more performance characterization work

DOCS.md

@@ -128,8 +128,65 @@ Use `RoutingProblem.initial_paths` to provide semantic per-net seeds. Seeds are
 | Field | Default | Description |
 | :-- | :-- | :-- |
 | `capture_expanded` | `False` | Record expanded nodes for diagnostics and visualization. |
+| `capture_conflict_trace` | `False` | Capture authoritative post-reverify conflict trace entries for debugging negotiated-congestion failures. |
+| `capture_frontier_trace` | `False` | Run an analysis-only reroute for reached-but-colliding nets and capture prune causes near their final conflict hotspots. |
 
-## 7. RouteMetrics
+## 7. Conflict Trace
+
+`RoutingRunResult.conflict_trace` is an immutable tuple of post-reverify conflict snapshots. It is empty unless `RoutingOptions.diagnostics.capture_conflict_trace=True`.
+
+Trace types:
+
+- `ConflictTraceEntry`
+  - `stage`: `"iteration"`, `"restored_best"`, or `"final"`
+  - `iteration`: Iteration index for `"iteration"` entries, otherwise `None`
+  - `completed_net_ids`: Nets with collision-free reached-target paths at that stage
+  - `conflict_edges`: Undirected dynamic-conflict net pairs seen after full reverify
+  - `nets`: Per-net trace payloads in routing-order order
+- `NetConflictTrace`
+  - `net_id`
+  - `outcome`
+  - `reached_target`
+  - `report`
+  - `conflicting_net_ids`: Dynamic conflicting nets for that stage
+  - `component_conflicts`: Dynamic component-pair overlaps for that stage
+- `ComponentConflictTrace`
+  - `other_net_id`
+  - `self_component_index`
+  - `other_component_index`
+
+The conflict trace only records dynamic net-vs-net component overlaps. Static-obstacle and self-collision details remain count-only in `RoutingReport`.
+
+Use `scripts/record_conflict_trace.py` to capture JSON and Markdown trace artifacts for the built-in trace scenarios. The default target is `example_07_large_scale_routing_no_warm_start`.
+
+## 8. Frontier Trace
+
+`RoutingRunResult.frontier_trace` is an immutable tuple of per-net post-run frontier analyses. It is empty unless `RoutingOptions.diagnostics.capture_frontier_trace=True`.
+
+Trace types:
+
+- `NetFrontierTrace`
+  - `net_id`
+  - `hotspot_bounds`: Buffered bounds around the net's final dynamic component-overlap hotspots
+  - `pruned_closed_set`
+  - `pruned_hard_collision`
+  - `pruned_self_collision`
+  - `pruned_cost`
+  - `samples`: First traced prune events near those hotspots
+- `FrontierPruneSample`
+  - `reason`: `"closed_set"`, `"hard_collision"`, `"self_collision"`, or `"cost"`
+  - `move_type`
+  - `hotspot_index`
+  - `parent_state`
+  - `end_state`
+
+The frontier trace is observational only. It reruns only the final reached-but-colliding nets in analysis mode, with scratch metrics, after the routed result is already fixed.
+
+Use `scripts/record_frontier_trace.py` to capture JSON and Markdown frontier-prune artifacts for the built-in trace scenarios. The default target is `example_07_large_scale_routing_no_warm_start`.
+
+Separately from the observational trace tooling, the router may run a bounded post-loop pair-local scratch reroute before refinement when the restored best snapshot ends with final two-net reached-target dynamic conflicts. That repair phase is part of normal routing behavior and is reported through the `pair_local_search_*` counters below.
+
+## 9. RouteMetrics
 
 `RoutingRunResult.metrics` is an immutable per-run snapshot.
 
@@ -148,11 +205,20 @@ Use `RoutingProblem.initial_paths` to provide semantic per-net seeds. Seeds are
 - `warm_start_paths_used`: Number of routing attempts satisfied directly from an initial or warm-start path.
 - `refine_path_calls`: Number of completed paths passed through the post-route refiner.
 - `timeout_events`: Number of timeout exits encountered during the run.
+- `iteration_reverify_calls`: Number of end-of-iteration full reverify passes against the final installed dynamic geometry.
+- `iteration_reverified_nets`: Number of reached-target nets reverified at iteration boundaries.
+- `iteration_conflicting_nets`: Total unique nets found in end-of-iteration dynamic conflicts.
+- `iteration_conflict_edges`: Total undirected dynamic-conflict edges observed at iteration boundaries.
+- `nets_carried_forward`: Number of nets retained unchanged between iterations.
 
