Measure constrained problems and mating opt-ins

.grok/skills/unsga3-oracle/SKILL.md

@@ -68,8 +68,18 @@ Checks sphere residual, niche occupancy, coverage vs true PF. Known historical r
 3. If algorithm code changed: `dotnet test Unsga3.slnx -c Release` must stay green.
 4. Do **not** commit `tools/oracle/out/*.csv` (gitignored).
 
+### 4. Constrained problems and mating opt-ins
+
+```powershell
+python tools/oracle/run_new_surfaces.py --seeds 15
+```
+
+Writes `docs/NEW-SURFACES-RESULTS.md`. Does **not** write `docs/WILCOXON-RESULTS.md`. C# defaults stay `RankNicheDistance` + `IndependentWithReplacement`. C1-DTLZ1 passes pymoo `n_var=7` (k=5). Every cell is measured or `skip:`.
+
 ## Do not
 
 - Claim bit-identical fronts (different RNG streams).
 - Tighten CI IGD bars without multi-seed evidence.
 - Run full n=15 when the user only asked for a smoke check.
+- Invent IGD, HV, or Wilcoxon numbers. A missing front or a seed with no feasible point is `skip:`.
+- Rewrite `docs/WILCOXON-RESULTS.md` from `run_new_surfaces.py` or from a mating-opt-in sweep.

CHANGELOG.md

@@ -9,6 +9,7 @@ and this project adheres to [Semantic Versioning](https://semver.org/).
 
 ### Added
 
+- Measured 15-seed tables for OSY, TNK, and C1-DTLZ1 (feasible non-dominated IGD vs pymoo `UNSGA3`) and for mating opt-ins (`TwoShuffledPasses`, `Algorithm2`, and both) in `docs/NEW-SURFACES-RESULTS.md`. Protocol is in `docs/EQUIVALENCE.md`. Package defaults are unchanged. `docs/WILCOXON-RESULTS.md` is not rewritten.
 - Optional **TypeSafe / Jev System One** pass (`tools/typesafe-pareto`) for Score + Choice judgments over a small Pareto candidate sample. Additive semantic layer — does not replace NSGA-III / U-NSGA-III objectives. Live calls only when `TYPESAFE_API_KEY` is set; CI uses mocks. Metrics append to `metrics/typesafe-runs.jsonl`.
 
