.com/wp-content/uploads/2016/07/CAUSALLY-MOTIVATED-ATTRIBUTION.pdf) answered 404 and the Semantic Scholar page 403; the abstract was read from OpenAlex. Semantic Scholar's API had no record for Anderl et al.; the abstract was read from EconPapers. Full texts of the journal papers in Published research items 9–17 were not read (publisher paywalls); claims are taken from abstracts only. The local `pdftotext` failed on a missing library; PDFs were read with PyMuPDF from the network stream, nothing saved.
+
+## Open questions
+
+1. Does unified reporting or v1 date KENP read to the click day (as "all conversion metrics" suggests) or to the reading day (as the KENP help page of 2023-03-30 says)?
+2. *Settled 2026-11-09 (F2a):* Amazon's NTB help page says new-to-brand data is not available for Sponsored Products, v3 agrees, and v1's eligibility list names SP even on view-attributed NTB fields SP cannot populate. What stays open is only whether a v1 query returns empty or zero for those fields on an SP row, which a report shows.
+3. Which denominator does a console Sponsored Products search-term download's "Purchase rate" column use today: clicks (its help page) or impressions (unified reporting)?
+4. Does the console search-term report still include impression-only terms (v3 FAQ) or click-only terms (unified reporting and the SP search-term help page)?
+5. Are returns after the 72-hour cancellation window ever removed from sponsored-ads attributed sales?
+6. What is the length of the shortened view window under the 2026 shopping-signal model?
+7. Is v1 concurrency per client ID (v1 overview) or per advertiser account (reporting overview and the 2026-09-30 release note)? And is the v1 account count per request 5 (Helpful concepts) or 2000 (the OpenAPI `maxItems`)?
+8. Is `timeZone.value` a v1 dimension a query can request (Dates and historical data retention (v1) names it; the dimension library list did not show it)?
+9. Does a downloaded bulksheet's computed CTR and conversion-rate column follow the inverted definitions on the bulksheet page, or only the text?
+10. When v3 stops on 2027-06-30, does any surface keep a 1-day, 14-day-for-sellers or 30-day Sponsored Products figure?
+11. (Gap fill.) Does a seller's Sponsored Products row in v1 return `metric.multiTouch*` values? The launch page's v3 sentence names "most report types" of v3, but the v3 column list has no multi-touch column and v1's fields map to v3 "N/A".
+12. (Gap fill.) What share of view-attributed conversions under the 2026 model rests on inferred rather than measured viewability? The IAB/MRC guidelines ask for such attribution to be labelled; no Amazon page read gives the share.
+13. (Gap fill.) Has Amazon published the accuracy of the machine-learning LTS against `metric.cohortAccumulatedSales*`, for example the ratio of estimate to actual by horizon? No page read gives it.
+14. (Gap fill.) Are Sponsored Brands or Sponsored Display impressions, viewable impressions or clicks in an MRC audit? They are not on the accreditation list read; whether an audit is under way is not stated.
+
+## Positions
+
+Each position is a fact with its source. Numbers inside a position are the value read on 2026-11-09; the skill reads the current one.
+
+**P-metrics-and-attribution-1.** Sponsored ads report an attributed conversion on the date of the shopper's ad interaction, which may differ from the conversion date; attribution metrics for a report date stay incomplete until the lookback window has ended. (Ad campaign attribution; Overview of the Reporting v1 API; Understand unified reporting) [official]
+
+**P-metrics-and-attribution-2.** Amazon attributes each conversion to no more than one ad interaction across all campaigns and all advertisers, ranked promoted-product clicks, brand-halo clicks, promoted-product views, brand-halo views, and then by last touch with the last click before the last view. (Ad campaign attribution) [official]
+
+**P-metrics-and-attribution-3.** Amazon's lookback and eligible products differ by ad product and advertiser: Sponsored Products for sellers is 7 days, clicks only, crediting the advertised products and any other product the Seller Central merchant sells; Sponsored Products for vendors and authors is 14 days, clicks only, crediting the advertised products and same-brand products, only those sold by Amazon, with no brand halo for books; Sponsored Brands, Sponsored Display, Streaming TV and DSP are 14 days and credit the brand halo, including sales by third parties and used items. (Ad campaign attribution) [official]
+
+**P-metrics-and-attribution-4.** Sponsored Products and CPC Sponsored Brands attribute to clicks only and report zero conversions from views; vCPM Sponsored Brands and Sponsored Display, Streaming TV and DSP also attribute to views that meet the MRC standard of 50% of pixels for 1 second (display) or 2 seconds (video). (Ad campaign attribution; Overview of the Reporting v1 API; Performance metrics) [official]
+
+**P-metrics-and-attribution-5.** From 2026-01-01 Amazon credits views for vCPM Sponsored Brands and Sponsored Display (including branded keyword campaigns) and DSP Store ads with a shopping-signal enhanced last-touch model with a shorter window, and keeps the original 14-day all-views model as separate "all views" metrics for comparison. (View attribution updates for Amazon Store ads; Ad campaign attribution; Purchases (all views)) [official]
+
+**P-metrics-and-attribution-6.** Amazon's impressions and clicks are net of invalid traffic (impressions = gross − invalid; clicks = gross − invalid), with gross and invalid counts reported as separate metrics. (Traffic metrics; Impressions (Delivery); Clicks (Delivery)) [official]
+
+**P-metrics-and-attribution-7.** Amazon defines CTR as clicks / impressions, CPC as cost / clicks, CPM as cost / impressions × 1,000, vCPM as cost / viewable impressions × 1,000 and ROAS as sales / total cost; its ACOS guide defines ACOS as ad spend / ad revenue × 100, the inverse of ROAS. (Clicks (Delivery); Impressions (Delivery); Viewability (Delivery); Purchases (v1); What is ACOS) [official]