 ### Cache Counters
 
 - `move_cache_abs_hits` / `move_cache_abs_misses`: Absolute move-geometry cache activity.
 - `move_cache_rel_hits` / `move_cache_rel_misses`: Relative move-geometry cache activity.
+- `guidance_match_moves`: Number of moves that matched the reroute guidance seed and received the guidance bonus.
+- `guidance_match_moves_straight`, `guidance_match_moves_bend90`, `guidance_match_moves_sbend`: Guidance-match counts split by move type.
+- `guidance_bonus_applied`: Total reroute-guidance bonus subtracted from move costs across the run.
+- `guidance_bonus_applied_straight`, `guidance_bonus_applied_bend90`, `guidance_bonus_applied_sbend`: Guidance bonus totals split by move type.
 - `static_safe_cache_hits`: Reuse count for the static-safe admission cache.
 - `hard_collision_cache_hits`: Reuse count for the hard-collision cache.
 - `congestion_cache_hits` / `congestion_cache_misses`: Per-search congestion-cache activity.
@@ -165,29 +231,50 @@ Use `RoutingProblem.initial_paths` to provide semantic per-net seeds. Seeds are
 - `static_tree_rebuilds`: Number of static dilated-obstacle STRtree rebuilds.
 - `static_raw_tree_rebuilds`: Number of raw static-obstacle STRtree rebuilds used for verification.
 - `static_net_tree_rebuilds`: Number of net-width-specific static STRtree rebuilds.
-- `visibility_builds`: Number of static visibility-graph rebuilds.
-- `visibility_corner_pairs_checked`: Number of corner-pair visibility probes considered while building that graph.
-- `visibility_corner_queries` / `visibility_corner_hits`: Precomputed-corner visibility query activity.
+- `visibility_corner_index_builds`: Number of lazy corner-index rebuilds.
+- `visibility_builds`: Number of exact corner-visibility graph rebuilds.
+- `visibility_corner_pairs_checked`: Number of corner-pair visibility probes considered while building the exact graph.
+- `visibility_corner_queries_exact` / `visibility_corner_hits_exact`: Exact-corner visibility query activity.
 - `visibility_point_queries`, `visibility_point_cache_hits`, `visibility_point_cache_misses`: Arbitrary-point visibility query and cache activity.
 - `ray_cast_calls`: Number of ray-cast queries issued against static obstacles.
 - `ray_cast_candidate_bounds`: Total broad-phase candidate bounds considered by ray casts.
 - `ray_cast_exact_geometry_checks`: Total exact non-rectangular geometry checks performed by ray casts.
 - `congestion_check_calls`: Number of congestion broad-phase checks requested by search.
+- `congestion_presence_cache_hits` / `congestion_presence_cache_misses`: Reuse of cached per-span booleans indicating whether a move polygon could overlap any other routed net at all.
+- `congestion_presence_skips`: Number of moves that bypassed full congestion evaluation because the presence precheck found no other routed nets in any covered dynamic-grid span.
+- `congestion_candidate_precheck_hits` / `congestion_candidate_precheck_misses`: Reuse of cached conservative per-span booleans indicating whether any candidate nets survive the net-envelope and grid-net broad phases.
+- `congestion_candidate_precheck_skips`: Number of moves that bypassed full congestion evaluation because the candidate-net precheck found no surviving candidate nets after those broad phases.
+- `congestion_candidate_nets`: Total candidate net ids returned by the dynamic net-envelope broad phase during routing.
+- `congestion_net_envelope_cache_hits` / `congestion_net_envelope_cache_misses`: Reuse of cached dynamic net-envelope candidate sets keyed by the queried grid-cell span.
+- `congestion_grid_net_cache_hits` / `congestion_grid_net_cache_misses`: Reuse of cached per-span candidate net ids gathered from dynamic grid occupancy.
+- `congestion_grid_span_cache_hits` / `congestion_grid_span_cache_misses`: Reuse of cached dynamic-path candidate unions keyed by the queried grid-cell span.
+- `congestion_lazy_resolutions`: Number of popped nodes whose pending congestion was resolved lazily.
+- `congestion_lazy_requeues`: Number of lazily resolved nodes requeued after a positive congestion penalty was applied.
+- `congestion_candidate_ids`: Total dynamic-path object ids returned by the congestion broad phase before exact confirmation.
 - `congestion_exact_pair_checks`: Number of exact geometry-pair checks performed while confirming congestion hits.
 
 ### Verification Counters
 
 - `verify_path_report_calls`: Number of full path-verification passes.
 - `verify_static_buffer_ops`: Number of static-verification `buffer()` operations.
+- `verify_dynamic_candidate_nets`: Total candidate net ids returned by the dynamic net-envelope broad phase during final verification.
 - `verify_dynamic_exact_pair_checks`: Number of exact geometry-pair checks performed during dynamic-path verification.
 
