Data and corrections: sources, refresh and errors

Liveness, link attributes and listings come from our own crawler reading directory pages with an identifying user agent. Indexation checks run through a search API, one sampled URL at a time. Domain rating would be the one third-party figure on the site, labeled as such; we hold no DR value for any directory today, so every DR field reads "Unknown". We keep the factual metadata of what we read, never copies of pages, and we buy no third-party backlink database. The per-number rules are on the methodology page.

The crawler and the indexation sampler run continuously in the background at polite rates, so rows refresh on a rolling schedule rather than all at once. A listing's indexation is rechecked at most once every 30 days, and a published rate is computed over the last 90 days of checks. Pages show when a value was last checked, and the changelog records what changed between checks. History is append-only: a value that changes closes its old record and opens a new one, so past states remain visible.

Directory operators: claim your listing with a meta tag or DNS record and the badge, then correct how your row is presented: its tagline, description, icon and submission URL. Product owners: claim your product and edit its name, tagline, icon, website link, screenshots and categories.

In both cases the correction is published beside what our crawler read, permanently labeled "operator reported", rather than replacing it. That label does not expire: an owner's description and a crawled one are different kinds of fact and stay distinguishable for good. No measured field is editable by anyone, so the way to correct a measurement is to tell us it is wrong and let us re-measure.

We fix the data, not the story. A confirmed error is corrected in the store, the correction flows to every page that reads it, and the change is visible in the record rather than papered over. What we do not do: retroactively rewrite history rows, adjust a number by hand outside the measurement path, or take payment to revisit a result. Community votes, comments, newsletter signups and paid submissions have no path into any published number, and a test in the suite fails the build if a scoring or sorting query so much as mentions those tables.