+
+**P-metrics-and-attribution-8.** Unified reporting and the Reporting API v1 have no ACOS field; Reporting v3 has `acosClicks7d` and `acosClicks14d` on the Sponsored Products targeting, search-term and advertised-product report types and on the prompt and video ad-extension report types, not on `spCampaigns`. (Reporting columns (version 3); Purchases (v1); Conversion metrics) [official]
+
+**P-metrics-and-attribution-9.** In unified reporting and v1, "Purchase rate" is purchases / impressions and "Purchase rate over clicks" is purchases from clicks / clicks (formerly "14 day conversion rate"); Amazon's Sponsored Products search-term help page (modified 2026-05-18) still calls the per-click rate "Purchase rate". (Conversion metrics; Purchases (v1); Search term report for Sponsored Products) [official]
+
+**P-metrics-and-attribution-10.** Amazon's bulksheet search-term page defines CTR as "calculated as impressions divided by clicks" and conversion rate as "The total number of orders divided by sales within 7 days from the date that your ad was clicked", the inverse of its definitions everywhere else, and fixes the bulksheet search-term lookback at 7 days for sellers and vendors regardless of the date range. (Bulksheets search term report; Clicks (Delivery); Conversion metrics) [official]
+
+**P-metrics-and-attribution-11.** Unified reporting and v1 drop window suffixes: Sponsored Products base metrics are 7-day for sellers and 14-day for vendors, other campaign types 14-day except DSP's flexible windows, and base metrics include halo, with promoted and halo as breakouts that replace "advertised SKU", "other SKU" and "brand halo". (Understand unified reporting; Overview of the Reporting v1 API; Conversion metrics) [official]
+
+**P-metrics-and-attribution-12.** Reporting v3 carries Sponsored Products 1-, 7-, 14- and 30-day click-attributed purchases, sales and units, and same-SKU and other-SKU variants; v1 carries no day-count variant; v3 is to be shut off after 2027-06-30. (Reporting columns (version 3); Purchases (v1); Amazon Ads reporting overview) [official]
+
+**P-metrics-and-attribution-13.** A purchase is an order event that can include several units; units sold is a separate metric. (Ad campaign attribution; Purchases (v1)) [official]
+
+**P-metrics-and-attribution-14.** Attributed sales exclude tax (VAT subtracted) and cart discounts, can include items sold by Amazon or other sellers and in used condition, and have payment failures and cancellations within 72 hours of order creation deducted, which can take up to 7 days. (Ad campaign attribution) [official]
+
+**P-metrics-and-attribution-15.** Amazon does not report impressions, clicks or other traffic metrics at the converted-product level, because one ad interaction can lead to several product conversions. (Data availability in unified reporting) [official]
+
+**P-metrics-and-attribution-16.** Amazon counts a purchase as new-to-brand when the customer has not bought from the brand within a 12-month lookback, from purchase history based on all shopping activity within the brand on Amazon, not limited to attributed purchases; its NTB help page says NTB data is not available for Sponsored Products (P-metrics-and-attribution-35 sets this against v1's eligibility list). (Sponsored ads new-to-brand metrics; Performance metrics; Purchases (new to brand) (v1)) [official]
+
+**P-metrics-and-attribution-17.** Detail page views, add to cart, viewable impressions and vCPM are not offered for Sponsored Products in v1; detail page views are not unique per customer, and in unified reporting include halo. (Detail page views (v1); Add to cart (v1); Viewability (Delivery); Performance metrics; Understand unified reporting) [official]
+
+**P-metrics-and-attribution-18.** KENP read is attributed to reading up to 14 days after an ad click and reported on the day the pages were read; offline reads can take a week to be reported; estimated KENP royalties multiply KENP read by the KENP rate at the time of reading, using the latest published rate when that day's rate is not yet published; KENP metrics appear once at least one Sponsored Products campaign promotes a book with at least one attributed KENP read. (KENP metrics; Ad campaign attribution) [official]
+
+**P-metrics-and-attribution-19.** Amazon Attribution reports events on the date they occurred, not on the date of the attributed click, with a 14-day last-click model in which the last-clicked tag gets full credit; it restates at 1, 7 and 28 days taking returns and refunds into account, and masks rows with fewer than 10 lifetime clicks. (Amazon Attribution OpenAPI model; Amazon Attribution how-to) [official]
+
+**P-metrics-and-attribution-20.** v1 aggregates report days in each campaign's time zone and expresses money in the campaign budget currency, requiring `budgetCurrency.value` with money metrics; converted-currency fields require `convertedCurrency.value`; console downloads use the account's registered time zone and AMC defaults to UTC. (Dates and historical data retention (v1); Helpful concepts (v1); Downloadable reports in sponsored ads; AMC data sources) [official]
+
+**P-metrics-and-attribution-21.** Amazon's stated settling schedule: reporting data within 12 hours, clicks changing for up to 3 days, conversions restated at 1, 7 and 28 days (up to 42 days for a 14-day window) and changing for up to 60 days, invalid-traffic detection for up to 30 days, with Amazon advising to wait at least 5 days. (Reporting v3 FAQ; Amazon Ads reporting overview; Adjustments to invalid spend; Traffic validation) [official]
+
+**P-metrics-and-attribution-22.** In the console, metrics update hourly, sales take up to 48 hours to populate, and cost is an estimate that can change until 3 days after the click or viewable impression. (Performance metrics) [official]