-## 8. Internal Modules
+### Local Search Counters
+
+- `pair_local_search_pairs_considered`: Number of final reached-target conflict pairs considered by the bounded post-loop pair-local-search phase.
+- `pair_local_search_attempts`: Number of pair-local-search reroute attempts executed across all considered pairs.
+- `pair_local_search_accepts`: Number of pair-local-search attempts accepted into the whole routed result set.
+- `pair_local_search_nodes_expanded`: Total A* node expansions spent inside pair-local-search attempts.
+
+## 10. Internal Modules
 
 Lower-level search and collision modules are semi-private implementation details. They remain accessible through deep imports for advanced use, but they are unstable and may change without notice. The stable supported entrypoint is `route(problem, options=...)`.
 
 The current implementation structure is summarized in **[docs/architecture.md](docs/architecture.md)**. The committed example-corpus counter baseline is tracked in **[docs/performance.md](docs/performance.md)**.
+Use `scripts/diff_performance_baseline.py` to compare a fresh local run against that baseline. Use `scripts/record_conflict_trace.py` for opt-in conflict-hotspot traces, `scripts/record_frontier_trace.py` for hotspot-adjacent prune traces, and `scripts/characterize_pair_local_search.py` to sweep example_07-style no-warm runs for pair-local repair behavior. The counter baseline is currently observational and is not enforced as a CI gate.
 
-## 9. Tuning Notes
+## 11. Tuning Notes
 
 ### Speed vs. optimality
 

docs/architecture.md

@@ -16,7 +16,7 @@
 - `inire/geometry/primitives.py`: Integer Manhattan ports and small transform helpers.
 - `inire/geometry/components.py`: `Straight`, `Bend90`, and `SBend` geometry generation.
 - `inire/geometry/collision.py`: Routing-world collision, congestion, ray-cast, and path-verification logic.
-- `inire/geometry/static_obstacle_index.py` and `inire/geometry/dynamic_path_index.py`: Spatial-index management for static obstacles and routed paths.
+- `inire/geometry/static_obstacle_index.py` and `inire/geometry/dynamic_path_index.py`: Spatial-index management for static obstacles and routed paths, including dynamic per-object indices, per-net grid occupancy, congestion grid membership, and per-net dynamic envelopes.
 - `inire/router/_search.py`, `_astar_moves.py`, `_astar_admission.py`, `_astar_types.py`: The state-lattice A* search loop and move admission pipeline.
 - `inire/router/_router.py`: The negotiated-congestion driver and refinement orchestration.
 - `inire/router/refiner.py`: Post-route path simplification for completed paths.
@@ -39,7 +39,10 @@ The search state is a snapped Manhattan `(x, y, r)` port. From each state the ro
 
 - Static obstacles and routed paths are treated as single-layer geometry; automatic crossings are not supported.
 - The danger-map implementation uses sampled obstacle-boundary points and a KD-tree, not a dense distance-transform grid.
+- The visibility subsystem keeps a lazy static corner index for default `tangent_corner` guidance and only builds the exact corner-to-corner graph on demand for `exact_corner` queries.
 - `use_tiered_strategy` can swap in a cheaper bend proxy on the first congestion iteration.
+- Negotiated congestion now re-verifies every reached-target path at the end of each iteration against the final installed dynamic geometry, and it stops early if the conflict graph stalls for consecutive iterations.
+- After best-snapshot restoration, the router runs a bounded pair-local scratch reroute on final two-net reached-target conflict pairs. That repair phase clones static obstacles from the live collision world, treats all outside-pair geometry as fixed blockers, tries both pair orders, and only keeps the result if whole-set reverify improves.
 - Final `RoutingResult` validity is determined by explicit post-route verification, not only by search-time pruning.
 
