We classified these failures as environment_state rather than a code regression. Both tasks fail with transient network errors (ECONNRESET during prerendering, fetch failed in the browser E2E test) that have no traceable connection to the PR's setResponseHeaders iteration fix — notably, the multiple Set-Cookie headers are preserved on redirect test, the closest proxy for the PR's changes, passes cleanly. With 0% historical flakiness and no similar failures found in other branches, these point to a transient CI network issue rather than anything introduced by this PR.
Self-Healing CI
An empty commit was applied to the branch to trigger a new CI pipeline execution to resolve the environment issue.
Both failing tasks belong to the tanstack-react-start-e2e-basic project, which is not in the list of touched projects (only root and @tanstack/start-server-core were modified by this PR).
The PR's changes are scoped to setResponseHeaders in packages/start-server-core/src/request-response.ts — fixing Headers iteration logic — and the accompanying unit tests. Nothing in this change could produce network-level TCP errors.
tanstack-react-start-e2e-basic:build:vite:prerender The vite client and SSR builds complete successfully. The failure occurs in the post-build prerender crawl phase, where the prerender process makes an HTTP request to the local server and receives: TypeError: fetch failed → caused by Error: read ECONNRESET (errno: -104) This is a TCP connection reset — the server dropped the connection mid-response. This is a transient infrastructure/environment failure with no logical connection to the header-iteration fix in this PR.
tanstack-react-start-e2e-basic:test:e2e--vite-spa 133 of 139 tests pass, including redirects › multiple Set-Cookie headers are preserved on redirect, which is the test most closely related to the PR's changes (and it passes). The single failing test — redirects › internal › internal target, navigation: thrower: beforeLoad, reloadDocument: true, preload: false — fails because the browser accumulates 35 Error: fetch failed console errors, causing the test's zero-errors assertion to fail. These are network fetch failures at the browser level, not header assertion failures.
Why not flaky_task Both tasks have a recorded flakiness rate of 0%, so the failures are not historically non-deterministic. This rules out flaky_task.
Why not code_change There is no traceable path from the PR's changes (iterating Headers entries instead of Object.entries) to an ECONNRESET or browser fetch failed error. The redirect tests most relevant to the PR (cookie preservation on redirect) all pass. The failures are network-layer issues in the CI environment.