docs(ci): final tally for the accel contrast — TCG 3/3 on macos-x86

[mobileskyfi]

Follow-up to #136. Refs #110, #76 — not linked to close either; the wedge mechanism is unexplained and #76's done-when is unsatisfied.

The accel contrast finished after #136 merged. All three accel=tcg legs completed 12/12 with job conclusion success (1939 / 2263 / 1862 s), with provisioning.test.ts passing at position 10 in 604 / 810 / 644 s. So the tally is HVF 0/7 against TCG 3/3 through the condition that has defeated this platform since the program began — Fisher exact two-sided p = 0.0083 (1/C(10,3)), down from 0.028 at 2/2. The merged 2/2 text quoted 0.056, which was the doubled one-sided convention; line 224 now uses the standard definition, matching the 0.029 already quoted at line 271.

The consequence worth writing down is the durable one: ci-data/tested-versions.json now carries macos-x86: pass for 7.23.3, the first row this platform has ever had. The fold accepts only scope:"full" records, which is exactly why it was absent across all ten versions — every prior macos-x86 ndjson is scope:"filtered" because no full suite had ever completed.

Docs only; no code change. Evidence: #76 (comment)

🤖 Generated with Claude Code

Summary by CodeRabbit

• Documentation
  • Updated CI findings to reflect three successful TCG full-suite runs on macOS x86.
  • Added run durations, provisioning timings, and statistical validation details.
  • Recorded RouterOS 7.23.3 as passing on macOS x86.
  • Clarified that only full-scope test records count toward tested-version results.

[coderabbitai]

📝 Walkthrough

Walkthrough

Updates the CI instructions doc to reflect three successful TCG full-suite runs on macos-x86 (up from two), with run durations, provisioning timings, and a Fisher exact test result. Adds the first macos-x86 pass entry for RouterOS 7.23.3 to tested-versions.json documentation.

Changes

CI documentation update

Layer / File(s)	Summary
TCG result and tested-versions entry .github/instructions/ci.instructions.md	Result changes from 2/2 to 3/3 successful TCG full-suite runs. Adds timings and a Fisher exact test result. Records the first full-scope macos-x86: pass entry for RouterOS 7.23.3 in tested-versions.json.

Estimated code review effort: 1 (Trivial) | ~2 minutes

Possibly related PRs

• tikoci/quickchr#75: Both PRs modify .github/instructions/ci.instructions.md for TCG full-suite behavior and the full-scope rollup rule.

🚥 Pre-merge checks | ✅ 5

✅ Passed checks (5 passed)

Check name	Status	Explanation
Docstring Coverage	✅ Passed	No functions found in the changed files to evaluate docstring coverage. Skipping docstring coverage check.
Linked Issues check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Description Check	✅ Passed	Check skipped - CodeRabbit’s high-level summary is enabled.
Title check	✅ Passed	Title accurately summarizes the main change: documentation of TCG 3/3 test results on macos-x86 following the accel contrast comparison.

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Commit unit tests in branch docs/tcg-contrast-final

---

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.}

[Copilot]

🟡 Not ready to approve

The updated doc states a “Fisher two-sided” p-value that appears inconsistent with the standard Fisher exact two-sided definition and should be corrected/clarified for accuracy.

Once you've addressed the issues Copilot identified, you can request another Copilot review.

This review doesn't count toward merge requirements. Sign up for the private preview to control whether Copilot approvals count.

Pull request overview

This PR updates the CI documentation (.github/instructions/ci.instructions.md) to record the final “HVF vs TCG” completion tally on macos-x86, including the claim that TCG achieved 3/3 full-suite successes and that tested-versions.json gained its first-ever macos-x86: pass row.

Changes:

• Updates the documented HVF/TCG win–loss tally to reflect TCG 3/3 full-suite successes.
• Adds supporting narrative details (durations, job conclusion status, and why scope:"full" matters for folding into tested-versions.json).

File summaries

File	Description
.github/instructions/ci.instructions.md	Updates CI instructions to record the completed accel-contrast results and their implications for macos-x86 historical metrics.

Review details

• Files reviewed: 1/1 changed files
• Comments generated: 1
• Review effort level: Lite

We're testing this review assessment. Please use 👍 or 👎 to tell us if it's correct.

[mobileskyfi]

Verified Copilot's finding rather than taking it at face value — it is correct, and the fix in 6efcaa1 is right.

For HVF 0/7 vs TCG 3/3 the margins are asymmetric (3 successes, 7 failures, n=10), so only the observed table has probability ≤ the observed one; the opposite extreme (all 3 successes in HVF) has p = 35/120 = 0.29 and is not summed in. Two-sided therefore equals one-sided:

k = TCG successes	P(k)
0	35/120 = 0.2917
1	42/120 = 0.5250
2	21/120 = 0.1750
3	1/120 = 0.00833

Two-sided p = 0.0083. The previous 0.0167 was 2× the one-sided value — the doubling convention, which the merged 2/2 text also used (0.056 where the standard definition gives 0.028).

The consistency point holds too. Line 271 (4/4 green outside position 10 vs 4/4 wedged at it) has symmetric 4/4 margins, so both extremes qualify and two-sided = 2 × 1/70 = 0.029 — the number already in the doc. Both conventions coincide there, which is why that line never looked wrong. Line 224 now uses the same standard definition.

Checked the rest of the repo for the superseded values (0.0167 / 0.017 / 0.056) — no remaining occurrences; the PR description has been corrected as well.