 ## Performance Visibility

docs/conflict_trace.md

@@ -0,0 +1,57 @@
+# Conflict Trace
+
+Generated at 2026-04-02T14:24:39-07:00 by `scripts/record_conflict_trace.py`.
+
+## example_07_large_scale_routing_no_warm_start
+
+Results: 10 valid / 10 reached / 10 total.
+
+| Stage | Iteration | Conflicting Nets | Conflict Edges | Completed Nets |
+| :-- | --: | --: | --: | --: |
+| iteration | 0 | 9 | 16 | 1 |
+| iteration | 1 | 8 | 12 | 2 |
+| iteration | 2 | 6 | 5 | 4 |
+| iteration | 3 | 4 | 2 | 6 |
+| iteration | 4 | 4 | 2 | 6 |
+| iteration | 5 | 4 | 2 | 6 |
+| restored_best |  | 4 | 2 | 6 |
+| final |  | 0 | 0 | 10 |
+
+Top nets by traced dynamic-collision stages:
+
+- `net_06`: 7
+- `net_07`: 7
+- `net_01`: 6
+- `net_00`: 5
+- `net_02`: 5
+- `net_03`: 4
+- `net_08`: 2
+- `net_09`: 2
+- `net_05`: 1
+
+Top net pairs by frequency:
+
+- `net_06` <-> `net_07`: 7
+- `net_00` <-> `net_01`: 5
+- `net_01` <-> `net_02`: 4
+- `net_00` <-> `net_02`: 3
+- `net_00` <-> `net_03`: 3
+- `net_02` <-> `net_03`: 3
+- `net_01` <-> `net_03`: 2
+- `net_06` <-> `net_08`: 2
+- `net_06` <-> `net_09`: 2
+- `net_07` <-> `net_08`: 2
+
+Top component pairs by frequency:
+
+- `net_06[2]` <-> `net_07[2]`: 6
+- `net_06[3]` <-> `net_07[2]`: 6
+- `net_06[1]` <-> `net_07[1]`: 6
+- `net_06[2]` <-> `net_07[1]`: 5
+- `net_00[2]` <-> `net_01[3]`: 4
+- `net_01[2]` <-> `net_02[2]`: 3
+- `net_01[2]` <-> `net_02[3]`: 3
+- `net_00[2]` <-> `net_01[2]`: 3
+- `net_07[3]` <-> `net_08[2]`: 2
+- `net_02[1]` <-> `net_03[1]`: 2
+

docs/frontier_trace.json

@@ -0,0 +1,120 @@
+{
+  "generated_at": "2026-04-02T14:24:39-07:00",
+  "generator": "scripts/record_frontier_trace.py",
+  "scenarios": [
+    {
+      "frontier_trace": [],
+      "metrics": {
+        "congestion_cache_hits": 31,
+        "congestion_cache_misses": 4625,
+        "congestion_candidate_ids": 9924,
+        "congestion_candidate_nets": 9979,
+        "congestion_candidate_precheck_hits": 2562,
+        "congestion_candidate_precheck_misses": 2165,
+        "congestion_candidate_precheck_skips": 71,
+        "congestion_check_calls": 4625,
+        "congestion_exact_pair_checks": 8122,
+        "congestion_grid_net_cache_hits": 2457,
+        "congestion_grid_net_cache_misses": 3942,
+        "congestion_grid_span_cache_hits": 2283,
+        "congestion_grid_span_cache_misses": 1948,
+        "congestion_lazy_requeues": 0,
+        "congestion_lazy_resolutions": 0,
+        "congestion_net_envelope_cache_hits": 2673,
+        "congestion_net_envelope_cache_misses": 4139,
+        "congestion_presence_cache_hits": 2858,
+        "congestion_presence_cache_misses": 2556,
+        "congestion_presence_skips": 687,
+        "danger_map_cache_hits": 16878,
+        "danger_map_cache_misses": 7425,
+        "danger_map_lookup_calls": 24303,
+        "danger_map_query_calls": 7425,
+        "danger_map_total_ns": 212814061,
+        "dynamic_grid_rebuilds": 0,
+        "dynamic_path_objects_added": 471,
+        "dynamic_path_objects_removed": 423,
+        "dynamic_tree_rebuilds": 0,
+        "guidance_bonus_applied": 11000.0,
+        "guidance_bonus_applied_bend90": 3500.0,
+        "guidance_bonus_applied_sbend": 625.0,
+        "guidance_bonus_applied_straight": 6875.0,
+        "guidance_match_moves": 176,
+        "guidance_match_moves_bend90": 56,
+        "guidance_match_moves_sbend": 10,
+        "guidance_match_moves_straight": 110,
+        "hard_collision_cache_hits": 0,
+        "iteration_conflict_edges": 39,
+        "iteration_conflicting_nets": 36,
+        "iteration_reverified_nets": 60,
+        "iteration_reverify_calls": 6,
+        "move_cache_abs_hits": 2559,
+        "move_cache_abs_misses": 6494,
+        "move_cache_rel_hits": 5872,
+        "move_cache_rel_misses": 622,
+        "moves_added": 8081,
+        "moves_generated": 9053,
+        "nets_carried_forward": 0,
+        "nets_reached_target": 60,
+        "nets_routed": 60,
+        "nodes_expanded": 1764,
+        "pair_local_search_accepts": 2,
+        "pair_local_search_attempts": 2,
+        "pair_local_search_nodes_expanded": 68,
+        "pair_local_search_pairs_considered": 2,
+        "path_cost_calls": 0,
+        "pruned_closed_set": 439,
+        "pruned_cost": 533,
+        "pruned_hard_collision": 0,
+        "ray_cast_calls": 5477,
+        "ray_cast_calls_expand_forward": 1704,
+        "ray_cast_calls_expand_snap": 46,
+        "ray_cast_calls_other": 0,
+        "ray_cast_calls_straight_static": 3721,
+        "ray_cast_calls_visibility_build": 0,
+        "ray_cast_calls_visibility_query": 0,
+        "ray_cast_calls_visibility_tangent": 6,
+        "ray_cast_candidate_bounds": 305,
+        "ray_cast_exact_geometry_checks": 0,
+        "refine_path_calls": 10,
+        "refinement_candidate_side_extents": 0,
+        "refinement_candidates_accepted": 0,
+        "refinement_candidates_built": 0,
+        "refinement_candidates_verified": 0,
+        "refinement_dynamic_bounds_checked": 0,
+        "refinement_static_bounds_checked": 0,
+        "refinement_windows_considered": 0,
+        "route_iterations": 6,
+        "score_component_calls": 8634,
+        "score_component_total_ns": 241025335,
+        "static_net_tree_rebuilds": 1,
+        "static_raw_tree_rebuilds": 1,
+        "static_safe_cache_hits": 2482,
+        "static_tree_rebuilds": 1,
+        "timeout_events": 0,
+        "verify_dynamic_candidate_nets": 2106,
+        "verify_dynamic_exact_pair_checks": 558,
+        "verify_path_report_calls": 190,
+        "verify_static_buffer_ops": 895,
+        "visibility_builds": 0,
+        "visibility_corner_hits_exact": 0,
+        "visibility_corner_index_builds": 1,
+        "visibility_corner_pairs_checked": 0,
+        "visibility_corner_queries_exact": 0,
+        "visibility_point_cache_hits": 0,
+        "visibility_point_cache_misses": 0,
+        "visibility_point_queries": 0,
+        "visibility_tangent_candidate_corner_checks": 6,
+        "visibility_tangent_candidate_ray_tests": 6,
+        "visibility_tangent_candidate_scans": 1704,
+        "warm_start_paths_built": 0,
+        "warm_start_paths_used": 0
+      },
+      "name": "example_07_large_scale_routing_no_warm_start",
+      "summary": {
+        "reached_targets": 10,
+        "total_results": 10,
+        "valid_results": 10
+      }
+    }
+  ]
+}

