feat(repo): give the site worker one d1 database, emulated locally under pnpm dev

[systemfsoftware-maker]

Lake 1, new layer 2 (Kiro ruling, 2026-10-06), stacked on #62. Restacked onto the RPC layer.

• apps/site/alchemy.run.ts declares Cloudflare.D1.Database('Database') and binds it to the Worker as DB. alchemy dev creates and serves it locally (under apps/site/.alchemy/local) with no cloud account; each cloud stage gets its own database.
• apps/site/src/cloudflare-env.d.ts types cloudflare:workers' env from Alchemy's InferEnv of the Site, so renaming the binding on one side fails the typecheck.
• The health procedure runs one D1 probe (env.DB.prepare('SELECT 1')) and hands its outcome to src/api/check-health.workflow.ts, a pure Workflow.make decision: healthy when the probe answered, otherwise the tagged refusal DatabaseUnreachable, which is the procedure's typed error. The home page shows unreachable on any failure, defects included.
• Queries go through the D1 binding itself: no effect/sql, no @effect/sql-d1.
• README: the token needs Workers and D1 edit rights.

Gate after the cycle 52 restack (Linux only)

Gate at 7f70d04, clean worktree, Linux, sandboxed:

$ pnpm bootstrap                    # exit 0
$ pnpm check:ci                     # exit 1, only check:sfs-sources:
check-sfs-sources: 11 @systemfsoftware/* package(s) resolve from the npm registry, not the systemfsoftware flake
 Tasks:    9 successful, 9 total   # lint, typecheck, test, build
 Tasks:    1 successful, 1 total   # dist
ok | 1 passed (9 steps) | 0 failed  # sandbox proofs
$ pnpm journeys                     # exit 0
Tests  2 passed (2)   # health over RPC, strict CSP

CI run 37573598160 on 7f70d04: 6 of 7 jobs pass (format, lint, typecheck, test, dist, journeys); check (sfs-sources) fails, the same red as the local gate above.

Sabotage: renaming the binding in alchemy.run.ts fails the typecheck (Property 'DB' does not exist on type 'Env'). Breaking the query (SELECT FROM) fails the health journey with Worker health: unreachable.

Cycle 35: the site's own test (Kiro rulings)

After F5 dropped --passWithNoTests, the guestbook's property test was the site's only test, so removing the guestbook left check:ci red on @endgame/site#test (No test files found). Ruling: the site gets a test of its own that survives removal, as a pure decision with a property, on this layer.

• apps/site/src/api/check-health.workflow.ts: command CheckHealth { probe: ProbeAnswered | ProbeUnanswered }, decision Healthy (private to the file), refusal DatabaseUnreachable. The handler only runs the probe and calls it.
• apps/site/src/api/__tests__/check-health.workflow.property.test.ts: ∀c_Healthy_=ProbeAnswered, drawing the command from its production schema: the workflow succeeds exactly when the probe answered, and otherwise refuses with DatabaseUnreachable.
• The site's test and mutation toolchain moves here from the guestbook layer with the first workflow: vitest, @systemfsoftware/vitest, the four @systemfsoftware/stryker-* packages, @systemfsoftware/effect-cell-types, vitest.config.ts, stryker.config.ts, the test (vitest run) and mutation scripts, and stryker.mutate: ["src/**/*.workflow.ts"]. The release-gate planner now plans @endgame/site (packages=["@endgame/site"]), and check:sfs-sources counts 11 packages here, the F1 gap that closes with stryker-js-effect#196's flake.

$ pnpm --filter @endgame/site test        # at 9d4a3a4
✓ src/api/__tests__/check-health.workflow.property.test.ts (1 test)
# sabotage: decide's arms swapped (answered → DatabaseUnreachable, unanswered → Healthy)
× ∀c_Healthy_=ProbeAnswered
PropertyRefuted: … Shrunk input: {"command":{"probe":ProbeUnanswered}}
# restored (the replay seed file the failure wrote deleted): 1 passed
$ pnpm journeys
Tests  2 passed (2)                       # health over RPC, strict CSP

Gate after cycle 83 (every @systemfsoftware/* package from our flakes, #52)

Gate at 89b434f, clean worktree, Linux, sandboxed:

$ pnpm bootstrap                    # exit 0
$ pnpm check:ci                     # exit 0
check-sfs-sources: all 25 @systemfsoftware/* packages resolve from the systemfsoftware flake
 Tasks:    9/9 successful, 9/9 total   # lint, typecheck, test, build
 Tasks:    1 successful, 1 total   # dist
ok | 1 passed (9 steps) | 0 failed  # sandbox proofs
$ pnpm journeys                     # exit 0
Tests  2 passed (2)   # health over RPC, strict CSP

CI run 37676551794 on 89b434f: all 7 jobs pass, check (sfs-sources) included. The exit 1 gates above are from before the stryker packages came from the stryker-js-effect flake.

[kiro-systemf]

Verified: 9/9 green on 85438ee, diff = d1's 13 files, 0 threads, hunt clean.