Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 97.34%. Comparing base (deca9c7) to head (aaf6fe7).
⚠️ Report is 2 commits behind head on main.

@@            Coverage Diff             @@
##             main     #694      +/-   ##
==========================================
+ Coverage   97.32%   97.34%   +0.02%     
==========================================
  Files         327      329       +2     
  Lines       42170    42297     +127     
==========================================
+ Hits        41042    41175     +133     
+ Misses       1128     1122       -6     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

Implementation Summary

Changes

• src/rules/minimummultiwaycut_qubo.rs — Reduction rule using Heidari et al. (2022) penalty Hamiltonian with kn binary variables
• src/unit_tests/rules/minimummultiwaycut_qubo.rs — 4 unit tests (closed-loop, small, sizes, terminal pinning)
• src/rules/mod.rs — Module registration and example-db integration
• docs/paper/reductions.typ — Paper entry with rule statement, proof, and worked example
• docs/paper/references.bib — Bibliography entry for Heidari et al. (2022)
• Regenerated exports: reduction_graph.json, problem_schemas.json, examples.json

Deviations from Plan

• None

Open Questions

• None

Agentic Review Report

Structural Check

Result: 11/11 structural checks passed.

Build: all tests pass, clippy clean.

Semantic Review:

• extract_solution correctness — OK. Correctly inverts the one-hot encoding: finds the active terminal for each vertex, then marks edges as cut/uncut. Output vector length matches source dims().
• Overhead accuracy — OK. num_vars = "num_terminals * num_vertices" matches the code (nq = n * k). Verified on concrete instance: n=5, k=3 → 15-variable QUBO.
• Mathematical correctness — OK. Traced on small instances: H_A (one-hot + terminal pinning) correctly penalizes invalid partitions with alpha = sum(|w|) + 1; H_B (cut cost) correctly counts each cut edge exactly once. Upper-triangular matrix convention consistent with QUBO evaluation. Both optimal QUBO solutions decode to minimum cuts matching brute-force on the source.
• Penalty coefficient — OK. Code uses alpha = sum(|w|) + 1.0 = 18.0 for the canonical instance (the issue said α=20, which appears to be an arithmetic error in the issue: 2+3+1+2+4+5 = 17, so sum+1 = 18). The code's formula is correct.
• Paper quality — OK. Proof sketch includes Construction, Correctness (forward + backward), and Solution extraction. Worked example is data-driven from fixture JSON. Reference (Heidari et al. 2022) properly added to references.bib.

Issue Compliance (#186): 6/6 checks passed — source/target, algorithm, solution extraction, correctness, overhead, and example all match the linked issue.

Quality Check

Design Principles:

• DRY: OK — The add_upper closure pattern is duplicated across QUBO reductions but is a 3-line inline closure; extracting it would add coupling without benefit.
• KISS: OK — Straightforward reduction logic following established QUBO encoding patterns. reduce_to is 60 lines.
• HC/LC: OK — Clean separation between solution mapping (ReductionResult) and matrix construction (reduce_to).

HCI: N/A (no CLI changes).

Test Quality Issues:

Agentic Feature Tests

Cross-checks: All direct-solve vs solve-via-reduction results match across all tested instances (canonical 5-vertex, minimal 2-terminal, disconnected, K4).

Issues found:

Both are caught by cargo fmt --check and trivially fixable with cargo fmt.

Generated by review-pipeline