What we measure, and what we don't
A tool that tells you what your ad platforms are missing is worth exactly what its own method is worth. So here is ours: every number we publish, where it was read, on what date, and the list of the ones we do not have yet.
Every figure wesaw publishes is one of three things: measured, meaning it was read off a real screen on a stated date; estimated, in which case the method is shown next to it; or not measured, in which case it says so and gives the reason. There is no fourth category, and nothing on this site is rounded in our favour.
Read this first: one test store, three orders, no customers
wesaw has no merchant customers, so there is no fleet of shops behind these numbers. Every figure below was read on a single Shopify store we run ourselves. The quality score and the deduplication check rest on three real test orders; the matching figure rests on nine, of which seven came from a normal shop journey. Those are the sample sizes, printed rather than left to be assumed, and none of this is a claim about what your store will get. It is evidence that the pipeline does what we say it does, on the only store we are entitled to show.
What is measured
Each of these was read off a screen on the date given, and each says which screen. Where a figure replaced an earlier one, the earlier one is shown too, because a number that improved is more informative than a number that appeared.
9.3 / 10Meta's Event Match Quality on Purchase
21 August 2026
Read off Meta Events Manager on our own test store, with the test event code removed, once conversions were being reported from both the browser and our server. The same score was 8.0 / 10 on 18 August, when only the server half was running. Meta grades the data it receives; it is not a result we produced for a merchant.
4 → 14match parameters sent with a purchase
18 August 2026
Counted in the payload we send to the Conversions API, against the four a stock Shopify pixel sends. Every one of the fourteen is a value the store already holds; none is invented to inflate the count, which is why the number is fourteen and not twenty.
3 and not 6purchase events counted for three test orders
21 August 2026
Three test orders, browser half and server half both reporting, checked in Events Manager: three Purchase events, integration shown as Multiple, no unmerged-event diagnostic. That is deduplication working, verified on Meta's side rather than on ours.
7 of 7orders matched back to the visit that produced them
22 August 2026
Nine test orders, of which seven came from a normal shop journey and all seven were matched to their originating session. The two failures were carts opened by typing a URL by hand, which is not something a shopper does. Small numbers, stated as small numbers. Those rows were deleted on 25 August 2026 when the store registry moved to the database, so this reading is the only record of them left: it was true when taken and it can no longer be re-checked, which is said here rather than left to be discovered.
What is not measured, and why
This is the part of a method page that usually does not exist. Every item below is something we would like to publish a number for and cannot yet, so it says not measured, and says what would change that.
How many sales are reported by both halves and merged
Not measured, and not measurable as the product is built. The browser half reports straight to your own ad-platform pixel without passing through our servers, so nothing in our database witnesses it, and the merge itself is decided by the ad platform, which does not expose it through any API. We would rather say this than show a merge rate we inferred.
What share of events an ad blocker actually stops
Not measured. On 22 August 2026 we checked our own paths against the public filter lists, which tells you what those lists contain, not what gets through in a real browser. The measurement that answers the question, a real Chrome with uBlock Origin running a full purchase journey, is planned and not done. Until then we do not put a blocking percentage on this site.
How long a visitor stays recognisable in Safari
Not measured. Safari's tracking prevention caps how long an identifier survives, and the exact cap depends on how the collection subdomain resolves. We have a specific question to answer and a test to run, over eight days without interaction, and we have not run it yet. Nothing on this site claims a retention window in the meantime.
How much each individual match parameter is worth
Not measured, and probably not measurable by us alone. The fourteen parameters were added together and the score moved together; Meta does not break its grade down by parameter. Anyone claiming a per-parameter contribution is modelling it.
What happens on Buy it now and dynamic checkout buttons
Not measured. These buttons skip the cart, which is where part of our matching lives, and we know it is a hole rather than having sized it. It is on the list, it is named here rather than discovered by the first merchant who uses those buttons.
The three rules this page enforces everywhere else
Measured
Traceable to a reading someone can name: a screen, a date, a store. If we cannot say where a number came from, it does not get published.
Estimated
Allowed, with the method printed next to it. An estimate that hides its method is a guess wearing a number's clothes.
Not measured
A legitimate answer, and the one this page exists to protect. The product itself uses it: where the dashboard cannot witness something, it says so and says why rather than filling the space.
Questions about the method
Why publish the things you have not measured?
Because the alternative is that you find them yourself, later, after paying. A tool whose whole argument is that your ad reporting is not to be trusted cannot ask to be trusted on its own numbers.
Do these figures mean my store will get the same?
No, and we will not suggest it. They were taken on our own test store, on a small number of orders. They show the pipeline works end to end; what your store gets depends on your traffic, your checkout and your customers.
Will these numbers change?
Yes, and when they do the old one stays visible next to the new one with both dates. A figure that quietly improves overnight is indistinguishable from one that was edited.
The terms on this page
Last reviewed: