feat(scaling): Stephenson rating system (PlayerRatings steph())

[seonghobae]

Iteration 61: Stephenson rating system (stephenson_rating)

Implements the Stephenson rating system as in CRAN PlayerRatings 1.1-0 steph().

Citation governance

• READ (normative): PlayerRatings 1.1-0 R/ratings.R lines 591-737 (steph() driver) and src/ratings.c lines 157-202 (stephenson_c kernel) — every formula line-cited in the Rust core header.
• NOT independently verifiable: the system's provenance (Alec Stephenson's winning entry, 2010 Kaggle chess-rating contest) is attributed by the package; no journal paper exists, so the R/C source is the sole normative authority.

What it adds over Glicko

• Per-game neighborhood variance term ngames * hval^2 inside the deviation update.
• Per-game bonus bval/100 added to each played game's score on both sides.
• Participants-only lambda drift toward opponents' ratings (lambda/100) * sum(r_opp - r_self) / ngames.
• (lag+1) * cval^2 per-period deviation-variance inflation clamped at rdmax^2 (note (lag+1), unlike Glicko-2's lag * sigma^2).
• Prior-run continuation via init_games / init_lag; per-game white advantage gamma.

Verification

• Spec-verify: APPROVED-WITH-CHANGES (4 mandatory changes, all adopted): u64 counts, explicit init_games/init_lag length validation, lag-continuation scope clarification, current-run-only W/D/L tallies.
• Oracle: faithful line-referenced Python port of the R driver + C kernel, EXECUTED; anchors S1 (heterogeneous init), S2 (two-period draw/lag), S3 (full knobs incl. asymmetric gamma), S4 (rdmax clamp + prior-run continuation), S5 (bval symmetry), S2-lambda0 contrast — all pinned at 1e-12 rel.
• Mutation kills (all EXECUTED, all KILLED): MU1 bval drop, MU2 lambda sign flip, MU3 ngames*hval^2 -> hval^2, MU4 (lag+1) -> lag, MU5 opponent-g -> own-g.
• Suites: cargo mlsirm-core --lib 802 pass; pytest test_paper_features.py 310 pass; st_mc_500 (500-rep MC, #[ignore]) pass.

Surface

• Rust: mlsirm_core::scaling::stephenson_rating (+ StephensonResult).
• PyO3: stephenson_rating (plain name, scaling-section precedent).
• Python: fast_mlsirm.stephenson_rating with PlayerRatings defaults init=(2200, 300), cval=10, hval=10, bval=0, lambda_=2, rdmax=350.

Stacked on #301 (Glicko-2).

---

Adversarial impl-review outcome

Round 1 (commit 3514293): FINDINGS ? 3 confirmed

• High: Python wrapper enforced the 2..=10000 n_players cap only in the Rust core, AFTER length-n allocations ? huge n_players attempted a huge allocation instead of raising ValueError.
• Medium: white/black player-id columns were cast through the float view, losing integer fidelity at/above the dtype's exact-integer bound before the u64 range checks.
• Medium: Rust core games/lag counter increments on caller-supplied u64 state could overflow (debug panic / release wrap) with init_games/init_lag near u64::MAX.

Fixes (commit 3e4fc99): early Python cap before any allocation; raw-integer-dtype direct u64 casts + fidelity rejection for float/object player ids; up-front u64::MAX - g overflow guard in the Rust core. Regression tests added on both sides (Rust st_error_contract overflow cases; Python fidelity + pre-allocation cap cases).

Round 2 (commit 3e4fc99): CLEAN ? all three fixes verified, including the negative integer player-id path; targeted Rust + Python error-contract suites re-run green.

[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: cba5e181-9573-4d3c-9ef8-79b82db091b3

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-stephenson

---

_{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.