fix(test): stop worker test from classifying the whole fork as production

docs/production-incident-baseline.md

@@ -21113,3 +21113,47 @@ Drizzle schema and runtime Zod schemas. Findings and remediations:
 - **Remaining risk / follow-up:** Group coupled CodeQL and Slack dependencies
   in Dependabot configuration so future updates arrive atomically. Continue to
   require a native Release build for Sentry React Native updates.
+
+## 2026-07-30 — Test Env Mutation Re-Enabled Production Classification for Local Runs
+
+- **Status:** Recurrence fix implemented in test/config paths; production
+  unaffected.
+- **Symptoms:** Every exception ingested into error tracking over the preceding
+  days was a unit-test fixture string (for example `test failure`,
+  `test worker error`, `invalid FIT`, `unrecoverable sync error`,
+  `Redis job log unavailable`) rather than a real failure. The events clustered
+  in a single short window, all attributed to the `dofek-server` distinct id,
+  with stack frames pointing at `@vitest/runner` and a local Conductor
+  workspace path.
+- **User impact:** No production user impact. The blast radius is the error
+  tracking surface itself: fixture noise buries genuine production exceptions
+  and each distinct fixture spawns its own "new issue" alert.
+- **Evidence:** `src/jobs/worker.test.ts` assigned
+  `process.env.DEPLOY_ENVIRONMENT = "prod"` and `process.env.SENTRY_DSN`
+  directly inside `vi.hoisted()`. Those raw assignments were never torn down,
+  so the "prod" classification persisted for the whole vitest fork and defeated
+  the production-only guard in `initProductionSentry()`
+  ([src/lib/sentry.ts](../src/lib/sentry.ts)) for every module that ran in the
+  same fork. `@sentry/node` is mocked in that test, so Sentry stayed clean, but
+  any un-mocked telemetry sink wired to the same guard emitted the fixture
+  errors as real production events.
+- **Root cause:** A recurrence of the failure mode fixed by commit 7eba92d
+  ("Prevent local errors from reaching production Sentry",
+  [#1882](https://github.com/Asherlc/dofek/pull/1882)), which added the
+  `DEPLOY_ENVIRONMENT`-gated guard. The worker test tricked that guard back
+  open by mutating the shared process environment and never restoring it.
+- **Fix / mitigation:** Replace the raw `process.env` assignments with tracked
+  `vi.stubEnv` calls and restore them with `vi.unstubAllEnvs()` in an
+  `afterAll`, so the "prod" classification cannot outlive the test file within
+  a reused fork. Add an explicit non-production `DEPLOY_ENVIRONMENT: "test"`
+  default to the shared test env in
+  [vitest.config.ts](../vitest.config.ts) so no test run is classified as a
+  production deployment unless it opts in. No production code was changed and
+  no `NODE_ENV === "test"` branch was added.
+- **Validation:** `src/jobs/worker.test.ts` and both `sentry.test.ts` suites
+  pass, the full Docker-free unit tier passes, and lint and TypeScript checks
+  pass.
+- **Remaining risk / follow-up:** When a PostHog (`posthog-node`) error
+  exporter lands on `main`, mock it in `worker.test.ts` the same way
+  `@sentry/node` is mocked so no test run can emit real telemetry regardless of
+  environment classification.

src/jobs/worker.test.ts

@@ -1,12 +1,19 @@
-import { describe, expect, it, vi } from "vitest";
+import { afterAll, describe, expect, it, vi } from "vitest";
 
 // All mock dependencies live inside vi.hoisted() so they are guaranteed to exist
 // before vi.mock() factories resolve and before the static import of worker.ts.
 // This satisfies vitest's "related" mode used by Stryker in CI, which requires a
 // static import dependency between the test and the source module.
 const hoisted = vi.hoisted(() => {
-  process.env.DEPLOY_ENVIRONMENT = "prod";
-  process.env.SENTRY_DSN = "https://test@sentry.io/123";
+  // Classify this fork as a production deployment so worker.ts initializes the
+  // production Sentry client on import. Use vi.stubEnv (not a raw process.env
+  // assignment) so the change is tracked and torn down after this file runs —
+  // otherwise the "prod" classification leaks into every other test module in
+  // the same vitest fork and defeats the production-only guard in
+  // initProductionSentry(), which is how local test-fixture errors previously
+  // reached production error tracking.
+  vi.stubEnv("DEPLOY_ENVIRONMENT", "prod");
+  vi.stubEnv("SENTRY_DSN", "https://test@sentry.io/123");
 
   function noOpExit(): never {
     throw new Error("process.exit called unexpectedly in test");
@@ -282,6 +289,12 @@ const {
 // test file as related to worker.ts, so mutations in worker.ts are covered.
 import "./worker.ts";
 
+// Restore the deployment-environment/DSN stubs so the "prod" classification does
+// not outlive this file within a reused vitest fork.
+afterAll(() => {
+  vi.unstubAllEnvs();
+});
+
 describe("worker module", () => {
   // 2 per-provider workers (strava, garmin) + 1 legacy sync + 1 import + 1 FIT import
   // + 1 FIT batch + 1 ZIP extract + 1 export + 1 scheduled-sync + 1 post-sync

vitest.config.ts

@@ -45,6 +45,14 @@ const sharedTestEnv = {
   CREDENTIAL_ENCRYPTION_KEY_NAMESPACE: "dofek-test",
   CREDENTIAL_ENCRYPTION_KEY_NAME: "provider-credentials-test",
   PUBLIC_URL: "https://app.example.test",
+  // Classify the test environment as non-production by default so no test run is
+  // ever treated as a production deployment. The production-only guard in
+  // initProductionSentry() keys off DEPLOY_ENVIRONMENT; an explicit non-prod
+  // default keeps a stray env mutation in one test from letting local
+  // test-fixture errors reach production error tracking. Tests that need to
+  // exercise the production path must opt in explicitly via vi.stubEnv and
+  // restore it afterwards.
+  DEPLOY_ENVIRONMENT: "test",
 };
 
 const configuredClickHouseUrl = process.env.CLICKHOUSE_URL?.trim();