benchmark: complete production dashboard QA and accessibility gate - #113
Conversation
TheHiddenObserver
left a comment
TheHiddenObserver
left a comment
There was a problem hiding this comment.
Fresh final review is complete on the current head. I rechecked the deployed-path server/base contract, progressive accessibility hooks, keyboard activation and focus semantics, metric-panel disclosure state, exact chart-data table fallback, generated asset materialization, production Playwright selectors, documentation/navigation integration, and CI wiring. No blocking correctness, accessibility, maintainability, or deployment-path findings remain. Benchmark Frontend CI #489 passes all six jobs, including the existing Chromium regression and the new production QA gate against the committed /docs/assets/benchmarks/ bundle in Chromium, Firefox, and WebKit. There are no unresolved review threads. Recommend squash merge into feature/benchmark-frontend-dashboard.
|
You have reached your Codex usage limits for code reviews. You can see your limits in the Codex usage dashboard. |
|
Bugbot is not enabled for your account, so this pull request was not reviewed. Enable Bugbot in the Cursor dashboard to get automatic reviews on future PRs. |
c815b89
into
feature/benchmark-frontend-dashboard
Summary
Completes #92 with a reproducible product-level QA and accessibility gate for the deployed benchmark dashboard on top of the current PR #76 staging branch.
This PR adds:
docs/assets/benchmarks/bundle rather than only the Vite development server;role=img/accessible descriptions and explicit text-contrast checks;frontend/e2e-production/README.md.No new benchmark family or unrelated statistical implementation is introduced.
Production deployment contract
The production Playwright server serves the repository root and opens the exact nested path:
The suite validates that the committed HTML, hashed JS/CSS, benchmark JSON, parse report, inventory, documentation links, and refresh behavior work from that location without relying on Vite dev-server routing.
Browser matrix
The production suite contains 5 product-level cases and runs them in:
The validated materialization run completed 15/15 production cases successfully (5 cases × 3 browsers). The authoritative clean-head
Benchmark Frontend CI #489repeats the production suite and passes.Interaction QA
The deployed-path suite exercises:
It verifies context-derived controls, deterministic upstream reset, stable scale selection, statgpu backend filtering, external-reference visibility, current rows in the overview table, explicit empty states, and the canonical
LogisticRegressionCVTorch failed disposition.Existing Chromium regression coverage remains in place and runs separately; the production suite extends rather than replaces it.
Accessibility
Primary improvements include:
<html lang="en">and a skip link to the main benchmark results;<main id="dashboard-main">as the primary landmark;:focus-visibletreatment;aria-sorton sortable headers;aria-expanded/aria-controlsfor metric panels;The accessibility enhancements are reapplied after each generated re-render, so newly created controls retain their keyboard/name contracts without accumulating handlers on removed DOM nodes.
Documentation
The English and Chinese benchmark guides now avoid treating mutable source/run counts as timeless prose. They point live counts to the generated inventory and retain the dated 2026-08-07 snapshot:
The dashboard guide documents the current canonical CV source, Cross-validation panel, accessible chart tables, keyboard behavior, and three-browser production QA procedure. The interactive dashboard was already linked from the repository documentation navigation and remains so.
Validation
Authoritative clean-head run on:
All six jobs pass:
The production QA test step itself completed successfully on the committed docs deployment. No temporary implementation/materialization/documentation workflow or helper script remains in the final diff.
Scope boundaries
Tracks and completes #92.