# How we check

Every fact has a source and a date. If we can't find one, we say unknown.

## From repository to listing
1. Discover: a hand-picked seed list and repositories people submit. Each night a scan builds a trending candidates list (GitHub trending daily and weekly, plus fast-growing new repositories described as alternatives or self-hosted, plus repositories linked from recent videos on YouTube channels that review open-source alternatives). The owner reviews candidates at /owner/candidates; nothing is listed automatically.
2. Fetch: GitHub metadata, licence files, model cards and official docs.
3. Extract: a person records each fact with a link (source_url), the date checked (checked_at) and a short quote. A fact with no source is shown as unknown.
4. Validate: a fact without a source cannot be marked yes; missing licence evidence blocks the open-source label; conflicts are flagged, never resolved in a project's favour.
5. Refresh: a nightly GitHub Actions run refreshes stars, forks, contributors, releases and archived status and writes a daily star snapshot; documentation facts are rechecked on a schedule and on licence or archive changes.

The nightly refresh warns the owner when a repository now resolves to a different owner or name (renamed or transferred) and when stars rise by more than 500 and more than 20% since the last snapshot (possible star burst). Warnings are logged for review; they do not change a listing by themselves.

## Labels
- Confirmed: a primary source states it directly.
- Partial: true with a condition (copyleft, revenue cap, partial coverage).
- Unknown: no source found. Never passes a strict filter, never shown as yes.
- No / conflict: the source says no, or sources disagree.

We review documentation; we have not installed or tested these projects.

## Ranking
- Best fit: job coverage, then evidence coverage, freshness and recent activity.
- Top ranked: stars 40%, recent activity 30%, evidence coverage 30%.
- Trending: new GitHub stars gained in 24 hours, 7 or 30 days, from daily snapshots.
Stars show popularity, not quality. Sponsors never change rankings, facts or agent results; paid placements are labelled Sponsored and never appear in search, the API or MCP.

## What we store
We keep facts, evidence (link, short quote, date), daily public GitHub numbers and a change history. We never keep code, model weights, full READMEs, contributor personal data, or anything from private repositories. Accounts, subscriptions, sponsor bookings and API usage are covered in the privacy policy (https://repoquarry.dev/privacy).

---
Source: RepoQuarry (https://repoquarry.dev/how-we-check). Data licence: CC BY 4.0. Quotes are from project documentation and belong to their authors.