ActorStack.dev
ActorStack.dev

How every number on this site was produced

Coverage percentages, throughput figures and accuracy claims all come from runs of the Actor being described. This page states how they were measured, what a source has to be before it can support a claim, and what gets published when a measurement turns out worse than the last one.

How the figures are measured

Every percentage on this site is a count, not an estimate. A fill rate is produced by running the Actor against a real sample, counting the rows in which a field has a non-null value, and dividing. The sample size, the number of cities or provinces it spans and the date are published next to the figure, because a coverage number without them cannot be checked or reproduced.

Two samples underlie most of what you will read here:

  • 741 distinct Nextdoor businesses across 5 cities and 12 categories, for the business and recommendation figures.
  • 550 coches.net listings across 35 makes and 38 provinces, dealer and private sellers, measured on 17 August 2026.

Throughput figures are end-to-end on the Apify platform — discovery included, not just the extraction stage — because a rate that excludes discovery is not the rate you will experience.

The correction that shaped this policy

The Nextdoor Actor's first fill-rate table was measured on 30 businesses. At 741, several fields came out lower, opening hours most sharply: 57.4% rather than the 77% the earlier sample suggested. The published table is the corrected one, and the earlier figures are not quietly gone — the article about it exists so the correction is part of the record.

The rule that came out of it: a number gets published with its sample size, and when the sample grows the number changes even if it changes for the worse. Anything not on a measured list is explicitly not promised, which is a stronger and more useful statement than leaving it out.

What counts as a source

Every article on this site cites its sources with a URL, a publisher and the date the URL was checked. Sources are ranked, and the rank governs what they are allowed to support:

LevelKindWhat it can support
1Law or regulatorThat something is required, forbidden, or officially defined.
2Platform documentationHow a platform behaves, what an API accepts, what a pricing model permits.
3The site's own declarationWhat a target site says about itself in machine-readable form: robots.txt, published schema.
4Operator claimOnly "X says Y". This includes my own Actors' READMEs and measurements.
5ContextBackground. Never the sole support for a factual claim.

My own Actors sit at level 4, deliberately. A README I wrote is evidence of what I claim, not proof that the claim is true — which is why the measurements are published with their method, so you can reproduce them rather than take my word.

URL verification

Every external URL cited on this site was requested and returned a page on the date shown next to it. That check is mechanical and it runs before a deploy, because a citation to a URL that does not resolve is worse than no citation: it looks like provenance and provides none.

Five things this site does not do

  1. Invent authorship. There is one author and he is a real person. No “editorial team”, no reviewer credited who did not review anything.
  2. Cite a source from memory. If a URL is on this site, it was opened.
  3. Round a measurement upward. 57.4% is not “around 60%”.
  4. Give legal advice. The compliance pages explain engineering decisions and cite the instruments they rest on.
  5. Emit structured data for content that is not on the page. An FAQ in JSON-LD whose answers are not visible is a misrepresentation, and the pre-deploy audit fails the build on it.

Corrections

If something here is wrong, tell me and it gets fixed with the change visible rather than silently overwritten: hello@actorstack.dev.

The full source catalogue

75 sources are referenced across the site. Each article lists the ones it relies on, with its level and check date.