+
+**P-metrics-and-attribution-23.** Amazon says an invalid-traffic rate (IVR) level on its own does not indicate undetected invalid traffic; for Sponsored Ads campaigns, a significant drop in IVR with an anomalous rise in CTR, impressions or clicks, together with a drop in conversion rate, is in Amazon's words "a higher confidence indicator for undetected IVT". (Traffic validation) [official]
+
+**P-metrics-and-attribution-24.** The MRC accreditation renewed on 2026-07-08 covers Amazon's Sponsored Products gross clicks and clicks net of SIVT on Amazon placements, in total reporting without environment segmentation; impressions and conversions are not accredited statistics. (MRC accreditation letter; MRC accredited data for Sponsored Products campaigns) [third party: MRC; official for the console filter]
+
+**P-metrics-and-attribution-25.** Amazon's multi-touch attribution is a US beta that splits credit across ad touches in proportion to their value using experiments, machine learning and shopping signals, reported as separate multi-touch purchases, sales and ROAS beside last-touch metrics. (Uncover ad value … multi-touch attribution (beta); Purchases (v1); Lewis, Zettelmeyer, Gordon et al. 2025) [official; research]
+
+**P-metrics-and-attribution-26.** The console's legacy sponsored ads and Display, Video and Audio report centres shut on 2026-12-31, with saved reports, schedules and historical data deleted; new legacy reports and edits stop on 2026-12-17; Amazon says these dates are subject to change. (Migrate your scheduled reports to unified reporting; Understand unified reporting) [official]
+
+**P-metrics-and-attribution-27.** Before unified reporting, Display, Video and Audio campaigns and legacy Sponsored Brands reported conversions on the conversion date; they now report on the traffic date, which moves conversions earlier in a flight without changing their total. (Understand unified reporting; Overview of the Reporting v1 API) [official]
+
+**P-metrics-and-attribution-28.** AMC cannot change a campaign's attribution window; its `conversions` table marks a conversion ad-attributed if the user had a traffic event in the prior 28 days, and it offers conversion-time and traffic-time tables whose traffic-time output can change for up to 30 days. (AMC conversions table; Amazon attributed events by conversion time) [official]
+
+**P-metrics-and-attribution-29.** Amazon's top-of-search impression share is the share of top-of-search impressions over eligible ones, covering the trailing 90 days and, for Sponsored Products, the first page only; unified reporting offers it at campaign level only. (Search (Impression share); Performance metrics; Migrate your scheduled reports to unified reporting) [official]
+
+**P-metrics-and-attribution-30.** Qin (2023, Amazon Ads, 15 US brands, 2021–2022) measured no Sponsored Products carry-over beyond 3 weeks, while about 40% of DSP display effect and about 30% of Sponsored Display effect came in week 5 and beyond. (Qin 2023, Applied Marketing Analytics) [research]
+
+**P-metrics-and-attribution-31.** Gordon, Moakler and Zettelmeyer's PIE study of 2,226 Meta experiments found 7-day last-click attribution predicted experimental incrementality with R² 0.19, against 0.88 for a model trained on experiments. (Gordon, Moakler and Zettelmeyer, arXiv 2304.06828) [research]
+
+**P-metrics-and-attribution-32.** Lily Tong, a director of measurement at Amazon Ads, reported by Digiday, says roughly 20% of a brand's sales value on Amazon arrives after the attribution window, from an Amazon analysis of more than 2 million June campaigns. (Relevant Audience citing Digiday) [opinion, Amazon spokesperson via press]
+
+**P-metrics-and-attribution-33.** Optmyzr (Lakshmi Padmanaban, 2026-04-08) reports that across 14,991 campaigns the top 5% gained at least 18.75% in attributed sales between day 1 and day 17, and treats the last 28 days as provisional. (Optmyzr blog) [opinion]
+
+**P-metrics-and-attribution-34.** Ad Badger (2026-08-13) describes Sponsored Products and Sponsored Brands as using 14-day click attribution and reports as built by order date; both statements contradict Amazon's attribution page. (Ad Badger, Attribution Window Blind Spots; Ad campaign attribution) [opinion; official]
+
+*Positions 35–54 were added in the gap fill of 2026-11-09.*
+
+**P-metrics-and-attribution-35.** Amazon's new-to-brand help page (modified 2026-08-19) states "New-to-brand metrics are available for Sponsored Brands and Display" and "New-to-brand data isn't available for Sponsored Products"; Reporting v3 offers NTB columns on no `sp*` report type; v1's NTB page names Sponsored Products in the eligibility of 90 of its 102 fields, including view-attributed fields such as `metric.newToBrandPurchasesFromViews`, although Amazon reports 0 view conversions for Sponsored Products. (Sponsored ads new-to-brand metrics; Reporting columns (version 3); Purchases (new to brand) (v1); Overview of the Reporting v1 API) [official]
+
+**P-metrics-and-attribution-36.** Amazon's multi-touch launch page says multi-touch metrics "can also be accessed in most report types for version 3 of the Amazon Ads API", but the v3 column list contains no multi-touch column, and all 56 v1 `metric.multiTouch*` fields give "N/A" as their v3 names and program eligibility; each multi-touch field requires its last-touch counterpart (for example `metric.multiTouchPurchases` requires `metric.purchases`). (Uncover ad value across the funnel with multi-touch attribution (beta); Reporting columns (version 3); Purchases (v1)) [official]
+
+**P-metrics-and-attribution-37.** Since 2026-09-25 Amazon's long-term sales estimates in the US "use machine learning models in place of the previous heuristic-based calculation", with 30-, 60-, 90- and 180-day horizons beside the 365-day `metric.longTermSales`, cohort estimates and `metric.cohortAccumulatedSales30d…365d` by the `longTermSalesCohort.value` month, which require `campaign.id` and "aren't summable across cohorts"; Amazon presents the accumulated-sales fields as a way "to validate estimate accuracy". (Amazon Ads advanced tools release notes, September 25, 2026; Long-term sales (v1)) [official]
