Measure constrained problems and mating opt-ins

[jkbennitt]

Summary

Measured 15-seed tables for the v0.2.0 surfaces. Every numeric cell is from python tools/oracle/run_new_surfaces.py --seeds 15 (2026-09-30, pymoo 0.6.2, .NET SDK 10.0.401). Package defaults are unchanged. docs/WILCOXON-RESULTS.md is not rewritten.

Tables and the protocol write-up: docs/NEW-SURFACES-RESULTS.md. Knob list: docs/EQUIVALENCE.md.

Constrained (seeds 1–15)

Constructor defaults (RankNicheDistance, IndependentWithReplacement), single-layer Das–Dennis, SBX η=30 / PM η=20. IGD is pymoo's mean nearest distance on the feasible non-dominated final population, both solvers, one shared reference front. res.F is a separate column and is not used in the tests.

Problem	n_var	Pop	Gens	Partitions
OSY	6	52	250	12
TNK	2	52	250	12
C1-DTLZ1	7 (k=5)	92	150	12

pymoo's default c1dtlz1 n_var is 12. This run passes n_var=7.

Each pymoo seed was run twice. TNK matched. OSY seed 11 and C1-DTLZ1 seed 14 did not; the table keeps the first run and lists the second. The two values are not averaged.

Mating opt-ins (seeds 1–15)

C# IGD on the feasible non-dominated front against ParetoFronts. Columns: default, TwoShuffledPasses (default tournament), Algorithm2 (default pool), both.

Problem	Pop	Gens
ZDT1	52	100
ZDT2	52	250
DTLZ2	92	150

ZDT1 default G6 strings match the published Wilcoxon C# column on 15/15 seeds. DTLZ2 algorithm2 G6 strings match the published PymooCompatible column on 15/15 seeds. That published file was not rewritten. ZDT2 and DTLZ2 default columns here are RankNicheDistance.

Commands

python tools/oracle/run_new_surfaces.py --seeds 15
dotnet run --project tools/OracleCompare -c Release -- \
  --problem zdt1 --partitions 12 --pop 52 --gens 100 --seed 1 --seed-count 15 \
  --tournament rank-niche --mating independent
dotnet run --project tools/OracleCompare -c Release -- \
  --problem osy --partitions 12 --pop 52 --gens 250 --seed 1 --seed-count 15
dotnet run --project tools/OracleCompare -c Release -- \
  --problem c1dtlz1 --partitions 12 --pop 92 --gens 150 --seed 1 --seed-count 15

Checklist

• dotnet test Unsga3.slnx -c Release passes locally (108 passed)
• CHANGELOG.md updated under Unreleased
• Public API changes documented (XML docs / README) — no API change
• If survival / normalization / metrics changed: noted oracle impact (or re-ran multi-seed) — no algorithm change; new surfaces measured

Test plan

• python tools/oracle/run_new_surfaces.py --seeds 15
• ZDT1 default column checked against the published G6 strings
• dotnet test Unsga3.slnx -c Release