Skip to content

fix(validation): preserve representable recovery metrics - #488

Draft
seonghobae wants to merge 541 commits into
mainfrom
fix/validation-bias-overflow-safe-mean
Draft

fix(validation): preserve representable recovery metrics#488
seonghobae wants to merge 541 commits into
mainfrom
fix/validation-bias-overflow-safe-mean

Conversation

@seonghobae

@seonghobae seonghobae commented Sep 3, 2026

Copy link
Copy Markdown
Contributor

Validation Evidence authority

Protected base is main@a243f18da4a4ca8a8d068c39922537f1f8ed6ad0. This Draft remains the owner-correct Validation Evidence landing vehicle. Production validation_core::bias_standard_error exact pair-distance admission is deliberately unchanged at n=4..=16; issue #491 and docs/research/bias-standard-error-wide-linear-admission-bound.md carry the measured exact-proof budget decision and scientific trace.

GAP-125 remains the latest public-value repair: RED 5da82b2d651706c191ca191c6c077d916cbfda25 → repair a509ae9e46c8ffc2cc3ef4f0e904774ad2516e1f. The authoritative exact-rounding primitive was subsequently widened for two-limb scaled comparisons by RED f7717361ad8c5f0592688c1514c104cc1b4adabe → repair e4a85f53a611922be7492fe906d62ce65787c18e, with tie-to-even edge contract 1240ace8eb41a01fa72a4bb99df842fd550a1288.

#491 resource finding now under test

7a2ab0a1ef7a72d8cc9b9253d7f92c493e578943 proves pairwise O(n²) and Wide256 O(n) exact-ratio equivalence on represented {0,1,2^53} geometries at n=4,16,17,65,257,2050, including the 129-bit cancellation-product boundary at n=2050.

New characterization 2bc1d2284d75154e020640adb573c1cfadf005fb isolates a different admission boundary inside the existing n=4 budget. Residuals [0,1,2^-54,2] have exact minimum-anchor coordinates, while non-anchor 1 - 2^-54 rounds in binary64 and makes the current pairwise-f64 proof refuse. On common unit 2^-54, exact integer coordinates [0,2^54,1,2^55] produce the same pair numerator through direct integer pair distances and n*Σc_i²-(Σc_i)²:

3569704090242693886528325169446915.

The public metric already returns the correctly represented 0x3fdea33e2c83c140 through the generic fallback. This is therefore a proof-admission/resource finding, not a public-value defect: exact non-anchor pair subtraction is sufficient for the current O(n²) reference but is not a scientific prerequisite when canonical anchor-relative coordinates are exact.

Doctoring is current through CHANGELOG fragment 8050b4918803e69d8746167948a216d90a0cbd5a and research trace/current head ea592b32a0733ede28390945e70c332a2cb4f3a5. Production routing has not yet moved to the anchor-linear/Wide256 numerator path, pairwise fail-closed comparison remains in place, and the sample budget has not widened.

Foundation prerequisite #492

#492 is Ready for Review at 794ba9e6dda9f043aa499920fdf609b81b075d7e. Its Rust Foundation, Documentation Quality, Security Scan, and Semgrep evidence is terminal GREEN, but CodeQL fails in the canonical owner control plane before the scan matrix starts. .github/main@fb2ae81dbeaacb0c630e51e9d772c6919fa220cf run 34014663536 reaches metadata binding with actor/sender opencode-agent[bot]; .github#1899 records the allowlist as github-actions[bot] and treats authorization of the actual app principal as an explicit owner security decision. TEPP will not change that allowlist, impersonate an allowed actor, or weaken the gate. #492 is not inherited here until it reaches protected main through the normal path.

Current delivery state

Exact #488 head is ea592b32a0733ede28390945e70c332a2cb4f3a5, Draft and mergeable on the protected base. Current-head required workflows were queued/pending at the latest sweep; no exact-head GREEN or qualifying independent approval is claimed. Pinned Rustfmt artifact 9982621569 remains predecessor-only evidence from 1f765a....

docs/product-technical-gap-baseline.md on #435 is still stale relative to this head, and canonical docs/TRACEABILITY.md still lacks a direct #491 resource-route row. They remain explicit doctoring gaps.

Merge policy

No force push, destructive rebase, self-approval, required-gate weakening, mutable sibling dependency, routine admin bypass, or fabricated benchmark/CI evidence. Before any production budget change: integrate canonical anchor-relative normalization and Wide256 numerator routing in the owning module under TDD; preserve pairwise comparison until same-domain equivalence and intended anchor-linear-only admissions are proven; retain exact midpoint/tie-to-even semantics; obtain Rust 1.98.0 release-mode raw CPU/allocator/RSS evidence; make TRACEABILITY/baseline current; and obtain exact-head Rust/rustdoc/100% line+branch/security/documentation GREEN plus qualifying independent review.

@coderabbitai

coderabbitai Bot commented Sep 3, 2026

Copy link
Copy Markdown

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

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

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

@seonghobae seonghobae changed the title fix(validation): preserve representable extreme bias fix(validation): preserve representable recovery metrics Sep 3, 2026
Non-force restack of PR #488 onto protected main after #490. Preserves the complete Validation Evidence lineage while inheriting the centralized hourly-development workflow unchanged.
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant