PS2SP
A static, Git-backed catalog of PlayStation 2 homebrew, kept current by scheduled GitHub Actions that sync known projects and discover new ones inside a fixed API budget.
I wrote the first version (15 commits on 28–29 Sept 2026: the Astro site, content schema, JSON API, sync and discovery scripts, the maturity gate and the promotion step). Another maintainer (NathanNeurotic) has contributed most of the commits since, and a bot account commits the automated catalog updates. The record count above is the combined result, not something I hand-wrote.
Sources
config/- Trusted PS2 organisations and developers
- Incremental search with a stored watermark
- Resumable historical bootstrap by month
Discovery
every 2 hours, hard request budget- Maturity gate on candidates
- Fork networks need PS2-specific identity
- Queued to discovery/pending/
Promotion
scripts/promote-pending.mjs- Eligible candidates become Markdown
- Human-written fields kept separate from synced fields
Sync + publish
24 shards, ETag caching- Each project checked about daily
- 304 responses cost no processing
- Astro builds static pages and a JSON API
Condensed from the repository README.
The problem
The PlayStation 2 homebrew scene is spread across thousands of repositories, forks and abandoned experiments, and nothing keeps a list of it that is both browsable and current. A hand-edited list goes stale; a crawler that brute-forces GitHub burns through rate limits.
Design
The rules I set at the start are in the README: records are plain, reviewable Markdown; Git is the durable source of truth; there is no application database and no always-on backend; the site deploys as static files. Human-written metadata stays separate from synchronised repository fields, so an automated job can refresh stars, releases and activity without overwriting a curated description.
The interesting engineering
- Rate-limit budget as a first-class input. Discovery runs with a maximum request count, a reserve held back for the core and search quotas, and a delay between search calls. When the budget is spent it stops cleanly instead of failing.
- Sharded sync with conditional requests. Known projects are split into 24 deterministic shards, so an hourly job touches each project about once a day. Stored ETags go out as
If-None-Match, so unchanged repositories return304and produce no commit. - Three discovery strategies combined. Trusted owners, recent incremental searches that remember where they stopped, and a historical scan that walks backward in bounded monthly windows across runs.
- A maturity gate. A repository that merely mentions the console is not published. Candidates are queued and promoted only when they pass checks, and forks are stricter still: copied upstream README text does not count as identity.
- Static output with a real API. Astro generates the pages plus
projects.json,categories.jsonandreleases.jsonendpoints from the same content collection.
Status
Live and being extended. Because most of the catalog is generated, the number to judge my work by is the architecture and the scripts above rather than the record count.