Add Kendall (1940) circular-triads consistency test and coefficient of agreement u

[seonghobae]

Iteration 57 of the autonomous paper-implementation loop. Stacked on #295 (seonghobae-top1).

Kendall & Babington Smith (1940): circular triads + coefficient of agreement u

Source governance: CRAN eba 1.10-0 R/circular.R, R/kendall.u.R, man/circular.Rd, man/kendall.u.Rd — READ (implementation source). Kendall & Babington Smith (1940, Biometrika 31, 324-345) and Alway (1962, exact null tables) — NOT READ (acquisition failed); cited as origins as described by the eba source.

What

• scaling::circular_triads(mat, n, alternative, correct) — number of circular triads T = C(n,3) - sum_j C(d_j,2) (integer form, proves integrality; algebraically equal to eba's n(n-1)(2n-1)/12 - sum(d^2)/2), T_max (parity-split), T_exp = C(n,3)/4, zeta = 1 - T/T_max; EXACT null p-values for n <= 10 from embedded null distributions (p-values are dyadic rationals k/2^C(n,2), pinned with assert_eq!), continuity-corrected chi-square for n >= 11 with tail inversion (less = upper chi-square tail per eba).
• scaling::kendall_u(mat, n, correct) — Sigma, u = 2 Sigma / (C(m,2) C(n,2)) - 1, min_u (-1/m odd, -1/(m-1) even), RAW chi-square (can be negative under the continuity correction — pinned at exactly -1 in a fixture; only the p-value clamps), df = C(n,2) m(m-1)/(m-2)^2.
• PyO3 bindings + Python wrappers with CircularTriadsResult/KendallUResult dataclasses; validation before casts (complex, object dtype, square 2-D, n cap).

Documented divergences from eba

• circular(): n = 2 rejected (T_max = 0 makes zeta undefined); incomplete/non-binary tournaments and nonzero diagonals rejected instead of returning garbage.
• kendall.u(): eba derives m from the FIRST pair only and never checks its equal-observations assumption — here every pair must satisfy M_ij + M_ji = m >= 3; entries capped at 1e6 so all binomials stay exact in u64.

Evidence discipline

• Exact-Fraction oracle (kendall_oracle.py) EXECUTED all-pass before spec review: 1940 dog example (T=5, zeta=3/8, p_less=1043/2048), transitive n=5 two-sided (9/64), n=12 chi-square path (chi2=121/8, df=165/8 vs scipy), agreement fixtures (Sigma=11, u=2/9, chi2=11, df=9; perfect u=1; negative raw chi2=-1).
• Adversarial spec-verify BEFORE implementation: APPROVED-WITH-CHANGES, all 8 mandatory changes adopted (incl. proof that the row/colsum mutant is algebraically invisible for complete tournaments — replaced with a drop-the-square mutant; two-sided break-loop = R cumsum equivalence proof).
• 11 Rust tests + 8 Python tests; every assert reads crate-returned values. Table integrity: sum(freq) = 2^C(n,2) for all n = 2..10.
• 5 mutants EXECUTED and killed: MU1 drop pairing (9/1 fail), MU2 T_max parity swap (8/2), MU3 drop opposite-tail extra (8/2), MU4 chi-square corr sign flip (9/1), MU5 drop Sigma continuity correction (6/4); baseline restored green.
• Full suites: cargo 769 pass / pytest 289 pass.

---

Adversarial impl-review outcome (post-merge-of-review)

Independent adversarial review (background agent, full report in session files kendall_impl_review.md): CLEAN — round 1, no confirmed findings.

Verified by execution: Rust kd_ tests 10 pass + MC-500 ignored pass; Python 8 pass; line-by-line fidelity vs eba circular.R/kendall.u.R (T integer form, T_max parity, exact-table indexing, chi-square + tail directions, kendall.u Sigma/u/min_u/chi2/df); adversarial probes (NaN/inf/complex/object/non-contiguous/Fortran/case-sensitive alternative/huge n) all ValueError or correct; brute-force exact p-values over all tournaments for small n match the crate for all alternatives; scipy chi-square cross-check n=15 all alternatives, correct=True/False; 2 mutation spot-checks re-executed and killed.

[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: a69db000-ae2f-4d56-a79c-597469037132

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-kendall

---

_{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.