feat(xtest): decision-first benchmark report with provenance and artifact links

[dmihalcik-virtru]

Stage 6 of the six-stage plan. Stacked on #660.

What it does

This brings back #621's report presentation, rebuilt on the structured result from stages 3–5:

• perf/presentation.py is rendering only: it never decides a verdict, a correction family or a winner. It shows:
  • One status: PASS / REGRESSION / INCONCLUSIVE / PARTIAL / UNTRUSTWORTHY / NOTHING MEASURED / NEUTRAL.
  • Compared builds, with requested refs, commits and diff links taken from the installation manifest (the same record locally and in CI).
  • A warning when the budget was too small for the requested precision.
  • The attention table with the tail-compressed effect strip.
  • Candidate comparisons and winners from perf.conclusions.
  • Braille round traces, every measurement, skipped cells, and run facts.
• perf/publish.py fills the artifact URL into the summary template once upload-artifact returns it, or drops those links locally. No placeholder survives either way.
• Output files: pytest writes <sdk>.md (the summary), <sdk>-table.md (the plain table) and <sdk>.summary.md (the template). CI defers the step summary until the upload's URL is known.

Provenance links use tree/<sha>, commit/<sha> and compare/<a>...<b>. Those are correct whatever kind of ref an arm was, which #621's release/PR URL guessing wasn't.

Validation

• 345 passed. ruff, ruff format and pyright are clean. actionlint reports no new findings.
• Tests cover:
  • each status word;
  • partial coverage listing what was skipped;
  • winner text matching the conclusion;
  • commit and diff links;
  • "provenance unavailable" when the manifest is missing;
  • huge effects and unavailable intervals rendering legibly;
  • markdown-sensitive characters and long arm names;
  • publishing with and without a URL, and with no GitHub env;
  • rendering leaving every gate decision and winner unchanged.
• Ran a real pytest session against a neutral manifest: the summary renders with NEUTRAL and no placeholder.
• Not yet run: a CI dispatch to check that the artifact link resolves.

[sonarqubecloud]

Quality Gate passed

Issues
2 New issues
0 Accepted issues

Measures
0 Security Hotspots
0.0% Coverage on New Code
0.0% Duplication on New Code

See analysis details on SonarQube Cloud

[coderabbitai]

Important

Draft PR not reviewed

Draft PRs are not automatically reviewed by default.

• Trigger a manual review

To automatically review draft PRs, update your CodeRabbit configuration:

reviews:
  auto_review:
    drafts: true

• Autofix · Keep fixing CodeRabbit findings and required CI, and resolving merge conflicts

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}

[github-actions]

X-Test Results

✅ java@v0.19.1-main
✅ go@main-main
✅ js@main-main
✅ go@main-v0.29.0
✅ java@v0.19.1-v0.29.0
✅ java@main-v0.29.0
✅ js@main-v0.29.0
✅ go@v0.38.1-v0.29.0
✅ go@v0.38.1-main
✅ js@v0.4.0-v0.29.0
✅ js@v0.4.0-main
✅ java@main-main