Find repos
How the Find repos trend feed collects public evidence, scores repositories from 0 to 100, and hands candidates to the normal launch review.
What Find repos is
/find-repos lists public GitHub repositories that are gaining attention, with the evidence behind each one. It helps discoverers find projects to launch. It is not a search engine for all of GitHub: it covers up to 48 fresh, verified candidates with a positive score.
Nothing launches automatically. Every launch still goes through the normal review and your own wallet signature.
Using the page
- Search matches names and descriptions of tracked candidates.
- Scope: All repos, No market yet, or Live markets.
- Activity: narrow to Gaining stars or Recent releases.
- Each card shows stars, forks, measured star changes, when it was last checked, market state, and an expandable breakdown of its score and source links.
Pasting owner/repo or a GitHub URL into the search box skips the feed and uses the normal resolver: it opens the existing market, or the launch form.
Cards link to the existing market when there is one. Candidates approved for launch show Review & launch. Other candidates link to GitHub, and you can still launch them by pasting their URL at /launch.
Sources
A background worker collects public evidence, without credentials, at most once every 30 minutes.
| Source | What it contributes |
|---|---|
| GitHub repository API | Numeric ID, current owner and name, stars, forks |
| GitHub latest release | Release publication time |
| GitHub default-branch commits | Commits and distinct authors in the last 24 hours versus the previous 24 hours |
| GitHub Trending | Presence on the daily trending page |
| GitHub Search | New seed candidates (created in the last 90 days, pushed in the last 7 days, at least 20 stars, not forks or archived); no points |
| Hacker News (Algolia) | Recent stories linking directly to the repository, within seven days |
| Curated links | Operator-supplied links with a note; labeled separately and worth no points |
Private, archived and disabled repositories are rejected. Observations expire after six hours, and stale candidates drop out of the public feed.
Score
The score runs from 0 to 100 and is the sum of these parts:
| Input | Points |
|---|---|
| Star velocity (per 24 hours) | min(25, floor(stars per day / 10)) |
| Star acceleration | 10 when the latest interval added at least 10 stars at twice or more the previous rate |
| Fork velocity (per 24 hours) | min(15, floor(forks per day)) |
| New contributors | 2 per additional distinct GitHub author versus the previous 24 hours, up to 10 |
| Commit acceleration | 5 when the last 24 hours had at least 5 commits and at least twice the previous 24 hours |
| Release in the last seven days | 10 |
| Hacker News stories in the last seven days | 5 per distinct story, up to 15 |
| On GitHub Trending now | 10 |
| Curated link | 0 |
Rules that keep the score honest:
- Velocity needs two real snapshots taken 1 to 48 hours apart; acceleration needs three. New candidates are shown as warming up until then.
- Unknown or incomplete inputs earn zero and show why. Total star count and search position earn nothing.
- Negative changes are shown and earn zero.
- Ties are sorted by numeric repository ID.
The score is a sorting aid. It is not a quality rating, a price forecast or a recommendation to trade.
Review states
Candidates move through these states:
detected -> reviewed -> approved -> launched -> active
Review and approval are done by gitpad operators and are audited. Approval checks the repository's identity, freshness, that no market exists yet, and the active launch configuration. Review & launch then opens the normal launch form, where any wallet can review, choose name, ticker and initial buy, and sign. That wallet becomes the discoverer.
A candidate becomes launched only after the launch is finalized and indexed, and active after its first real trade. A market launched through another path is shown as an existing market and is not credited to the feed.
Optional natural-language search
A deployment can enable natural-language matching through an external model provider. When enabled, only the search text and public repository names and descriptions are sent, and the page discloses it. The model can only filter and match supplied candidates; it cannot change scores, order or market state. When it is disabled or unavailable, keyword search and filters work as usual.
Found something wrong? Edit this page on GitHub.