METHODOLOGY
How to interpret Sightings data
Sightings is an independent observatory of app ↔ model relationships visible in OpenRouter-published rankings and model pages.
What Sightings measures
Sightings records model catalog data, closed-day token rankings, app ranking context, and app ↔ model relationships exposed on public OpenRouter pages. A recorded relationship is public evidence of an association observed by Sightings, not an audit of an app’s full or private infrastructure and not a claim of exclusive model usage.
Sources and collection
The primary sources are OpenRouter public rankings and OpenRouter model pages. The scheduled collector runs every six hours. Historical metrics use closed UTC days; pages display the latest applicable evidence date or an explicit timestamp-unavailable state.
First-observed dates
A first-observed date is when Sightings first recorded an entity or pairing in the available evidence. It is not the start of usage. Records reconstructed from the initial bootstrap or historical backfill are labeled Historical import; only post-bootstrap observations can be labeled new sightings.
Coverage limits
OpenRouter model pages expose at most five apps per model-page scrape. Daily public app rankings cover about the top 200 apps per day. An absent relationship therefore means “not observed in the covered evidence,” not “does not exist.”
Identity and enrichment
Canonical app identity prefers an OpenRouter slug, then a verified external origin, then a normalized name. Conflicting or unresolved identities remain separate. App descriptions and own-site links appear only when bounded public crawling supplies sufficient identity evidence; uncertain enrichment is suppressed rather than guessed.
How to interpret a sighting
“App A was first observed with Model B on 2026-08-28” means Sightings first recorded that pairing on that date from OpenRouter-published model-page evidence. When citing a relationship, retain “first observed by Sightings,” the displayed evidence date, and the relevant coverage limit; do not restate the date as the start of usage.