Elin Chara Viewer

This is an application for viewing character data in the Elin.

Development

• npm run dev for development server
• npm run build for production build
• npm run check:archive -- <archive-dir> parses every archived version with the app's schemas (defaults to tmp/archive)
• npm run check:history holds the change history's PROVENANCE to the models
• npm run check:hidden verifies Chara#isHidden's id list against the game data; see docs/hidden-charas.md

Deployment

The site is a static export (output: 'export') served by Cloudflare Workers static assets, with no Worker script. Workers Builds builds and deploys it from this repository.

• Build command npm run build, deploy command npx wrangler deploy; non-production branches use the preview command npx wrangler preview, which creates or updates a Preview per branch. The Worker name must match name in wrangler.toml.
• npm run build runs script/prune-export.cjs afterwards. It deletes the __next.*.txt segment files, which are most of the export and only serve prefetching (NoPrefetchLink never prefetches), and fails the build when out/ still holds 20,000 files or more, the Workers Free limit per version.
• public/_headers sets the Cache-Control values and public/_redirects holds the redirects and the rewrites for past versions (below).
• npx wrangler dev serves out/ locally the way Cloudflare does, including _redirects and _headers. npm run dev rewrites past version URLs itself.
• Build variables:
  • NEXT_PUBLIC_ARCHIVE_BASE_URL: the archive host (see below). Required when building on Workers Builds (WORKERS_CI=1).
• Settings > Build > Build cache is enabled. Workers Builds then restores .next/cache between builds, which holds Turbopack's build cache (turbopackFileSystemCacheForBuild, about 320MB). Locally a build of a new commit on a warm cache compiles in about 1s instead of 30s. The cache is shared by every branch, including dependabot's, so a dependency's install script can write to the cache a production build reads. Clear the cache from the same page when the build output looks stale or the cache keeps growing.
• next build skips its type check on Workers Builds (WORKERS_CI=1). The Build step of the GitHub Actions CI type checks the same code, and master requires that check to pass. The ruleset does not require branches to be up to date, so two pull requests that pass separately can still deploy a master that fails the type check. A branch pushed without a pull request gets a Preview that nothing has type checked.
• Cloudflare Web Analytics uses the automatic setup for elin.pocke.me: Cloudflare injects the beacon into the HTML it serves on that hostname, so the site does not embed it.

Past versions

Only the two current versions live in db/ and are bundled into the build. Every past version lives in elin-chara-viewer-data and is fetched by the browser, so the build produces no pages for them.

• A past version's URL is /<lang>/<slug>/{charas,charas/<id>,feats,feats/<alias>,sources} with a slug such as EA-23.306. public/_redirects rewrites each of them (status 200) to a shell page under src/app/[lang]/archive/, which reads the slug and the id from the URL, looks the slug up in index.json and renders the page in the browser. Clicking a link between past version pages reloads the whole page, because the router's <url>.txt request is answered with the shell HTML. A slug or id that does not exist shows "not found" with status 200. Slugs always start with EA-, which keeps EA and Nightly out of the rewrites.

• /versions also reads index.json in the browser, so a release shows up without rebuilding the site.

• NEXT_PUBLIC_ARCHIVE_BASE_URL points at the archive. Production must set it: the default is the repository on raw.githubusercontent.com, which is meant for development and is not a CDN to serve readers from. The data repository syncs itself to Cloudflare R2 and that is what production points at; the bucket needs the CORS policy kept in that repository, because the browser reads the CSVs directly.

• Nothing reads the archive at build time, so a build never depends on the archive host.

• .github/workflows/archive.yml adds the versions named in versions/ to the data repository on every push to master, rebuilds the change history there, and pushes both as one commit. It needs an ARCHIVE_REPO_TOKEN secret with write access there.

• To add a version by hand — an export made outside these scripts:

  $ ruby script/archive_release.rb ../elin-chara-viewer-data 'EA 23.306' \
      --channel stable --release-date 2026-05-10 --db /path/to/exported/csv
  $ ruby script/extract_feat.rb --archive ../elin-chara-viewer-data --version 'EA 23.306'
  $ npm run build:history -- ../elin-chara-viewer-data
  

  --db points at the directory holding <version>/*.csv, and --channel is required for a version that no versions/ file names.

• script/restore_archive_from_history.rb imported the versions that only existed in this repository's git history. It was a one-shot; the release flow above is what keeps the archive current.

• The versions older than that were restored from Steam itself, by downloading each build of depot 2135153 and running it so the exporter writes its CSVs. That sweep was a one-shot; its scripts and the manifest list it worked from are attached to #303.

Change history

The character detail pages show what changed about a character, version by version. Computing that from every archived version on request is out of the question, so npm run build:history -- <archive-dir> does it offline and writes one file per character page to history/charas/ in the data repository; the page fetches its own when the accordion is opened. The whole history is rebuilt from scratch on every release, so a version archived out of order or a corrected release date needs nothing else done.

• src/lib/history/viewModel.ts holds PROVENANCE, which says what each raw column feeds. The history hides a raw column once the display group it feeds has already reported the change, so a column that quietly starts feeding a group would show the same change twice. npm run check:history sets each column to another value the game uses and checks that nothing moves undeclared.
• A version that gains a column makes every character look changed, because the schemas fill the older versions in with a default. Those are dropped, and a column addition outside KNOWN_NEW_COLUMNS stops the build instead.
• The build refuses to write when it would lose pages or change the entry count by more than a quarter. --allow-large-change says the difference is meant.

License

MIT License for the source code.

Note that the files under db/ are not covered by this license. They are imported from the Elin game. Also, the files under src/generated/ are generated files from the Elin source code and are not covered by this license.