Add a Bun health script that exercises top-level, dynamic, and representative video routes against one or more hosts so prod regressions are visible from a single command.
Device pages now fall back to the bundled device list when the external API misses a slug, and orphaned tv slugs redirect to /benchmarks instead of returning a 500. Video fallback logic reuses the existing YouTube-to-listing builder so route reconstruction stays aligned with the current build logic.
Constraint: The external API host can lag behind the frontend build and omit per-slug JSON files that public routes still expect
Rejected: Import the generated video list directly | static/video-list.json is too large for a safe SSR fallback
Rejected: Leave missing tv routes as 500s | a stale public URL should degrade to a useful redirect instead of breaking the request
Confidence: medium
Scope-risk: moderate
Reversibility: clean
Directive: Keep route fallbacks tied to build-time artifacts from the same repo so frontend and fallback data stay in sync
Tested: bun scripts/health http://127.0.0.1:4322; vitest ./test/prebuild/config-node.test.js ./test/prebuild/site-listings.test.js; pnpm run netlify-build
Not-tested: live production deploy before push
Dynamic Astro routes were reading Netlify redirect config through a cwd-relative path, which is fragile inside a serverless runtime and was taking detail pages down with 500s before render.
Resolve netlify.toml by searching from the module directory and current working directory, and fail open in request-time redirect lookup so a config read problem does not block page rendering.
Constraint: Netlify serverless cwd is not guaranteed to be the repo root
Rejected: Inline redirects into route modules | would duplicate platform config and drift from source of truth
Rejected: Leave redirect lookup hard-failing | one config read failure should not take down unrelated pages
Confidence: medium
Scope-risk: narrow
Reversibility: clean
Directive: Keep redirect config lookup independent of process cwd anywhere server code reads deploy config files
Tested: vitest ./test/prebuild/config-node.test.js; pnpm run netlify-build
Not-tested: live Netlify production deploy before push
Madge treated the local plist parser module name as a circular dependency during the Netlify build lane after the TypeScript refactor. Rename the local module to plist-parser so the internal file no longer collides with the external plist package name, while keeping parser behavior unchanged.
Constraint: Must restore the deploy gate without changing parser semantics
Rejected: Disable the circular-dependency check | would hide a useful guard instead of fixing the naming conflict that triggered it
Confidence: high
Scope-risk: narrow
Reversibility: clean
Directive: Avoid naming local modules after direct external package imports when the repo relies on static dependency graph tooling
Tested: pnpm run typecheck; pnpm exec vitest run test/scanner/plist.test.ts test/scanner/file-api.test.ts test/scanner/client.test.ts test/prebuild/load-sitemap-endpoints.test.ts; pnpm run test-prebuild
Not-tested: Full production deploy completion before push
The TypeScript helper refactor exposed a separate production risk: builds can fail when the sitemap-endpoints API returns a transient 5xx during Pagefind index generation. Move that fetch into a small helper with retry logic and add a focused prebuild test so this failure mode is caught without waiting on a full deploy.
Constraint: Must keep Pagefind index generation behavior the same when the endpoint is healthy
Rejected: Ignore the failure as external-only | transient 5xx responses can still block deploys and CI repeatedly
Confidence: high
Scope-risk: narrow
Reversibility: clean
Directive: When external build inputs can fail transiently, add a small retryable helper plus a direct test before relying on full deploy verification
Tested: pnpm exec vitest run test/prebuild/load-sitemap-endpoints.test.ts; pnpm run typecheck; pnpm run test-prebuild
Not-tested: Full production redeploy completion before push
Convert the scanner's plist parser and Node-style file shim to TypeScript and add small unit tests so common parser and file-reader failures are caught before we need to lean on Playwright or the higher-level scanner test.
Constraint: Must preserve current scanner behavior while tightening the lowest-level helper surface
Rejected: Jump straight to Mach-O parser conversion | harder to isolate regressions without first proving the smaller helper-test pattern
Confidence: high
Scope-risk: narrow
Reversibility: clean
Directive: Add small module-level regression tests for repeatable scanner breakage before expanding browser coverage
Tested: pnpm run typecheck; pnpm exec vitest run test/scanner/plist.test.ts test/scanner/file-api.test.ts test/scanner/client.test.ts; pnpm run test; pnpm run test:browser
Not-tested: Production deploy behavior prior to push
Move the worker scanner surface into TypeScript, add a direct worker regression, and make the version=2 app-test path populate the same visible result data and final status as the legacy scanner. This keeps the refactor bounded while making the worker route safe to exercise.
Constraint: Must preserve the existing Apple Silicon app-test behavior while changing the worker internals
Rejected: Flip production to the worker path immediately | still needs the normal deploy path and broader production soak
Confidence: medium
Scope-risk: moderate
Reversibility: clean
Directive: Keep the version=2 adapter using the shared finishFileScan path until the legacy scanner can be removed entirely
Tested: pnpm run typecheck; pnpm exec vitest run test/scanner/client.test.ts; pnpm run test:browser (original workspace); netlify build --context deploy-preview (original workspace)
Not-tested: Browser suite from the clean clone environment (local Astro dev server startup timed out there)
Add a typed Playwright harness for Pagefind and the Apple Silicon app-test flow so scanner work has browser-level protection. Keep the rollout plan in the same stack so the TypeScript conversion stays staged and reviewable.
Constraint: Must not change production runtime behavior in this commit
Rejected: Leave the old JS browser test and add a second harness | duplicates setup and leaves the targeted browser script broken
Confidence: high
Scope-risk: narrow
Reversibility: clean
Directive: Keep browser-only helpers under test/playwright/support until the runtime scanner surface is fully typed
Tested: pnpm run typecheck; pnpm run test:browser; pnpm run test:browser:pagefind
Not-tested: Live browser checks against doesitarm.com
Cover the Native Support filter with a Playwright-backed Vitest case
that can boot the local dev server or attach to a deployed URL so the
same regression can gate post-deploy verification.
Add Pagefind indexing and browser search adapters behind a provider switch.
This lets prebuild generate either Stork or Pagefind search artifacts and lets the existing search UI run against Pagefind while preserving scoped filters, excerpts, and result metadata.
Share filter parsing and provider mapping logic between Stork and upcoming search adapters.
This keeps the existing Stork API stable while adding Pagefind-oriented filter serialization under test.
