[isPANN]

Stacked on #1194. The base is fix/simple-graph-invariant, so this diff shows only the comparator commit. When #1194 merges, GitHub retargets this PR to main.

Summary

pred path <S> <T> --compare compares the listed reduction paths by the polynomial certificates of their parameter relations. Smaller certificates are better.

Generic, batch-relative order.

• Each path is treated as a family x_i = c_i·t^e_i, and the order compares the top faces of the Newton polytopes of its certificates. f ≤gen g holds when f is eventually bounded by g for almost every valid growth family, in each cell of each stratum.
• The order is a preorder only within one comparison batch. Adding paths can change verdicts, so results from different batches cannot be chained.
• Unrelated parameters, such as m vs n², come out incomparable. The output keeps a Pareto front of non-dominated paths instead of forcing a single winner.
• Dominance, proven incomparability and UNRESOLVED are reported as separate verdicts.

Known limitations, documented.

• The order ignores exceptional exact-cancellation families, such as complete graphs with zero complement edges.
• A generic "yes" is not a worst-case guarantee.

What the commit adds:

• Library:
  • exact polynomial algebra in src/polynomial.rs;
  • partial-hull parameter composition in src/parameters.rs;
  • the comparator in src/poly_compare/: Newton coverage with checked LP, regular-valid witnesses, conservative final-front filtering, and rendering;
  • ReductionGraph::path_polynomial_profile.
• CLI: pred path ... --compare, plus --proofs for exact proofs and witness curves, and --json for the schema-5 structured certificates.
• Docs:
  • the user guide docs/src/polynomial-comparator.md, linked from SUMMARY.md and design.md;
  • the theory note docs/theory/polynomial-comparator.typ;
  • the examples/polynomial_compare_audit.rs audit example.

Test plan

• Unit tests for polynomial algebra, parameter composition, the comparator and rendering
• CLI tests for --compare, --proofs and --json
• After rebasing onto [Critical] Make graph semantics explicit: simple vs multi graph types #1194: make fmt-check, make clippy and make check pass (7,097 tests)
• Coverage before the rebase: comparator 98.37% lines, polynomial 99.38% lines
• Re-run coverage once [Critical] Make graph semantics explicit: simple vs multi graph types #1194 has merged

🤖 Generated with Claude Code