+
+**P-metrics-and-attribution-38.** Amazon defines `metric.longTermSales` as "Attributed sales plus the estimated 365-day value of new-to-brand shopper engagements attributed to the campaign, based on the brand's historical shopping patterns" and calls LTS "ad-attributed measures that estimate the incremental sales value"; its help page credits the step in return on engagement to "the responsible campaign that drove the new customer engagement", and no Amazon page read describes a holdout or unexposed comparison in the LTS calculation. (Long-term sales (v1); Amazon Ads advanced tools release notes, June 5, 2025; Long-term sales help page) [official; the absence is as searched]
+
+**P-metrics-and-attribution-39.** Athey, Chetty, Imbens and Kang show that the treatment effect on a surrogate index of short-term outcomes equals the effect on the long-term outcome "Under the Prentice surrogacy assumption, which requires that the primary outcome is independent of the treatment conditional on the surrogates", in a setting where an experimental sample links treatment to surrogates, and they characterise the bias when the assumptions fail. (Athey, Chetty, Imbens & Kang, NBER w26463, 2019, rev. 2024) [research]
+
+**P-metrics-and-attribution-40.** The IAB/MRC Retail Media Measurement Guidelines (January 2024) state that attribution windows "must be empirically supported", "must be disclosed before campaign execution and measurement", should have "a reasonable maximum" applied consistently, and that day-level data on the lag from ad view or click to conversion "is encouraged to allow reconciliation across differing windows". (IAB/MRC Retail Media Measurement Guidelines, p. 27) [third party: IAB/MRC]
+
+**P-metrics-and-attribution-41.** The IAB/MRC guidelines state that "The MRC requires viewable impressions for attribution of outcomes to ad exposures", that attribution on non-viewable or unmeasurable impressions "must be properly labeled with a clear disclaimer", and that a click can stand in for viewability; Amazon's attribution page says that where direct viewability measurement is not available, "viewability is inferred based on historical data and placement characteristics". (IAB/MRC Retail Media Measurement Guidelines, p. 25 [third party: IAB/MRC]; Ad campaign attribution [official])
+
+**P-metrics-and-attribution-42.** The IAB/MRC guidelines state that when a measurer forecasts a campaign's full 12-month effect, "The basis for such forecasts must be disclosed and the methodology for extending shorter period results to 12 months or longer periods should be re-evaluated periodically based on empirical evidence". (IAB/MRC Retail Media Measurement Guidelines, p. 27) [third party: IAB/MRC]
+
+**P-metrics-and-attribution-43.** The IAB/MRC guidelines state that "MRC Standards highly recommends using RCTs" for incrementality; the IAB and IAB Europe guidelines of November 2025 state that attribution and ROAS "show what happened, not whether marketing caused the result" and grade "New-to-brand percentage, baseline vs. exposed analysis, platform-reported incrementality, Simple Multi-Touch Attribution" as "Weak" in causal strength, experiments (RCTs, holdouts or ghost ads, matched markets) as "Strong". (IAB/MRC Retail Media Measurement Guidelines, p. 32; Guidelines for Incremental Measurement in Commerce Media, pp. 2–3) [third party: IAB/MRC; IAB/IAB Europe]
+
+**P-metrics-and-attribution-44.** The MRC's digital accreditation list names, of Amazon's own measurement, Sponsored Products clicks (display and search, total only) with SIVT filtration, and Amazon DSP impressions, clicks and viewable impressions by supply source and environment on a GIVT basis; Sponsored Brands and Sponsored Display do not appear on the list. (MRC Accredited Services — Digital; MRC letters to Amazon Ads of May 29 and July 8, 2026) [third party: MRC]
+
+**P-metrics-and-attribution-45.** Danaher and van Heerde (JMR 2018) show that attribution weights are proportional to a medium's marginal effectiveness times its number of exposures, while the profit-maximising allocation of a fixed budget depends on effectiveness and not on past exposure levels, so attribution-based allocation can be misleading. (Danaher & van Heerde 2018) [research]
+
+**P-metrics-and-attribution-46.** Li and Kannan (JMR 2014) and Anderl et al. (IJRM 2016) found channel credit from carryover-and-spillover and Markov-path models to differ substantially from last-click credit, and Li and Kannan's field study found cases in which retargeting can lower conversion probability. (Li & Kannan 2014; Anderl, Becker, von Wangenheim & Schumann 2016) [research]
+
+**P-metrics-and-attribution-47.** Dalessandro et al. (ADKDD 2012) frame attribution as causal estimation and state that game-theoretic approximations of it, when causal assumptions are violated, "can be interpreted as variable importance measures". (Dalessandro, Perlich, Stitelman & Provost 2012) [research]
+
+**P-metrics-and-attribution-48.** Zantedeschi, Feit and Bradlow (Management Science 2017), using randomised campaign holdouts, found single-campaign effects imprecise, estimated separate ad-stock decay rates by medium, and found catalogs' effects last substantially longer than emails'. (Zantedeschi, Feit & Bradlow 2017) [research]
+
+**P-metrics-and-attribution-49.** In a Yahoo! display experiment with 1.6 million customers, Lewis and Reiley found that 78% of the purchase increase came from consumers who never clicked the ads, and that an observational cross-sectional estimator was biased negative by three times the experimental effect. (Lewis & Reiley 2014) [research]
+
