[joshunrau]

Fixes 26 small, independent Difficulty: Low bugs on one branch. Each fix is its own commit with its own Closes #N, so this PR can be reviewed and reverted one commit at a time, and each commit becomes one CHANGELOG entry.

How this was produced

1. Scout. I listed the 230 open issues and the 15 open PRs. I dropped issues an open PR already fixes: Date limit to date of birth #1820 (Add dob limit #1821), Dashboard Records & Sessions chart has no legend, so its two series can only be told apart by colour #1782 (fix(web): add accessible legend to dashboard trends #1819), CSV upload silently truncates RECORD_ARRAY values with a missing closing parenthesis or a colon in the value #1705 (fix(web): stop the record array upload parser truncating values #1818), The playground instrument selector's search matches instrument UUIDs instead of labels, so typing a name finds nothing #1764 (fix(playground): search instruments by label #1817), Blog index page title reads "Open Data Capture - Open Data Capture" #1775 (fix(outreach): use the blog title in metadata #1815), odc-cli crashes with a TypeError traceback when an error response has an empty body #1779 (fix(odc-cli): report the status for an error response with an empty body #1809), pnpm test:e2e replays a cached pass after app code changes because the task hash covers only the testing package #1766 (fix(testing): update turbo task graph for test:e2e cache #1808), and the upgrade chores in build: upgrade dependencies held back by breaking changes #1660. I also dropped issues whose fix would touch a source file an open PR also changes, to avoid merge conflicts: the datahub work in feat: add the instrument hub and record provenance #1807, email templates in fix(web): scope email template languages to active languages #1810, odc-cli in fix(odc-cli): report the status for an error response with an empty body #1809, and the playground selector in fix(playground): search instruments by label #1817. Outreach-docs and CI-script issues were left out because testing/ cannot exercise them end to end.
2. Triage. 8 read-only agents checked 55 candidates against main. For each one they recorded whether the bug still reproduces (all but Stroop Task never calls done, so it can never be completed #1634 do; the Stroop task was removed), a 1–5 simplicity score, open decisions, and concrete fix, unit-test and e2e plans. I kept the 26 that scored as trivial, had no open design question, and had a feasible e2e test.
3. Implement. 10 agents worked in parallel, one per area, each in its own git worktree. Each wrote one commit per issue: the fix, a unit test that fails on the old code, and an e2e spec appended to an existing file. During this step they ran only scoped vitest with coverage and scoped lint; Playwright was held for one run at the end.
4. Adversarial review. Every commit was checked by 2 independent reviewers told to assume it was wrong. One looked at correctness and completeness. The other checked whether the tests actually pin the bug and whether the commit follows the repo rules. 5 commits had blocking findings (About page shows the Core API uptime from when the app was loaded, not when the page is opened #1724, Login page heading shows "Open Data Capture" on narrow screens when the instance name is not set in the reader's language #1704, Profile form on /user discards unsaved edits whenever the user query refetches, such as on returning to the tab #1720, Measures declaring the deprecated hidden: false are hidden from the instrument summary #1694, End Current Session in the mobile menu ends the session without confirmation and discards an instrument in progress #1709). Each was revised once and passed a second two-reviewer round. For example, the first About page shows the Core API uptime from when the app was loaded, not when the page is opened #1724 fix failed on any revisit within the router's 5-minute match cache; the fix now uses staleReloadMode: 'blocking'.
5. Integrate and verify. I cherry-picked all commits onto this branch, autosquashed the revisions into their issue commits, then ran the full gate. It caught 3 broken e2e tests and 2 pieces of doc drift, all fixed into their issue commits (details under Verification).

