refactor(api)!: replace raw TDS access with owner topology queries

Cargo.toml

@@ -55,6 +55,7 @@ uuid = { version = "1.23.4", features = [ "v4", "serde", "fast-rng" ] }
 [dev-dependencies]
 approx = "0.5.1"
 criterion = { version = "0.8.2", features = [ "html_reports" ] }
+markov-chain-monte-carlo = "0.4.0"
 pastey = "0.2.3"
 proptest = "1.11.0"
 serde_json = "1.0.150"

README.md

@@ -151,7 +151,7 @@ and [`docs/numerical_robustness_guide.md`](docs/numerical_robustness_guide.md).
 | Level | Validates | Primary API |
 |---|---|---|
 | 1 | Vertex, simplex, and facet element invariants | `vertex.is_valid()` / `simplex.is_valid()` |
-| 2 | TDS keys, incidences, and neighbor links | `dt.tds().is_valid()` / `dt.tds().structure_report()` |
+| 2 | TDS keys, incidences, and neighbor links | `dt.validate_structure()` / `dt.structure_report()` |
 | 3 | Manifold topology, ridge links, and Euler consistency | `dt.as_triangulation().is_valid_topology()` / `dt.as_triangulation().topology_report()` |
 | 4 | Faithful embedding | `dt.as_triangulation().is_valid_embedding()` / `dt.as_triangulation().embedding_report()` |
 | 5 | Delaunay property via local predicates | `dt.is_valid_delaunay()` / `dt.delaunay_report()` |

benches/README.md

@@ -12,8 +12,8 @@ predicates fast across 2D-5D.
 | `ci_performance_suite.rs` | Public workflow regression contract | Calibrated 2D-5D canaries | ~5-10 min | CI, baselines, `just perf-no-regressions` |
 | `circumsphere_containment.rs` | Compare circumsphere predicate methods | 2D-5D fixed, 3D random, edge cases | ~5 min | Predicate tuning, summaries |
 | `cold_path_predicates.rs` | Track hot/cold predicate paths | 2D-5D hot queries, near-boundary cases | ~2-5 min | Predicate optimization work |
-| `delaunay_repair.rs` | Flip-based Delaunay repair on prepared Levels 1-4 fixtures | 2D-5D repair-convergent fixtures | ~2-5 min | Repair tuning |
-| `pachner_stress.rs` | Unified Pachner move API stress | Accepted 4D k=1/k=2/k=3 forward/inverse moves | <1 min | Monte-Carlo move workflow tuning |
+| `delaunay_repair.rs` | Flip-based Delaunay repair plus transaction-pressure cases | 2D-5D repair-convergent fixtures | ~2-5 min | Repair tuning |
+| `pachner_stress.rs` | Unified Pachner move API stress | Accepted 4D microcases plus 3D/4D Monte Carlo sequences | Manual | Monte-Carlo move workflow tuning |
 | `pl_manifold_repair.rs` | Over-shared facet and targeted topology repair | 2D/3D synthetic repair fixtures | <1 min | PL-manifold repair tuning |
 | `profiling_suite.rs` | Large-scale construction, memory, query, validation profiling | 2D/3D 10k, 4D 3k, 5D 1k | ~2-3 hr | Manual/monthly |
 | `delete_vertex.rs` | Vertex deletion and rollback cost | 2D-5D fixed cases | ~1-5 min | Vertex deletion |
@@ -42,7 +42,8 @@ predicates fast across 2D-5D.
 | Predicate comparison | `cargo bench --profile perf --bench circumsphere_containment -- --noplot` |
 | Predicate cold-path work | `cargo bench --profile perf --bench cold_path_predicates -- --noplot` |
 | Flip-based Delaunay repair | `cargo bench --profile perf --bench delaunay_repair -- --noplot` |
-| Unified Pachner move stress | `cargo bench --profile perf --bench pachner_stress -- --noplot` |
+| Flip-repair transaction pressure | `cargo bench --profile perf --bench delaunay_repair -- repair_transaction_pressure --noplot` |
+| Unified Pachner move stress | `just pachner-stress` |
 | PL-manifold repair path | `cargo bench --profile perf --features bench --bench pl_manifold_repair -- --noplot` |
 | Large-scale scaling suite | `cargo bench --profile perf --bench profiling_suite -- --noplot` |
 | Vertex deletion mutation baseline | `cargo bench --profile perf --bench delete_vertex -- --noplot` |
@@ -184,6 +185,56 @@ DELAUNAY_BENCH_EXPORT_METRICS=1 \
 Use `just bench-perf-summary` for release summaries; it runs the full perf
 profile summary workflow and captures the construction metrics automatically.
 
+## Pachner Stress
+
+```bash
+just pachner-stress
+just pachner-stress-3d
+just pachner-stress-4d
+```
+
+`pachner_stress.rs` contains two layers:
+
+- accepted-move microcases for the unified 4D Pachner API facade
+- manual Monte Carlo stress cases for 3D and 4D long-run topology stability
+
+The `just` recipes run the Monte Carlo cases at the default issue-scale target:
+10,000 vertices in 3D and 1,000 vertices in 4D, with 100,000 attempted random
+Pachner moves per Criterion sample and topology validation every 1,000
+attempts. Criterion requires at least 10 samples, so a default dimension-specific
+recipe measures at least ten 100K-move sequences. These recipes enable
+`DELAUNAY_PACHNER_STRESS_REPORT=1`, causing each measured sequence to emit a
+`pachner_stress_metric` line with accepted/rejected attempts, proposal
+diagnostics, validation time, final simplex count, and RSS memory counters.
+
+The stress cases validate topology plus the Level 4 embedding invariant that
+arbitrary Pachner moves are expected to preserve; Level 5 Delaunay validity is
+not a postcondition of random topology edits. The move stream runs through the
+`markov-chain-monte-carlo` delayed proposal API with a flat target, so
+successfully planned Pachner proposals commit with 100% acceptance while invalid
+local candidates are recorded as no-proposal self-loops. Validation failures
+include recent MCMC trace rows so long chains can be diagnosed by step, outcome,
+and topology size.
+
+Useful overrides:
+
+```bash
+just pachner-stress-4d 10000 250 1000 10
+
+DELAUNAY_PACHNER_STRESS_REPORT=1 \
+DELAUNAY_PACHNER_STRESS_ATTEMPTS=10000 \
+DELAUNAY_PACHNER_STRESS_VERTICES_4D=250 \
+cargo bench --profile perf --bench pachner_stress -- "monte_carlo/4d" --noplot
+```
+
+Supported override families are `DELAUNAY_PACHNER_STRESS_VERTICES`,
+`DELAUNAY_PACHNER_STRESS_ATTEMPTS`,
+`DELAUNAY_PACHNER_STRESS_VALIDATE_EVERY`,
+`DELAUNAY_PACHNER_STRESS_KEY_REFRESH_EVERY`, and
+`DELAUNAY_PACHNER_STRESS_SEED`. Append `_3D` or `_4D` for a
+dimension-specific value. Use `MONTE_CARLO_SAMPLE_SIZE` to change Criterion's
+sample count; the benchmark enforces Criterion's minimum of 10 samples.
+
 ## Circumsphere Containment
 
 ```bash