+**P-metrics-and-attribution-50.** Johnson, Lewis and Nubbemeyer's ghost-ads design compares exposed users with their randomised would-have-been-exposed counterparts; in a retailer's display retargeting campaign it measured lifts of 17.2% in website visits and 10.5% in purchases. (Johnson, Lewis & Nubbemeyer 2017) [research]
+
+**P-metrics-and-attribution-51.** In Qin's Amazon study the two brands with a Sponsored Products row show no SP effect after week 3, with 37% and 69% of the DPV effect (48% and 62% of revenue) in week 1 and the rest in weeks 2–3. (Qin 2023, Table 1) [research, Amazon-authored]
+
+**P-metrics-and-attribution-52.** Lily Tong of Amazon Ads, quoted by Digiday on 2026-10-01, says roughly 20% of a brand's sales value shows up after the attribution window, a figure Amazon drew from more than 2 million campaigns across its ad products in June; Digiday reports agencies split on long-term sales, some treating it as "a forecast that shifts whenever the campaign mix changes". (Digiday, Seb Joseph, 2026-10-01) [opinion, Amazon spokesperson and agencies via press]
+
+**P-metrics-and-attribution-53.** Dekimpe and Hanssens's persistence modelling tests whether sales are stationary or evolving to tell temporary from permanent marketing effects; Qin, applying it to Amazon brands, found every KPI stationary, which in her words "implies that no marketing effects last forever". (Dekimpe & Hanssens 1995; Qin 2023) [research]
+
+**P-metrics-and-attribution-54.** Amazon says its 2026 shopping-signal view model is "based on our continuous research on shopping journeys in the Amazon store"; no Amazon page read publishes an incrementality figure for view-attributed purchases under the new model or the "all views" model. (View Attribution Updates for Amazon Store ads [official]; the absence is as searched)
+
+### Seat's positions (enter no skill)
+
+- S1. [seat] A seller's ACOS read from v1 is `metric.totalCost / metric.sales` on the same rows, and equals 1 / `metric.roas`.
+- S2. [seat] The most recent lookback-length of days in any traffic-dated report reads worse on ROAS, ACOS and conversion rate than it will once settled.
+- S3. [seat] A comparison across 2026-01-01 for vCPM Sponsored Brands or Sponsored Display, or across 2025-10-01 for branded searches, compares two definitions; the all-views family is Amazon's bridge for the first.
+- S4. [seat] "Halo" for a seller's Sponsored Products means other products of the same merchant of any brand; for everyone else it means same-brand products.
+- S5. [seat] Reported cost and the invoiced ad charge in SP-API Finances differ in dating, tax and adjustments.
+- S6. [seat] Amazon Attribution and sponsored-ads figures with the same 14-day name are not additive or comparable without noting event-date against click-date.
+- S7. [seat] Long-term sales reads as a campaign's causal 12-month value only if the attributed engagements were caused by the campaign, the surrogacy condition holds, and the historical engagement-to-sales relation transfers; the accumulated-sales fields test only the last of these (F11).
+- S8. [seat] Moving a Sponsored Brands or Sponsored Display campaign from CPC to vCPM adds view-attributed purchases to its reported ROAS by definition; whether those purchases are incremental is not shown by any metric the seller reads (C9).
+- S9. [seat] On Amazon data, Sponsored Products effects have been measured to fade within about three weeks, with a material share in weeks 2–3, which is after a seller's 7-day window; the non-Amazon evidence of effects building over months is not Sponsored Products evidence (Published research, carry-over paragraph).
+
+## Verification
+
+Verified 2026-11-09 by an adversarial verifier seat.
+
+**Method.** Every cited page was opened again by plain GET as a browser would: Amazon Ads help pages as HTML with the embedded content JSON decoded; Ads API guides as markdown on `https://d3a0d0y2hgofx6.cloudfront.net/en-us/<path>.md`; the OpenAPI models (OfflineReport, merged, merged ALL, Amazon Attribution) as JSON on Amazon's CloudFront hosts; the SP-API Finances and Data Kiosk models on GitHub raw; the MRC letter and Qin 2023 as PDFs; practitioner and press pages as HTML; research through arXiv abstracts, with journal metadata through search. Each quoted string was checked by script against the decoded text (whitespace and curly-quote normalised); v1 field pages were parsed into field / eligibility / description / v3-name tables and checked against the v3 column list per report type. Searches for 2025–2026 announcements covered view attribution, multi-touch, long-term sales, impression counting and AMC attribution. No position was phrased as behaviour; none needed moving to the seat's list.
+
+**Corrections (was → now).**
+1. Tags line, D7, Stable 14, P-24, Pages read: MRC letter "[official, third party]" → "[third party: MRC]" (the Amazon MRC-filter help page stays [official]).
+2. H1 author quotes: "Attribution reflects orders when created" / "Paperback sales are reported after … printed and shipped" / "credits, borrows, and cash" → Amazon's verbatim "when they're created", "KDP sales reports for paperback and hardcover books reflect sales after printing and shipping", "credit redemption, borrows, and cash purchases".
+3. H1 KENP visibility and royalty quotes (not verbatim) → the page's own sentences ("…if, as a minimum, 1 of your Sponsored Products campaigns promotes a book that has at least 1 KENP Read attributed to it"; royalties "calculated by multiplying KENP read by the KENP rate at the time the pages were read…"); P-18 reworded to match.
+4. E1 console quote → the page's verbatim text, adding the other-campaign-types sentence.
+5. E1 API quote "Interval suffixes removed: … standardizes on a single attribution window per ad product … DSP: Flexible" (not on the page) → the v1 overview's actual "Removal of interval suffixes (1d, 7d, 14d, 30d)" text; release-note sentence replaced by its actual wording.
+6. "Amazon Ads API release notes" → page title "Amazon Ads advanced tools release notes", throughout.
