[isPANN]

Registered numeric variants were missing direct conversions, forcing path search through unrelated models or leaving variants disconnected. Add 20 missing edges: 15 One → i64 conversions (including three decision wrappers) and five i64 → f64 conversions. All 29 matching numeric variant pairs now have direct executable edges.

Reuse the existing variant-reduction macro and declare exact identity parameter transforms. Unit-weight conversions preserve witnesses, objective values, and decision thresholds. The five new float conversions reject coefficients or accumulated objective bounds outside the exact-integer float range; Steiner forest also checks its scaled costs. No new variants or dependencies are introduced. Existing ILP rules qualify their shared trait helpers where the additional implementations would otherwise make type inference ambiguous.

Add a registry-wide test that discovers matching weight/coefficient variants automatically and requires direct witness-executable conversions. Behavior tests compare complete converted instances against independently deserialized target instances, solve both sides with the real brute-force solver, recover source optima and aggregate values, and exercise precision and overflow boundaries. Two unit-weight model examples supply inputs for the existing executable parameter-formula checks.

Remove Typst explanations of pure numeric casts, including the existing Spin Glass, Set Packing, and QUBO entries and numeric-cast paragraphs in MIS and ILP. Keep graph transformations, fixed-to-generic variants, and bounded-integer encodings. The paper completeness filter excludes only numeric promotions that leave every other variant dimension unchanged; a build-time assertion checks that a graph-changing MIS rule remains nontrivial. Repository guidance records this distinction.

Validation:

• The new tests failed before the missing conversions were implemented.
• Formatting and final workspace Clippy passed.
• Workspace unit/integration tests passed, including 6,247 library tests and the executable parameter-formula checks. Final focused numeric tests: 7 passed.
• Full library LLVM coverage passed: 95.67% line coverage. Final focused coverage gives 100% line coverage for the new exact-float helper and scaled forest cost checks.
• make paper passed, including the numeric/nontrivial classification assertions.
• make check still ends at the existing problemreductions-macros/src/lib.rs:211 doctest failure: its standalone reduction attribute has no attached item. That file is unchanged from main.

[codecov]

Codecov Report

❌ Patch coverage is 99.24051% with 15 lines in your changes missing coverage. Please review.
✅ Project coverage is 97.02%. Comparing base (35f40b4) to head (fbce255).
⚠️ Report is 26 commits behind head on main.

Files with missing lines	Patch %	Lines
src/unit_tests/rules/weight_casts.rs	98.00%	5 Missing ⚠️
src/unit_tests/parameter_formula_validation.rs	97.93%	3 Missing ⚠️
src/rules/travelingsalesman_ilp.rs	94.59%	2 Missing ⚠️
src/unit_tests/rules/travelingsalesman_ilp.rs	99.00%	2 Missing ⚠️
...ules/ksatisfiability_decisionminimumvertexcover.rs	87.50%	1 Missing ⚠️
src/rules/minimumtardinesssequencing_ilp.rs	87.50%	1 Missing ⚠️
src/unit_tests/example_db.rs	97.72%	1 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #1209      +/-   ##
==========================================
+ Coverage   96.90%   97.02%   +0.11%     
==========================================
  Files        1095     1101       +6     
  Lines      144972   145516     +544     
==========================================
+ Hits       140484   141183     +699     
+ Misses       4488     4333     -155     

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

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.
• 📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.

[GiggleLiu]

Agentic Review Report

Reviewed PR #1209 at 85175e3bdaff7454750a5ffc84ea700ba466a036 against merge base 875b6e86d5a87e5f66c2f6743be9d22abe917a09.

Structural and semantic

No blocking findings against this PR's base. Inspected all 20 added variant conversions and registrations. Unit-weight conversions preserve all instance fields and decision bounds. The shared float helper checks each value before taking its absolute value, so i64::MIN cannot overflow that operation. Its running absolute sum bounds every possible subset sum within the exactly represented integer range. The forest conversion additionally checks scaled prizes, edge costs, and vertex penalties with checked integer arithmetic. Separate comparative-containment sums remain exact on each side of the comparison.

The five floating conversions register aggregate mappings only where source and target evaluation types agree. Numeric metadata is copied exactly. The new registry audit requires all corresponding concrete numeric variants to have direct executable edges. Paper filtering excludes only a numeric dimension change with identical remaining dimensions.

Quality

git diff --check passed. Reviewed tests compare serialized target instances, solve source and target by brute force, check all optimal target witnesses, preserve decision bounds, and reject inexact sums and scaled products. Nine checks on the exact PR head are green, including tests, Clippy, formatting, coverage, and platform builds. No local Rust compilation or paper rendering was rerun by this reviewer.

Feature test

No CLI installation was performed. These registry-driven conversions are exercised by direct path construction and runtime reduction in the committed tests, covered by existing green CI.

Findings

No blocker in isolation. There is a definite integration dependency if #1194 lands first: src/unit_tests/rules/weight_casts.rs:8 imports the removed DirectedGraph and MixedGraph, with constructors later in that file. Rename them to MultiDigraph and MultiMixedGraph when updating this PR onto the graph-type change, then run a combined build. A textually clean merge alone will not catch this incompatibility.

Safe to merge after this integration adjustment if needed and a current-head recheck.

GitHub PR audit

GATE: PASS

• harm.secrets: OK. No credentials introduced.
• harm.dangerous-files: OK. New files are purposeful Rust conversion modules and tests.
• harm.generated-artifacts: OK. No generated exports or build products.
• harm.cache: OK. No cache directories.
• harm.debug-leftovers: OK. No production probes.
• pr.description-quality: OK. The body describes conversion limits and verification.
• pr.traceable: OK. The missing graph edges and the added audit are explained.

Integration update

Head fbce2550d88a47b78de5e67259b896728a8139f4 merges the reviewed graph API change and accepted main comment cleanup. The only additional adaptation changes three references in src/unit_tests/rules/weight_casts.rs: DirectedGraph becomes MultiDigraph, and MixedGraph becomes MultiMixedGraph. These match the constructors required by MinimumFeedbackVertexSet and MixedChinesePostman. No fixture data or production cast behavior was changed. Formatting and diff-whitespace checks pass. Fresh CI must pass before merge.