docs/frontier_trace.md

@@ -0,0 +1,23 @@
+# Frontier Trace
+
+Generated at 2026-04-02T14:24:39-07:00 by `scripts/record_frontier_trace.py`.
+
+## example_07_large_scale_routing_no_warm_start
+
+Results: 10 valid / 10 reached / 10 total.
+
+| Net | Hotspots | Closed-Set | Hard Collision | Self Collision | Cost | Samples |
+| :-- | --: | --: | --: | --: | --: | --: |
+
+Prune totals by reason:
+
+- None
+
+Top traced hotspots by sample count:
+
+- None
+
+Per-net sampled reason/move breakdown:
+
+- None
+

docs/pair_local_characterization.md

@@ -0,0 +1,30 @@
+# Pair-Local Search Characterization
+
+Generated at 2026-04-02T15:53:29-07:00 by `scripts/characterize_pair_local_search.py`.
+
+Grid: `num_nets=[6, 8, 10]`, `seed=[41, 42, 43]`, repeats=2.
+
+| Nets | Seed | Repeat | Duration (s) | Valid | Reached | Pair Pairs | Pair Accepts | Pair Nodes | Nodes | Checks |
+| :-- | :-- | :-- | --: | --: | --: | --: | --: | --: | --: | --: |
+| 6 | 41 | 0 | 0.5462 | 1 | 6 | 0 | 0 | 0 | 500 | 674 |
+| 6 | 41 | 1 | 0.5256 | 1 | 6 | 0 | 0 | 0 | 500 | 674 |
+| 6 | 42 | 0 | 0.5241 | 1 | 6 | 0 | 0 | 0 | 503 | 683 |
+| 6 | 42 | 1 | 0.5477 | 1 | 6 | 0 | 0 | 0 | 503 | 683 |
+| 6 | 43 | 0 | 0.5199 | 1 | 6 | 0 | 0 | 0 | 493 | 654 |
+| 6 | 43 | 1 | 0.5160 | 1 | 6 | 0 | 0 | 0 | 493 | 654 |
+| 8 | 41 | 0 | 1.8818 | 8 | 8 | 2 | 2 | 38 | 1558 | 4313 |
+| 8 | 41 | 1 | 1.8618 | 8 | 8 | 2 | 2 | 38 | 1558 | 4313 |
+| 8 | 42 | 0 | 1.4850 | 8 | 8 | 1 | 1 | 19 | 1440 | 3799 |
+| 8 | 42 | 1 | 1.4636 | 8 | 8 | 1 | 1 | 19 | 1440 | 3799 |
+| 8 | 43 | 0 | 1.0652 | 8 | 8 | 0 | 0 | 0 | 939 | 1844 |
+| 8 | 43 | 1 | 1.0502 | 8 | 8 | 0 | 0 | 0 | 939 | 1844 |
+| 10 | 41 | 0 | 2.8617 | 8 | 10 | 2 | 2 | 41 | 2223 | 6208 |
+| 10 | 41 | 1 | 2.8282 | 8 | 10 | 2 | 2 | 41 | 2223 | 6208 |
+| 10 | 42 | 0 | 2.0356 | 10 | 10 | 2 | 2 | 68 | 1764 | 4625 |
+| 10 | 42 | 1 | 2.0052 | 10 | 10 | 2 | 2 | 68 | 1764 | 4625 |
+| 10 | 43 | 0 | 50.1863 | 10 | 10 | 2 | 2 | 38 | 61259 | 165223 |
+| 10 | 43 | 1 | 50.4019 | 10 | 10 | 2 | 2 | 38 | 61259 | 165223 |
+
+## Recommendation
+
+No smaller stable pair-local smoke scenario satisfied the rule `valid_results == total_results`, `pair_local_search_accepts >= 1`, and `duration_s <= 1.0` across all repeats.