+7. E3 v1 `metric.purchases` long definition (no longer on the page) → current "The number of purchases attributed to an ad interaction."; v3 names line with its "[SP: 7d sellers / 14d vendors]" note; "Purchases (formerly Orders, …)" construction → the console table's "former name" cells; `metric.unitsSold` description updated.
+8. E4 `metric.sales` description and v3 names updated; "Sales (formerly …)" construction replaced the same way.
+9. E5 `metric.purchasesPromoted` v3 names: removed `attributedConversions…SameSKU` (on neither page) → "purchasesPromoted, purchasesSameSku7d, purchasesSameSku14d"; "Purchases (halo) (formerly …)" construction → the table's definition and former names; halo rename quote made verbatim.
+10. E7 "Purchase rate … The percentage of purchases per impression" → "The rate of purchases relative to ad impressions. Purchase rate = purchases / impressions"; per-click definition made verbatim.
+11. E8 truncated quote → full "…occurring within 7 days of your ad being clicked"; added the page's second label error (vendor "Purchases" column defined as a rate).
+12. E9/I2/E11 v3 `roas` "(SB/SD)" → on Sponsored TV and DSP report types only (`stCampaigns`, `stTargeting`, `dspCampaign`, `dspInventory`); ROAS "ratio" → "an index (multiplier) rather than a percent (%)"; returnPerThousandImpressions formula "sales / impressions * 1,000" → "sales / (impressions / 1,000)"; costPerPurchase "formerly eCPP / cost per purchase" → former DVA name "Total effective cost per purchase (total eCPP)".
+13. E10, Stable 8, P-8: `acosClicks7d/14d` "only on spTargeting, spSearchTerm, spAdvertisedProduct" → also on `spPromptAdExtension`, `sbPromptAdExtension`, `spVideoAdExtension`; not `spCampaigns`.
+14. E10 bulksheet ACOS quote completed; E10 list of v1 pages read corrected (no separate Multi-touch or New-to-brand page).
+15. E12 `purchasesPromoted`/`salesPromoted` "SB/SD" → Sponsored Brands report types only (salesPromoted also `dspProduct`), with their actual descriptions.
+16. F1 NTB quotes (not verbatim; "last 365 days") → the NTB page's 12-month look-back sentences, the attribution page's "previous 365 days", Performance metrics' purchase-history sentence, the start-date sentences and the legacy-SB "0"/"XX" sentence as written; P-16 reworded.
+17. F2 v3 `newToBrandPurchases` "SB, SD and ST only" → also DSP report types, `conversionPath`, `sbPromptAdExtension`; no `sp*` type.
+18. F2, P-16, Pages read: "New to brand (v1)" at …/new-to-brand.md (404) → "Purchases (new to brand) (Amazon retail conversions)" at …/purchases-new-to-brand.md.
+19. F3 "Detail page views now include halo" → "Detail page views now includes promoted and halo conversions".
+20. F5 "trends may shift" → "Advertisers may notice new trends in their branded search metrics after this update."; 2026-08-26 retention sentence made verbatim ("Maximum report date ranges are unchanged").
+21. F6 LTS description → the help page's definition and formula; "in closed beta since October 2024" → "unveiled in closed beta in October 2024", with the 2025-06-05 release note putting LTS in campaign reporting.
+22. F6, practitioner 10, P-32: "Amazon's director of measurement" → "a director of measurement at Amazon Ads".
+23. G1 rewind quote → "…we use the total percentage of unduplicated video that was viewed to calculate the completion rate."; "5 second views … available for Sponsored Brands only" → "This is only available for Sponsored Brands video ads.", with v3's "Sponsored Brands and Sponsored Products video-only metric".
+24. G3 pixel-metrics quote attributed to Migrate your scheduled reports, not Engagement metrics.
+25. H3 mapping error stated in full: the FromClicks and FromViews v3 names are swapped both ways.
+26. I2 promoted row: noted v1's `salesSameSku7d/14d` is absent from v3; halo row: `salesBrandHalo` is Sponsored Display vCPM and `dspProduct` only; KENP row adds unsuffixed `kindleEditionNormalizedPagesRead`.
+27. I3 "an attribute requires its primary key" and "CSV empty string, JSON null" (paraphrases) → verbatim; error codes quoted from Helpful concepts (the OpenAPI model does not carry them).
+28. J2 "v3 reports expose `advertiserTimezone`" → DSP report types only (`dspCampaign`, `dspInventory`, `dspAudience`); "Profiles guide" → "Profiles".
+29. J3: added that Helpful concepts names `metric.convertedSales` while the field page carries `metric.salesConverted`.
+30. K1 bulksheet quotes → the page's column R and W text; P-10 reworded.
+31. K1/E10/Pages read bulksheet URL …/guides/no-code-tools/… (404) → …/en-us/no-code-tools/bulksheets/bulksheets-search-term-report.md.
+32. K2 "does not affect" / "with at least 1 click" → "does not affect the search term report metrics" / "includes only the terms that resulted in at least one ad click".
+33. K3 v3 FAQ "had impressions … only search terms with clicks" → verbatim; SP help page quote made exact.
+34. K7 v1 retention "120 days for some" etc. (the second figure is the maximum report date range, not a subset of metrics) → the table as written; page title "Dates (v1)" → "Dates and historical data retention", throughout.
+35. L2/P-23 "higher-confidence signal" and "not an indicator" → "IVR level on its own doesn't indicate undetected IVT" and "a higher confidence indicator for undetected IVT", scoped "For Sponsored Ads campaigns"; the 5% discrepancy scoped to Display, Video, and Audio.
