[pomponchik]

Benchmark results now preserve separate runs of the same tooling version instead of overwriting previous measurements. Clients can send run_id to replay an upload safely; conflicting replays return 409, and incremental suite uploads retain their original server timestamp. Migration 0018 preserves existing rows and rejects downgrades that would lose run history.

The computer page shows dated run history in the browser timezone and improves responsive layouts. History navigation follows delayed data and browser layout adjustments while preserving manual scrolling, keyboard input and focus. A viewport resize regression covers scroll clamping without user input. API, browser and visual coverage and documentation are updated.

Hugo and the platform pin reboop 0.1.37, and HTTP lint pins reboop-eslint 0.1.37. Both npm publications and GitHub release archives were verified; their code matches 0.1.34. No other dependencies changed.

The non-expiring token rollback regression starts at revision 0017 and checks the complete reflected schema, revision and credential rows before returning to head, keeping its data-loss assertions independent of later migrations. Production token migrations are unchanged.

Validation:

• On the final source snapshot, lint, production build and 87 unit tests pass. All 230 benchmark browser scenarios pass across desktop, mobile, Firefox, WebKit and iPhone, including manual scroll, focus and keyboard checks. The two affected mobile navigation scenarios also pass 20 repeated cases.
• The browser scroll failure was reproduced before the fix. Diagnostics confirmed that layout changed scrollY with history focused and no user interruption; the new resize regression fails on the previous implementation.
• All 131 targeted SQLite API tests pass, including retry/conflict/concurrency/recovery and populated migration checks. Ruff passes and PostgreSQL upgrade SQL compiles. The token rollback failure was reproduced before its fix; six Python 3.11 and twelve Python 3.12 migration cases pass, and deliberate schema/data corruption is caught before restoration.
• Dependency consumer checks pass six lint configuration tests, 87 unit tests, lint, and production builds of both the platform and Hugo.

Full push and pull_request CI, including real PostgreSQL and Linux visual comparisons, must pass on the final head before merging. Production deployment and Ansible verification are required after merge.

Header screenshot diagnostics persist mismatched before/after PNG files directly in test output so CI artifacts retain them with the default reporter. Strict byte comparison, the header renderer and frozen reference are unchanged. A previously observed high-contrast focus mismatch did not recur in either complete current x64 CI run: all 290 header cases pass in each run, alongside ten macOS repeats and 34 Linux ARM cases. Its rendering cause remains unconfirmed; no rendering fix is claimed. A deliberate outline mismatch confirms that both PNG files are retained.