Issues

Area	Issues
web: About	#1724 uptime fetched fresh on every visit · #1643 counter no longer a second ahead
web: audit logs	#1663 first Time-header click sorts oldest first · #1661 loader prefetches the key the table reads
web: login / branding	#1704 narrow-screen heading falls back through languages · #1676 hidden login image no longer blocks Save
web: account	#1720 refetch keeps unsaved profile edits · #1735 rejected password explained in the dialog · #1698 (schemas) international and grouped phone numbers accepted
web: mutations	#1738 repository import/sync no longer aborted at 10 s · #1731 five self-handling mutation hooks no longer hit the error boundary
web: misc	#1638 upload list render loop · #1644 paste parse errors numbered by file line · #1667 Add Group keeps the form on failure · #1709 mobile End Session asks for confirmation
react-core	#1694 deprecated hidden: false measures shown · #1743 dropzone reports "too many files" and keeps the earlier selection
playground	#1805 static-assets example submits · #1754 Save disabled for examples · #1753 closing a background tab keeps the active file · #1740 deleting the selected instrument loads the default template
instrument-bundler	#1800 no spurious reading 'inputs' error · #1799 scoped-package subpaths resolve to .js
instrument-library	#1678 lowercase postal code and immigration age 0 accepted · #1677 copy errors · #1674 Happiness error shown under its field

Verification

• pnpm lint: green, with no eslint --fix rewrites.

• pnpm test:coverage: 388 files and 4409 tests green, still at 100%. vitest list --filesOnly collects all 4 new test files.

• pnpm test:e2e: 308 of 309 passed. The one failure was a pre-existing accessible-instruments test that timed out in browser.newContext while its fixture was being set up, before the test body ran. It passes when rerun on its own.

• The full run caught 3 broken new e2e tests, all fixed in their issue's commit:

  • Profile form on /user discards unsaved edits whenever the user query refetches, such as on returning to the tab #1720: useSuspenseQuery clamps staleTime to 1 s, so an early focus event never refetched. The test now retries until the refetch happens.
  • Closing a background tab in the playground editor switches the active file to a different tab #1753: the click landed on a pointer-events-none input.
  • About page uptime counter runs one second ahead #1643: clock.pauseAt could target a time that had already passed.

  The About page shows the Core API uptime from when the app was loaded, not when the page is opened #1724 revisit test was also flaky through the user dropup, so it now revisits with page.goForward().

• The Profile form on /user discards unsaved edits whenever the user query refetches, such as on returning to the tab #1720 and About page shows the Core API uptime from when the app was loaded, not when the page is opened #1724 e2e tests were also run with the fix removed. Both fail, showing the stale value, so they really pin the bug.

• Doc drift fixed with the code: .agents/docs/playbooks/add-web-data-hook.md used the very loader call Audit logs loader prefetches under a different query key than the table reads, so every visit fetches the logs twice #1661 corrects as its example, and the disableDefaultTimeout comment in apps/web/src/services/axios.ts now names repository imports.

Worth a look

• Playground static-assets example throws ReferenceError on Submit because message is undefined, hidden by another example's ambient global #1805 has no unit test, because the playground's example instruments have no unit tier. It is covered by the playground e2e spec only.
• About page shows the Core API uptime from when the app was loaded, not when the page is opened #1724 loads About with fetchQuery({ staleTime: 0 }) and staleReloadMode: 'blocking' rather than ensureQueryData. This is deliberate, because the uptime must be live, and a comment beside the loader says so. A failing /v1/setup now shows the error page on About instead of a stale value.
• Adding or syncing an instrument repository reports failure after 10 seconds while the import keeps running #1738 removes the client timeout entirely for repository create and sync, matching setup. A stalled server now leaves the dialog waiting.
• Five mutation hooks that handle their own errors still replace the page with the error page on failure #1731 has e2e coverage only for the refused-import path. The other four hooks are covered by unit tests.
• Add Group navigates away before the create request settles, so a rejected group loses the form and shows only a generic 409 #1667 shows the API's conflict message, which is English only, under a translated fallback.

Co-Authored-By: Claude Opus 5.5 noreply@anthropic.com