BREAKER DELTA
浪尖レーダー
OverviewGetting StartedReading the SiteCLI ReferenceConfigurationSkill & MCPExtensionDeploy Your Own

日本語版は準備中です。このページは英語版を表示しています。

Deploy Your Own

One principle first: you own your data

The engine repo (daiyanze/breaker-delta) holds code only — never any instance's collected data. Your editions, ledger, track record and config live in your own data repo; its CI checks out the engine, builds the site from your data, and deploys to the data repo's own GitHub Pages. Upgrading the engine never touches your data.

The data repo's shape

your-data-repo/
  editions/            # one file per edition
  state/               # ledger, vocabulary, sitemap baselines
  track-record.md      # the pick ledger
  config.yaml          # your config (optional; the engine example is the fallback)
  .github/workflows/pages.yml   # build and deploy to this repo's Pages

The working reference is daiyanze/breaker-delta-data: clone its shape, point it at your own data, and the radar is yours.

How the GitHub Pages build flows

  1. In the data repo's Settings, set the Pages source to GitHub Actions.
  2. The data repo's workflow does four things: check out the data repo → check out the engine repo → bun install && bun run bundle → run publish against your data dir and deploy.
  3. Two environment variables point the build at you: TS_BASE_URL (your Pages URL) and TS_DATA_REPO (your data repo). The rest of TS_* is in the CLI Reference.
  4. From then on, every push of new data (for example a scan pushing a new edition) triggers a rebuild — the site updates itself, no manual deploys.

The official instance now runs on Cloudflare Pages (projects connected straight to the repos, built on every push): the main project connects to the data repo and clones the engine during its build, the docs project connects to the engine repo and builds directly, each with its own domain (breakerdelta.com and docs.breakerdelta.com). The build commands are exactly the ones above — only the runner changed from GitHub to Cloudflare; self-hosters can pick either platform.

Domain layout: main site, docs and app live apart

The official domains are split on purpose: the main site sits at the apex (breakerdelta.com — the boards and topic pages are public content that lives on sharing and search, and the apex is the cleanest address for both); the docs are a separate site (docs.breakerdelta.com — this site, built and deployed by its own docs repo running the docs command, never bundled into the main build; the main site's Docs tab simply links here); and the app. subdomain is reserved for a future logged-in product. For your own instance the same shape applies: point TS_BASE_URL at your apex, and either link to the official docs or build your own copy with the docs command.

A real failure worth knowing

A data repo's CI uses its own default token, which cannot read a private repo. If the engine repo is made private, the data repo's build fails at the "check out engine repo" step (repository not found) and the site stops updating — including the rebuilds triggered by scan data pushes. Two fixes: keep the engine repo public (simplest), or store a read-only token for the engine repo in the data repo's Secrets and use it for that checkout.

Updates and day-to-day

  • Engine updates: the data repo's CI builds from the engine's main every time, so a new engine release reaches your site at its next rebuild. Pin the workflow's ref to a tag if you prefer locked versions.
  • Config changes: edit config.yaml in the data repo and push; after the rebuild, verify each source's effective config on the site's Sources page.
  • Verify locally first: run the same publish command locally before pushing (see Getting Started) — if it builds locally, CI almost always builds too.
  • Where a failed build usually breaks: engine checkout = visibility / token; bundling = Bun version (needs ≥ 1.2); publish = run doctor against the data dir first.