Add Thurstone (1927) Case V paired-comparison scaling

[seonghobae]

Thurstone (1927) Case V paired-comparison scaling

Iteration 51 of the autonomous paper-implementation loop.

What

mlsirm_core::scaling::thurstone_case_v (NEW module) + PyO3 binding + fast_mlsirm.thurstone_case_v wrapper returning a ThurstoneResult dataclass (scale, gof, model, residual).

Given an n x n choice-probability matrix (choice[i,j] = P(column j preferred over row i), psych convention):

• z = qnorm(choice); scale_j = colmean_j(z) - min_k colmean_k(z) (minimum exactly 0)
• fitted model model_ij = pnorm(scale_j - scale_i); residual = model - choice
• goodness of fit GF = 1 - sse/ssc over the full model matrix including the diagonal — this pins the psych code behavior; the .Rd prose claiming "lower off diagonal" is stale (lower-only GF on fixture A would be 0.99868280449920059, numerically distinct from the pinned 0.99986967677023893).

Citation governance

• READ: psych thurstone.R, matrix.addition.R, thurstone.Rd (Revelle). The %+% argument was hand-derived: -scale %+% t(scale) gives S_j - S_i (reviewer independently confirmed).
• NOT READ (as-cited): Thurstone (1927), Psychological Review 34(4), 273-286 — original unacquirable; cited as the model origin per the psych source.
• Deliberate divergence: entries must be strictly in (0, 1). psych's direct path admits qnorm(0)/qnorm(1) = +-Inf; we reject boundaries as a safety contract.

Verification

• Oracle: mpmath 50-digit, EXECUTED. Fixtures: A (3x3 asymmetric, nonzero residuals), B (2x2 exactly Case-V-consistent, GF = 1 round-trip), C (4x4 intransitive, min colmean not in first column).
• Spec-verify (adversarial, before implementation): APPROVED-WITH-CHANGES; all 4 mandatory changes applied (full-matrix GF documented + pinned; MC scales bounded to [0,2]; strict-(0,1) divergence marked; accuracy claims fixtures-only). Reviewer recomputed all pins at 80 digits; measured Rust-kernel error <= 2.2e-8 on fixtures (pins at 1e-6).
• Mutation kills (all EXECUTED on clean baseline, all KILLED): MU1 transpose index choice[j*n+i]; MU2 drop min-shift; MU3 sign flip S_i - S_j; MU4 GF denominator ssc without mean-centering (killed by fixture C — fixture A's delta 4.3e-7 is sub-tolerance, documented); MU5 residual sign choice - model.
• Suites: cargo 734 pass (incl. MC-500 #[ignore] verified separately); pytest 267 pass.

Limitation (stated, not hidden)

Fixture B is zero-residual and thus variance-mutation-blind; variance-sensitive mutants are anchored by fixtures A and C nonzero-residual pins.

---

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

VERDICT: CLEAN (first pass, no fix cycles needed).

Independent reviewer executed: Rust scaling tests (5 pass) + MC-500 #[ignore] (pass); 12 Python-binding probes (near-boundary 1e-300/1-1e-16, 50x50 consistent round-trip max scale err 8.54e-10, Fortran-order, non-contiguous transpose, float32, int-matrix rejection, NaN rejection, asymmetric non-complementary input, 1x1/0x0 rejection, raw-binding length/n-mismatch errors); re-verified all 15 numeric pins against the mpmath 50-digit oracle; confirmed citation-governance header honesty (psych READ / Thurstone 1927 NOT READ / .Rd-vs-code GF conflict / strict-(0,1) divergence); confirmed every assert reads crate/binding outputs; rustfmt --check clean.

[coderabbitai]

Warning

Review limit reached

@seonghobae, you've reached your PR review limit, so we couldn't start this review.

Next review available in: 13 minutes

Enable usage-based reviews in Billing to review now. Otherwise, wait until the next included review is available.
You're only billed for reviews past your plan's rate limits ($0.25/file).

How can I continue?

After more reviews become available, a review can be triggered using the @coderabbitai review command as a PR comment. Alternatively, push new commits to this PR.

To avoid repeated limits, reduce automatic review volume by pausing incremental auto-reviews earlier, using label-based review opt-in, excluding WIP or generated PR titles, or requesting reviews manually when the PR is ready. If your team needs uninterrupted high-volume reviews, an organization admin can enable usage-based reviews.

How do review limits work?

CodeRabbit enforces per-developer PR review limits for each organization. Most developers receive the normal plan review availability.

For paid Pro and Pro+ PR reviews, CodeRabbit uses adaptive limits for sustained high-volume activity. When a developer's recent PR review activity reaches the 95th percentile or higher among CodeRabbit users, additional reviews become available more gradually as earlier reviews age out of the rolling window.

Please refer docs for additional details.

Review details

⚙️ Run configuration

Configuration used: Organization UI

Review profile: CHILL

Plan: Pro Plus

Run ID: d71b5d4a-b19a-4a1d-bef3-7a0cbbc0982e

📥 Commits

Reviewing files that changed from the base of the PR and between 3c1390e and 6d1fd5b.

📒 Files selected for processing (8)

• CHANGELOG.md
• crates/fast-mlsirm-py/src/lib.rs
• crates/mlsirm-core/src/lib.rs
• crates/mlsirm-core/src/scaling.rs
• python/fast_mlsirm/__init__.py
• python/fast_mlsirm/scaling.py
• tests/test_paper_features.py
• tests/unit/scaling_tests.rs

✨ Finishing Touches

🧪 Generate unit tests (beta)

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

---

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

[Copilot]

Pull request overview

Adds a new Thurstone (1927) Case V paired-comparison scaling feature implemented in the Rust core, exposed through the PyO3 extension, and surfaced as a Python API returning a structured result. This fits the codebase’s pattern of implementing psych-algorithm-compatible psychometrics routines in Rust with Python wrappers and pinned oracle-based fixtures.

Changes:

• Introduces mlsirm_core::scaling::thurstone_case_v with strict (0, 1) input validation, returning scale/model/residual and a full-matrix GF.
• Adds PyO3 binding and a Python wrapper (fast_mlsirm.thurstone_case_v) returning a ThurstoneResult dataclass with NumPy-shaped outputs.
• Adds Rust + Python fixture pins and error-contract tests, and documents the feature in the changelog.

Reviewed changes

Copilot reviewed 8 out of 8 changed files in this pull request and generated 1 comment.

Show a summary per file

File	Description
tests/unit/scaling_tests.rs	New Rust fixture pins and error-contract tests for Thurstone Case V scaling.
tests/test_paper_features.py	Adds Python-level fixture pins and error tests for fast_mlsirm.thurstone_case_v.
python/fast_mlsirm/scaling.py	New Python wrapper + ThurstoneResult dataclass around the Rust core implementation.
python/fast_mlsirm/init.py	Exposes thurstone_case_v and ThurstoneResult at the package top level.
crates/mlsirm-core/src/scaling.rs	New Rust implementation of Thurstone Case V scaling and result struct.
crates/mlsirm-core/src/lib.rs	Exports the new scaling module.
crates/fast-mlsirm-py/src/lib.rs	Adds the PyO3 thurstone_case_v function and registers it in the module.
CHANGELOG.md	Documents the new scaling feature and its contract/fixtures.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.