+36. L3, N1, N2, Stable 12, P-12, P-21: "Reporting overview" → "Amazon Ads reporting overview".
+37. M1 page title → "… multi-touch attribution (beta)"; added "Announced at unBoxed 2024", multi-touch ACoS, breakouts, SP via "most report types" of the v3 API and Stream, console downloads for DSP/SD/SB/ST.
+38. M2, Stable 15, P-25, Pages read: "Multi-touch (v1)" at …/multi-touch.md (404) → the 56 `metric.multiTouch*` fields on the Purchases page.
+39. N2: added the 2026-09-30 release note's per-advertiser statement (three Amazon statements, two readings).
+40. N3 "up to 5 accounts" → Helpful concepts says 5 while the OpenAPI `accessRequestedAccounts` `maxItems` is 2000; v1 warnings dated to the 2026-10-05 release note; Open question 7 extended.
+41. O1 page names → "How does Amazon Marketing Cloud work", "Overview - Amazon Marketing Cloud", "Get access to AMC using self-service APIs"; terms "must be accepted once in each region (NA, EU, and FE)".
+42. O2 "retention 13 months to 5 years by instance" → "by data view", "most tables" backfilled to June 2023.
+43. O3: added Amazon's 2022-03-25 announcement "Amazon Marketing Cloud now supports custom attribution analysis" (query lookback up to 28 days; first-touch, last-touch, equal-weight, position-based) [official].
+44. P1 and the run-time list: `subType` `AMAZON_ATTRIBUTION` as the way the profile appears → a legacy profile that "New integrators should filter out", Amazon Attribution reached through the ad console account; eligibility split between the API overview (brand-registered professional sellers and vendors) and the product page (adds KDP authors, agencies, free).
+45. C3 quotes made verbatim ("giving credit to ad views…", "Conversion reporting for all other campaign types and inventory types remains unchanged. Click-based attribution remains unchanged.", the all-views sentence); page title "View Attribution Updates for Amazon Store ads", dated January 01, 2026.
+46. Practitioner 1: spend up to 14 days after pause qualified to "Ad types, like lock screen Ads"; exact URL added.
+47. Practitioner 3 and P-33: "grow 18.75%" / "36.67%" → "by at least"; "Method and data selection are not published" → published in Frederick Vallaeys's companion post (302 accounts, 14 marketplaces, 14-day window, day 17 final); Optmyzr's returns claim noted against A7; URLs added.
+48. Practitioner 5: removed TACoS (the podcast page does not mention it) → its own words "look way worse" and "bad optimization decisions".
+49. Practitioners 6, 7, 8, 9: exact URLs; Helium 10 dated 2022-03-24, updated 2022-08-26.
+50. Practitioner 9: PPC Land's scope (SB, SD and DSP without vCPM) noted as broader than Amazon's page.
+51. Research 3 "median absolute errors of 83% and 173% …" → median estimated lifts by DML (SPSM) against median RCT lifts, from the arXiv v2 abstract (2022-10-04); journal figures not read.
+52. Research 4 title "Predictive Incrementality…" → "Predicted Incrementality by Experimentation (PIE) for Ad Measurement".
+53. Research 7 "confidence intervals … often wider than 100 percentage points" → "the median confidence interval on ROI is over 100 percentage points wide".
+54. P-3 vendor row: "same-brand products sold by Amazon" → "same-brand products, only those sold by Amazon" (the cell's own condition).
+55. P-11: "all other ad products 14-day" → "other campaign types 14-day except DSP's flexible windows".
+56. Pages read: exact paths filled in for every row given as a directory (KENP, KU, video, LTS, AMC tables, retention, thresholds, Openbridge, Optmyzr, Helium 10, Intentwise, Code3, PPC Land); rows added for Detail page views (new to brand), the AMC custom-attribution announcement, the Optmyzr method post, the PPC Land impression-count article, Lewis and Rao, and Chapelle.
+
+**Additions from the 2025–2026 search (none overturns a claim).** PPC Land (2026-11-03) reports an undated Amazon notice that from 2027-01-01 DVA display impressions on the Amazon Store count only once on screen, about 63% fewer; no Amazon page read states it and it names no sponsored-ads product (practitioner 9). The 2026-08-26 retention change and the 2026-10-05 v1 warnings were already within the file's dates.
+
+**Removals.**
+1. P4 "top 100 products" for the Amazon Attribution PRODUCTS report: on neither the how-to nor the OpenAPI model; replaced by the page's "No click-throughs".
+2. Practitioner 9 "counts about 54 all-views fields": not in the PPC Land article (the 54 count stands from Amazon's own page, C4).
+3. E5 `attributedConversions…SameSKU`: on neither the v1 page nor the v3 list.
+4. Practitioner 3 "Method and data selection are not published": the method is published.
+5. Practitioner 5 "TACoS": not on the page.
+
+**Stayed unreachable.** Seller Central help (JavaScript shell; content call forbidden); Amazon's conversion attribution GJKPFEUSF9236MMW (empty body, confirmed again); Search Engine Land and MarketingCharts (403); INFORMS PubsOnline full texts (403), so Research 3, 5 and 6 rest on arXiv abstracts or abstract metadata, and Research 7 and 8 on search-surfaced abstracts and metadata; IDEAS/RePEc for Lewis and Rao (404). Not re-opened in this pass: the cx-internal local files (door.md, listing-playbook research), which are cx's own records and were not changed.
+
+**Hosts.** No Ads API host (advertising-api*.amazon.com), Login with Amazon (api.amazon.com/auth), Amazon Ads MCP server endpoint, SP-API host, or the Seller Central `mons-api/GetHelpTopic` call was contacted; no page needing sign-in was opened; no seller data was read.
