Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 97.55%. Comparing base (bb726f8) to head (1e4157d).
⚠️ Report is 1 commits behind head on main.

@@            Coverage Diff             @@
##             main     #707      +/-   ##
==========================================
+ Coverage   97.53%   97.55%   +0.01%     
==========================================
  Files         373      375       +2     
  Lines       47182    47272      +90     
==========================================
+ Hits        46020    46116      +96     
+ Misses       1162     1156       -6     

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

Implementation Summary

Changes

• Added GraphPartitioning -> MaxCut with a weighted-complete-graph construction in src/rules/graphpartitioning_maxcut.rs.
• Added focused rule tests in src/unit_tests/rules/graphpartitioning_maxcut.rs covering closed-loop optimality, target weights/size, identity extraction, and penalty overflow guarding.
• Registered the rule and its canonical example in src/rules/mod.rs.
• Added a reduction-rule("GraphPartitioning", "MaxCut", ...) entry with a worked 6-vertex example in docs/paper/reductions.typ.
• Regenerated and verified the gitignored reduction graph, problem schemas, and paper example export.

Deviations from Plan

• The plan referenced make regenerate-fixtures and tracked JSON fixture paths that do not exist in this checkout. This repo regenerates the needed example data via cargo run --features "example-db" --example export_examples, and the generated outputs under docs/src/reductions/ and docs/paper/data/ are gitignored.
• Added one small hardening follow-up after code review: the reduction now checks that num_edges + 1 fits in i32 before using it as the MaxCut penalty weight.

Open Questions

• None.

Agentic Review Report

Structural Check

Structural Review: rule graphpartitioning_maxcut

Structural Completeness

Build Status

• make test: PASS
• make clippy: PASS

Semantic Review

• extract_solution correctness: OK for the intended even-vertex domain.
• Overhead accuracy: OK.
• Example quality: OK.
• Paper quality: OK for even-vertex instances.
• Correctness domain: ISSUE — odd-vertex GraphPartitioning instances are infeasible in the source model, but reduce_to() still constructs an ordinary MaxCut instance and extract_solution() returns an infeasible source assignment.

Issue Compliance

Summary

• 11/11 structural checks passed
• 5/6 issue compliance checks passed
• ISSUE: the reduction does not enforce or encode the even-vertex precondition required by GraphPartitioning, so it is not correct for odd-vertex source instances.

Quality Check

Quality Review

Design Principles

• DRY: OK.
• KISS: ISSUE — the rule looks simple, but it silently depends on an unstated runtime precondition (num_vertices must be even).
• HC/LC: OK.

Test Quality

• Naive test detection: ISSUE
  • The new tests cover the happy-path example and overflow, but not the reachable odd-vertex boundary case.

Issues

Critical (Must Fix)

• Odd-vertex source instances are not handled correctly. GraphPartitioning treats odd-vertex graphs as infeasible, but this reduction still produces a valid MaxCut instance and identity-extracts a source assignment. Because odd-vertex GraphPartitioning instances are publicly creatable, this is a real correctness hole.

Important (Should Fix)

• Add a regression test covering odd-vertex rejection/handling.

Minor (Nice to Have)

• None.

Summary

• Hidden even-vertex precondition makes the reduction incomplete over the public GraphPartitioning API.
• The new tests miss the boundary case that exposes the bug.

Agentic Feature Tests

Agentic Feature Tests

Summary

• Discoverability: pass.
• Setup: pass.
• Functionality: pass on the canonical example path.

Findings

• None confirmed on the canonical example flow.
• Manual edge-case reproduction found a separate correctness problem outside the happy path:
  • pred solve on an odd-vertex source instance fails with Error: No solution found.
  • pred reduce of that same instance to MaxCut, followed by pred solve, returns a source evaluation: Invalid with an intermediate MaxCut optimum, showing the reduction does not preserve infeasibility.

Commands Exercised

• pred list
• pred show GraphPartitioning
• pred show MaxCut
• pred create --example GraphPartitioning
• pred solve <canonical-source> --solver brute-force
• pred reduce <canonical-source> --to MaxCut
• pred solve <canonical-reduction-bundle> --solver brute-force
• Edge case: pred create GraphPartitioning --graph 0-1,1-2 | pred reduce - --to MaxCut