docs/performance.md

@@ -1,25 +1,41 @@
 # Performance Baseline
 
-Generated on 2026-03-31 by `scripts/record_performance_baseline.py`.
+Generated on 2026-04-02 by `scripts/record_performance_baseline.py`.
 
 The full machine-readable snapshot lives in `docs/performance_baseline.json`.
+Use `scripts/diff_performance_baseline.py` to compare a fresh run against that snapshot.
+
+The default baseline table below covers the standard example corpus only. The heavier `example_07_large_scale_routing_no_warm_start` canary remains performance-only and is tracked through targeted diffs plus the conflict/frontier trace artifacts.
+
+Use `scripts/characterize_pair_local_search.py` when you want a small parameter sweep over example_07-style no-warm runs instead of a single canary reading.
+The current tracked sweep output lives in `docs/pair_local_characterization.json` and `docs/pair_local_characterization.md`.
 
 | Scenario | Duration (s) | Total | Valid | Reached | Iter | Nets Routed | Nodes | Ray Casts | Moves Gen | Moves Added | Dyn Tree | Visibility Builds | Congestion Checks | Verify Calls |
 | :-- | --: | --: | --: | --: | --: | --: | --: | --: | --: | --: | --: | --: | --: | --: |
-| example_01_simple_route | 0.0042 | 1 | 1 | 1 | 1 | 1 | 2 | 22 | 11 | 7 | 2 | 2 | 0 | 3 |
-| example_02_congestion_resolution | 0.3335 | 3 | 3 | 3 | 1 | 3 | 366 | 1176 | 1413 | 668 | 8 | 4 | 0 | 35 |
-| example_03_locked_paths | 0.1810 | 2 | 2 | 2 | 2 | 2 | 191 | 681 | 904 | 307 | 5 | 4 | 0 | 14 |
-| example_04_sbends_and_radii | 2.0151 | 2 | 2 | 2 | 1 | 2 | 15 | 18218 | 123 | 65 | 4 | 3 | 0 | 6 |
-| example_05_orientation_stress | 0.2438 | 3 | 3 | 3 | 2 | 6 | 286 | 1243 | 1624 | 681 | 12 | 3 | 412 | 12 |
-| example_06_bend_collision_models | 4.1636 | 3 | 3 | 3 | 3 | 3 | 240 | 40530 | 1026 | 629 | 6 | 6 | 0 | 9 |
-| example_07_large_scale_routing | 1.3759 | 10 | 10 | 10 | 1 | 10 | 78 | 11151 | 372 | 227 | 20 | 11 | 0 | 30 |
-| example_08_custom_bend_geometry | 0.2437 | 2 | 2 | 2 | 2 | 2 | 18 | 2308 | 78 | 56 | 4 | 4 | 0 | 6 |
-| example_09_unroutable_best_effort | 0.0052 | 1 | 0 | 0 | 1 | 1 | 3 | 13 | 16 | 10 | 1 | 0 | 0 | 1 |
+| example_01_simple_route | 0.0040 | 1 | 1 | 1 | 1 | 1 | 2 | 10 | 11 | 7 | 0 | 0 | 0 | 4 |
+| example_02_congestion_resolution | 0.3378 | 3 | 3 | 3 | 1 | 3 | 366 | 1164 | 1413 | 668 | 0 | 0 | 0 | 38 |
+| example_03_locked_paths | 0.1929 | 2 | 2 | 2 | 2 | 2 | 191 | 657 | 904 | 307 | 0 | 0 | 0 | 16 |
+| example_04_sbends_and_radii | 0.0279 | 2 | 2 | 2 | 1 | 2 | 15 | 70 | 123 | 65 | 0 | 0 | 0 | 8 |
+| example_05_orientation_stress | 0.2367 | 3 | 3 | 3 | 2 | 6 | 299 | 1284 | 1691 | 696 | 0 | 0 | 149 | 18 |
+| example_06_bend_collision_models | 0.1998 | 3 | 3 | 3 | 3 | 3 | 240 | 682 | 1026 | 629 | 0 | 0 | 0 | 12 |
+| example_07_large_scale_routing | 0.2005 | 10 | 10 | 10 | 1 | 10 | 78 | 383 | 372 | 227 | 0 | 0 | 0 | 40 |
+| example_08_custom_bend_geometry | 0.0176 | 2 | 2 | 2 | 2 | 2 | 18 | 56 | 78 | 56 | 0 | 0 | 0 | 8 |
+| example_09_unroutable_best_effort | 0.0058 | 1 | 0 | 0 | 1 | 1 | 3 | 13 | 16 | 10 | 0 | 0 | 0 | 1 |
 