+
+### Gap fill 2026-11-09
+
+Done by the research seat that owns this file, on three completeness critics' lists; same sourcing rules, every page opened by plain GET on 2026-11-09, PDFs read from the network stream with PyMuPDF and nothing saved. No Ads API host, Login with Amazon, Amazon Ads MCP endpoint, SP-API host or Seller Central `mons-api` call was contacted; no page needing sign-in; no seller data.
+
+**Gaps added.**
+1. *Industry measurement standard (gap 1).* Added §R (R1–R8) from the IAB/MRC Retail Media Measurement Guidelines (January 2024): attribution windows, weights and halo disclosure, viewable impressions for outcome attribution, 12-month forecasts, extrapolated conversions, IVT, RCTs and ITT, NTB and disclosure; and the IAB/IAB Europe incrementality guidelines (November 2025) with their four graded method families. Accreditation: added D7a from the MRC digital accreditation list and the Amazon DSP letter of 2026-05-29; re-opened the Sponsored Products letter of 2026-07-08 (scope unchanged; TAG addendum noted). Sponsored Brands and Sponsored Display are not on the list. Stable items 18–19, positions 40–44, run-time item on accreditation, Open questions 12 and 14.
+2. *Attribution-method literature (gap 2).* Added Published research 9–13 (Danaher & van Heerde 2018; Li & Kannan 2014; Anderl et al. 2016; Dalessandro et al. 2012; Zantedeschi, Feit & Bradlow 2017), from abstracts; stable item 20; positions 45–48.
+3. *Surrogates and long-term value (gap 3).* Added F8–F11 (the 2026-09-25 machine-learning switch, the 24 v1 LTS fields with the cohort and accumulated-sales fields, the dated help page's heuristic and its use of "incremental", and the surrogacy conditions LTS needs), Published research 14 (Athey, Chetty, Imbens & Kang) and 15 (Dekimpe & Hanssens, with Qin's use of persistence modelling); stable item 17; positions 37–39, 42, 53; seat S7; run-time item on LTS fields; Open question 13. The gap's wording "LTS is ad-attributed and not incremental" was kept as a reading, not a quotation: Amazon itself calls LTS "ad-attributed measures that estimate the incremental sales value", and the file now quotes both words and states what "incremental" means in Amazon's own worked example (F10).
+4. *Views (gap 4).* Added C8–C9, Published research 16 (Lewis & Reiley 2014) and 17 (Johnson, Lewis & Nubbemeyer 2017) beside the existing Gordon et al. items; positions 49–50, 54; seat S8. Lewis & Reiley's abstract also shows an observational estimator biased *negative*, so the file states that observational bias can run either way, not that it always overstates.
+
+**Contradictions settled.**
+1. *Multi-touch column names (owner: this file).* Re-opened the launch page, the v3 column list, v1's Purchases page, the Stream data guide and v2 overview, and the release notes. Facts: the v3 column list has no multi-touch column; v1 has 56 `metric.multiTouch*` fields, every one with v3 names and eligibility "N/A"; Stream pages read show none; the release notes have no multi-touch entry; the launch page claims "most report types for version 3". M1 was corrected (it had read the launch page as saying Sponsored Products reaches multi-touch through v3; the sentence's ad-product list may govern Stream only) and M1a added; M2, stable 15 and P-36 updated. For `method/change-measurement.md`: its "No multi-touch column was found on the Reporting columns (version 3) page" stands; its "no API column name was found on the pages read" does not, since v1 names 56 fields.
+2. *Sponsored Products new-to-brand (owner: this file).* Re-opened the NTB help page (G5EWB37XK4RS4J4T, lastModified 2026-08-19): "New-to-brand metrics are available for Sponsored Brands and Display" and "New-to-brand data isn't available for Sponsored Products." stand verbatim. The v1 eligibility list names SP on 90 of 102 NTB fields, including view-attributed ones SP cannot populate, so it is no evidence against the help page; the attribution page's NTB definition names no ad product. F2's "not settled" sentence was replaced by F2a; P-16 reworded; P-35 added; Open question 2 marked settled. For `areas/ads-organic-and-total.md`: its Open question 1 and P-9 are answered by the help page it does not cite; its "defines NTB without excluding SP" is accurate about the attribution page, which simply does not address ad products.
+3. *How long a Sponsored Products effect lasts (owner: `method/change-measurement.md`).* Re-opened Qin 2023 (PDF), Digiday (now read directly), and Hershaff. Facts: Qin's appendix has SP rows for only two brands, with 37% and 69% of the DPV effect in week 1, the remainder in weeks 2–3, and none after; Tong's ~20% after the window pools all ad products and windows of "seven to 30 days", method unpublished; Hershaff's MMM decay of ~0.35 is per day for one small seller. This file now weighs them (carry-over paragraph after Published research 17; P-51, P-52; S9): on Amazon data SP effects fade within about three weeks, with a material share after a seller's 7-day window; the months-long builds come from non-Amazon settings. This file's P-30 stands as written. The owner file's "a short pause understates the loss" is consistent with Qin for pauses shorter than about three weeks; the owner should weigh the three sources itself.
+
+**Declined.** None. Every gap was right on the facts as far as the pages read show; the two wording refinements are in gaps 3 and 4 above.
+
+**Corrections made in passing.** M1's reading of the launch page (above); F2's open sentence (above); P-16's second clause; the Lily Tong item is now sourced to Digiday directly (practitioner 10a) as well as to Relevant Audience.
+
+**Still unreachable.** Dstillery's copy of Dalessandro et al. (404), Semantic Scholar's paper page (403); journal full texts for Published research 9–17 (abstracts only).
+