 ### Changed

README.md

@@ -8,7 +8,7 @@
 
 > **v0.2.0** — opt-in two-shuffle mating, Algorithm 2 tournament, odd-N pairing, feasible-only niching, equal-CV Deb fix, two-layer Das–Dennis, and OSY/TNK/C1-DTLZ1. Defaults stay unconstrained `RankNicheDistance`, independent mating, and single-layer Das–Dennis. The published 15-seed table is not rewritten.  
 > **15-seed IGD vs pymoo `UNSGA3`:** ZDT1 **median 0.053 vs 0.070** (MWU *p*≈0.05) compares the full C# non-dominated front with pymoo `res.F`. DTLZ2 **median 0.0045 vs 0.0028** (~1.6×) compares C# n_var=12 with pymoo's default n_var=10. Neither pair is a same-set, same-problem ranking. Notes: [`docs/ORACLE-RESULTS.md`](docs/ORACLE-RESULTS.md).  
-> Details: [`docs/WILCOXON-RESULTS.md`](docs/WILCOXON-RESULTS.md) · single-seed notes: [`docs/ORACLE-RESULTS.md`](docs/ORACLE-RESULTS.md)
+> Details: [`docs/WILCOXON-RESULTS.md`](docs/WILCOXON-RESULTS.md) · single-seed notes: [`docs/ORACLE-RESULTS.md`](docs/ORACLE-RESULTS.md) · constrained + mating opt-ins: [`docs/NEW-SURFACES-RESULTS.md`](docs/NEW-SURFACES-RESULTS.md)
 
 ```text
 https://github.com/AppSprout-dev/Unsga3
@@ -123,6 +123,7 @@ API docs used: [HTTP](https://docs.typesafe.ai/api.md) · [Python SDK](https://d
 |-----|----------|
 | [docs/ORACLE-RESULTS.md](docs/ORACLE-RESULTS.md) | Single-seed C# vs pymoo |
 | [docs/WILCOXON-RESULTS.md](docs/WILCOXON-RESULTS.md) | Multi-seed Mann–Whitney / Wilcoxon |
+| [docs/NEW-SURFACES-RESULTS.md](docs/NEW-SURFACES-RESULTS.md) | 15-seed constrained IGD and mating opt-ins |
 | [docs/EQUIVALENCE.md](docs/EQUIVALENCE.md) | Protocol & intentional deltas (ZDT2 quality A/B = gens=250, `PymooCompatible`; matches unsga3-bend) |
 | [docs/RESEARCH-STANDARDS.md](docs/RESEARCH-STANDARDS.md) | Literature + indicator standards |
 | [docs/NOTICE.md](docs/NOTICE.md) | Attribution (papers + validation tools) |

docs/EQUIVALENCE.md

@@ -45,13 +45,13 @@ ZDT2 **gens=100** is an early-stress snapshot (collapse on Bend, C#, and pymoo),
 | Bi | ZDT1–4, ZDT6 | 2 | yes |
 | Many | DTLZ1–4, DTLZ7 | 3+ | yes |
 | Hard | WFG1, WFG2, WFG9 | 3–5 | planned |
-| Constrained | OSY, TNK, C1-DTLZ1 | 2–3 | yes (self-tests; no IGD table) |
+| Constrained | OSY, TNK, C1-DTLZ1 | 2–3 | yes (self-tests; 15-seed IGD in [NEW-SURFACES-RESULTS.md](NEW-SURFACES-RESULTS.md)) |
 
 ## Unit checks
 
 - Reference association & niche counts  
 - Non-dominated ranks / constraint domination (equal CV is mutual non-domination)  
-- Feasible niching, CV fill, and OSY / TNK / C1-DTLZ1 (`ConstrainedSurvivalTests`, `ConstrainedProblemTests`). No IGD table.  
+- Feasible niching, CV fill, and OSY / TNK / C1-DTLZ1 (`ConstrainedSurvivalTests`, `ConstrainedProblemTests`). Multi-seed IGD is in [NEW-SURFACES-RESULTS.md](NEW-SURFACES-RESULTS.md).  
 - **Normalization intercepts / ASF axis extremes** (`NormalizationTests`)  
 - Tournament pressure  
 - `populationSize: null` ⇒ `|refs|`  
@@ -95,6 +95,39 @@ pymoo 0.6.2 `TournamentSelection` (`pressure=2`) **is** the two-shuffled-pass po
 
 Odd N (Das–Dennis M=2, p=12 → 13; M=3, p=12 → 91): `CreateOffspring` pairs the leftover parent with parent 0, so `parents[N-1]` enters SBX, then the cursor restarts at 0. Even N stays consecutive pairs `(0,1), (2,3), …` and wraps to `(0,1)` with no extra random draw. Published ZDT1 (52) and DTLZ2 (92) are even, so their pair sequence is unchanged.
 
+## New surfaces (constrained + mating opt-ins)
+
+Measured tables live in [NEW-SURFACES-RESULTS.md](NEW-SURFACES-RESULTS.md). They are written by `tools/oracle/run_new_surfaces.py`. [WILCOXON-RESULTS.md](WILCOXON-RESULTS.md) is not a target of that script. Package defaults stay `RankNicheDistance` and `IndependentWithReplacement`.
+
+Seeds are 1..15, the same indexes as the published ZDT1 and DTLZ2 tables. C# and pymoo do not share an RNG. Pairing is by seed index. Operators: SBX η=30, p_c=1, PM η=20, p_m=1/n, duplicate elimination on, single-layer Das–Dennis.
+
+### Constrained
+
+| Problem | M | n_var | Partitions | Pop | Gens | C# settings |
+|---------|---|------:|------------|----:|-----:|-------------|
+| OSY | 2 | 6 | 12 | 52 | 250 | constructor defaults |
+| TNK | 2 | 2 | 12 | 52 | 250 | constructor defaults |
+| C1-DTLZ1 | 3 | **7** (k=5) | 12 | 92 | 150 | constructor defaults |
+
+Pop and partitions follow the ZDT (M=2, pop=52) and DTLZ2 (M=3, pop=92) oracles. OSY and TNK use the ZDT2 quality generation count (250). C1-DTLZ1 uses the DTLZ2 generation count (150). `C1Dtlz1Problem` defaults to k=5. pymoo `get_problem("c1dtlz1")` defaults to n_var=12; the harness passes **n_var=7**.
+
+IGD is pymoo's mean nearest distance on the **feasible non-dominated subset of the final population**, both solvers, one shared reference front. OSY and TNK use pymoo's shipped front file. C1-DTLZ1 uses the DTLZ1 simplex at the same Das–Dennis partitions (the C1 inequality does not remove that front). `res.F` is a different set and is not the comparison. A seed with no feasible point is `skip:`, not a filled-in IGD.
+
+pymoo `UNSGA3` still uses its own tournament and two-shuffle pool. This is default Unsga3 against default pymoo, on a shared front. It is not a claim that the mating operators match. A second pymoo run of the same seed is not always bit-identical with SciPy OpenBLAS. The results file keeps the first run in the table and lists the second run when it differs. The two values are not averaged.
+
+### Mating opt-ins
+
+Same populations and generation counts as the quality protocols above. ZDT1 gens=100, ZDT2 gens=250, DTLZ2 gens=150. IGD is the C# feasible non-dominated front against `ParetoFronts` (the published ZDT1 C# definition).
+
+| Column | Tournament | Pool |
+|--------|------------|------|
+| default | `RankNicheDistance` | `IndependentWithReplacement` |
+| two_shuffled | `RankNicheDistance` | `TwoShuffledPasses` |
+| algorithm2 | `Algorithm2` | `IndependentWithReplacement` |
+| both | `Algorithm2` | `TwoShuffledPasses` |
+
+The ZDT2 and DTLZ2 default columns in that table are `RankNicheDistance`. The published DTLZ2 Wilcoxon column is `PymooCompatible`. ZDT2 has no published Wilcoxon column. Neither published table is rewritten.
+
 ## DTLZ2 gap history
 
 | Stage | C# pymoo-mode IGD | vs pymoo 0.0035 |