 ## Full Counter Set
 
 Each scenario entry in `docs/performance_baseline.json` records the full `RouteMetrics` snapshot, including cache, index, congestion, and verification counters.
+These counters are currently observational only and are not enforced as CI regression gates.
+
+For the current accepted branch, the most important performance-only canary is `example_07_large_scale_routing_no_warm_start`, which now finishes `10/10/10` after a bounded post-loop pair-local scratch reroute. The relevant counters for that phase are:
+
+- `pair_local_search_pairs_considered`
+- `pair_local_search_attempts`
+- `pair_local_search_accepts`
+- `pair_local_search_nodes_expanded`
+
+The latest tracked characterization sweep confirms there is no smaller stable pair-local smoke case under the `<=1.0s` rule, so the 10-net no-warm-start canary remains the primary regression target for this behavior.
 
 Tracked metric keys:
 
-nodes_expanded, moves_generated, moves_added, pruned_closed_set, pruned_hard_collision, pruned_cost, route_iterations, nets_routed, nets_reached_target, warm_start_paths_built, warm_start_paths_used, refine_path_calls, timeout_events, move_cache_abs_hits, move_cache_abs_misses, move_cache_rel_hits, move_cache_rel_misses, static_safe_cache_hits, hard_collision_cache_hits, congestion_cache_hits, congestion_cache_misses, dynamic_path_objects_added, dynamic_path_objects_removed, dynamic_tree_rebuilds, dynamic_grid_rebuilds, static_tree_rebuilds, static_raw_tree_rebuilds, static_net_tree_rebuilds, visibility_builds, visibility_corner_pairs_checked, visibility_corner_queries, visibility_corner_hits, visibility_point_queries, visibility_point_cache_hits, visibility_point_cache_misses, ray_cast_calls, ray_cast_candidate_bounds, ray_cast_exact_geometry_checks, congestion_check_calls, congestion_exact_pair_checks, verify_path_report_calls, verify_static_buffer_ops, verify_dynamic_exact_pair_checks
+nodes_expanded, moves_generated, moves_added, pruned_closed_set, pruned_hard_collision, pruned_cost, route_iterations, nets_routed, nets_reached_target, warm_start_paths_built, warm_start_paths_used, refine_path_calls, timeout_events, iteration_reverify_calls, iteration_reverified_nets, iteration_conflicting_nets, iteration_conflict_edges, nets_carried_forward, score_component_calls, score_component_total_ns, path_cost_calls, danger_map_lookup_calls, danger_map_cache_hits, danger_map_cache_misses, danger_map_query_calls, danger_map_total_ns, move_cache_abs_hits, move_cache_abs_misses, move_cache_rel_hits, move_cache_rel_misses, guidance_match_moves, guidance_match_moves_straight, guidance_match_moves_bend90, guidance_match_moves_sbend, guidance_bonus_applied, guidance_bonus_applied_straight, guidance_bonus_applied_bend90, guidance_bonus_applied_sbend, static_safe_cache_hits, hard_collision_cache_hits, congestion_cache_hits, congestion_cache_misses, congestion_presence_cache_hits, congestion_presence_cache_misses, congestion_presence_skips, congestion_candidate_precheck_hits, congestion_candidate_precheck_misses, congestion_candidate_precheck_skips, congestion_grid_net_cache_hits, congestion_grid_net_cache_misses, congestion_grid_span_cache_hits, congestion_grid_span_cache_misses, congestion_candidate_nets, congestion_net_envelope_cache_hits, congestion_net_envelope_cache_misses, dynamic_path_objects_added, dynamic_path_objects_removed, dynamic_tree_rebuilds, dynamic_grid_rebuilds, static_tree_rebuilds, static_raw_tree_rebuilds, static_net_tree_rebuilds, visibility_corner_index_builds, visibility_builds, visibility_corner_pairs_checked, visibility_corner_queries_exact, visibility_corner_hits_exact, visibility_point_queries, visibility_point_cache_hits, visibility_point_cache_misses, visibility_tangent_candidate_scans, visibility_tangent_candidate_corner_checks, visibility_tangent_candidate_ray_tests, ray_cast_calls, ray_cast_calls_straight_static, ray_cast_calls_expand_snap, ray_cast_calls_expand_forward, ray_cast_calls_visibility_build, ray_cast_calls_visibility_query, ray_cast_calls_visibility_tangent, ray_cast_calls_other, ray_cast_candidate_bounds, ray_cast_exact_geometry_checks, congestion_check_calls, congestion_lazy_resolutions, congestion_lazy_requeues, congestion_candidate_ids, congestion_exact_pair_checks, verify_path_report_calls, verify_static_buffer_ops, verify_dynamic_candidate_nets, verify_dynamic_exact_pair_checks, refinement_windows_considered, refinement_static_bounds_checked, refinement_dynamic_bounds_checked, refinement_candidate_side_extents, refinement_candidates_built, refinement_candidates_verified, refinement_candidates_accepted, pair_local_search_pairs_considered, pair_local_search_attempts, pair_local_search_accepts, pair_local_search_nodes_expanded

