feat(scaling): Glicko rating system with deviation inflation

[seonghobae]

Iteration 59 of the autonomous paper-implementation loop. Stacked on #297 (Elo).

Summary

Implements the Glicko rating system (Glickman) as a Rust core mlsirm_core::scaling::glicko_rating with a thin PyO3 binding and Python wrapper fast_mlsirm.glicko_rating.

Citation governance

• READ: Glickman, M. E. (n.d.). The Glicko system [Technical note]. http://www.glicko.net/glicko/glicko.pdf — Step 1b/Step 2 formulas; the worked example (r=1500, RD=200 vs three opponents → r'=1464, RD'=151.4) is reproduced to full float64 precision by gk_paper_anchor_ga.
• READ: Stephenson, A., & Sonas, J. (2020). PlayerRatings 1.1-0 [R package]. R/ratings.R glicko() + src/ratings.c glicko_c — batch-per-period reference (cval inflation, rdmax clamp, gamma, W/D/L + lag bookkeeping).
• NOT READ (as-cited): Glickman, M. E. (1999). Parameter estimation in large dynamic paired comparison experiments. Applied Statistics, 48(3), 377–394.

Model

Per period (ascending label, batch semantics): participant-only variance inflation v = min(v + (lag+1)c², rdmax²); g = 1/sqrt(1 + 3(q/π)²v) post-inflation; per-game accumulation with opponent g; variance update first, then rating with the new variance. Documented non-identity: no rating-sum conservation (asymmetric opponent-g weighting; pinned ≠ 6600 by test).

Verification

• Executed float64 oracle (fixtures GA–GF); spec-verify verdict APPROVED-WITH-CHANGES, all 6 mandatory changes adopted (fractional-score fixture, expanded mutation plan, inherited elo fidelity contract, n≥2, Result<_,String>, no-status divergence documented).
• 7 executed mutation kills: MU1 opponent-g swap (heterogeneous-RD anchor; uniform-RD fixtures are blind), MU2 inflation off-by-one, MU3 clamp drop, MU4 stale-variance update, MU5 missing q², MU7 gamma sign (exact pins), MU9 all-player inflation (full dev-vector pin; idle player must not inflate).
• Every assert reads crate-returned values; gk_mc_500 (#[ignore]) executed once, pass.
• Full suites: cargo 785 pass; pytest 300 pass.
• Python wrapper inherits the Elo period-label fidelity contract (integer-dtype lossless u64 path; dtype-derived float bound — float32 2^24, float64 2^53), regression-tested.

Adversarial implementation review (2 rounds)

Round 1 ? FINDINGS (files/glicko_impl_review.md): confirmed all core semantics clean vs PlayerRatings R/C reference (participant-only pre-period inflation with rdmax clamp, post-inflation g for all players, opponent-g accumulation, variance-before-rating update order, gamma signs, exact-only tallies, lag order; GB deviation-vector pin distinguishes MU9, GD pins kill gamma sign swap). One Medium defect: the float period-fidelity bound used 2.0 ** np.finfo(dtype).nmant ? nmant excludes the implicit leading bit, so exactly representable labels were rejected one power of two early (float32 at 2^23, float64 at 2^52) in both the elo and glicko wrappers.

Fix (bb2a441): bound is now 2.0 ** (np.finfo(dtype).nmant + 1) (float64 2^53, float32 2^24, float16 2^11) with the >= comparison kept (2^53 itself stays rejected: 2^53 + 1 aliases onto it). Boundary tests added for both elo and glicko pin acceptance of 2^24 - 1 (float32) and 2^53 - 1 (float64) via crate game tallies; these asserts FAIL under the old 2**nmant code, killing that mutant.

Round 2 ? CLEAN (files/glicko_impl_review_round2.md): fix verified correct for all three float dtypes, >= semantics confirmed, boundary tests confirmed mutation-killing, TestGlicko + TestElo 11/11 pass, no new defects in the fix diff.

[coderabbitai]

Important

Review skipped

Auto reviews are disabled on base/target branches other than the default branch.

Please check the settings in the CodeRabbit UI or the .coderabbit.yaml file in this repository. To trigger a single review, invoke the @coderabbitai review command.

⚙️ Run configuration

Configuration used: Organization UI

Review profile: CHILL

Plan: Pro Plus

Run ID: efaf8670-9060-4719-b69c-6ef3e70a467d

You can disable this status message by setting the reviews.review_status to false in the CodeRabbit configuration file.

Use the checkbox below for a quick retry:

• 🔍 Trigger review

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Commit unit tests in branch seonghobae-glicko

---

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

[seonghobae]

Note: this PR was squash-merged into a non-main stacked base, so its commits did not land on main via this PR. The feature set reached main via #374 (stack tip integration) after #290.