inire/__init__.py

@@ -14,7 +14,15 @@ from .model import (
     RoutingProblem as RoutingProblem,
     SearchOptions as SearchOptions,
 )  # noqa: PLC0414
-from .results import RoutingResult as RoutingResult, RoutingRunResult as RoutingRunResult  # noqa: PLC0414
+from .results import (  # noqa: PLC0414
+    ComponentConflictTrace as ComponentConflictTrace,
+    ConflictTraceEntry as ConflictTraceEntry,
+    FrontierPruneSample as FrontierPruneSample,
+    NetConflictTrace as NetConflictTrace,
+    NetFrontierTrace as NetFrontierTrace,
+    RoutingResult as RoutingResult,
+    RoutingRunResult as RoutingRunResult,
+)
 from .seeds import Bend90Seed as Bend90Seed, PathSeed as PathSeed, SBendSeed as SBendSeed, StraightSeed as StraightSeed  # noqa: PLC0414
 
 __author__ = 'Jan Petykiewicz'
@@ -37,16 +45,23 @@ def route(
         results_by_net=results,
         metrics=finder.metrics.snapshot(),
         expanded_nodes=tuple(finder.accumulated_expanded_nodes),
+        conflict_trace=tuple(finder.conflict_trace),
+        frontier_trace=tuple(finder.frontier_trace),
     )
 
 __all__ = [
     "Bend90Seed",
     "CongestionOptions",
+    "ComponentConflictTrace",
+    "ConflictTraceEntry",
     "DiagnosticsOptions",
     "NetSpec",
+    "NetConflictTrace",
+    "NetFrontierTrace",
     "ObjectiveWeights",
     "PathSeed",
     "Port",
+    "FrontierPruneSample",
     "RefinementOptions",
     "RoutingOptions",
     "RoutingProblem",

inire/geometry/collision.py

@@ -1,5 +1,6 @@
 from __future__ import annotations
 
+from dataclasses import dataclass
 from typing import TYPE_CHECKING
 
 import numpy
@@ -28,6 +29,50 @@ def _intersection_distance(origin: Port, geometry: BaseGeometry) -> float:
     return float(numpy.sqrt((geometry.coords[0][0] - origin.x) ** 2 + (geometry.coords[0][1] - origin.y) ** 2))
 
 
+def _bounds_overlap(
+    left: tuple[float, float, float, float],
+    right: tuple[float, float, float, float],
+) -> bool:
+    return (
+        left[0] < right[2]
+        and left[2] > right[0]
+        and left[1] < right[3]
+        and left[3] > right[1]
+    )
+
+
+def _has_non_touching_overlap(left: BaseGeometry, right: BaseGeometry) -> bool:
+    return left.intersects(right) and not left.touches(right)
+
+
+def _span_to_bounds(
+    gx_min: int,
+    gy_min: int,
+    gx_max: int,
+    gy_max: int,
+    cell_size: float,
+) -> tuple[float, float, float, float]:
+    return (
+        gx_min * cell_size,
+        gy_min * cell_size,
+        (gx_max + 1) * cell_size,
+        (gy_max + 1) * cell_size,
+    )
+
+
+@dataclass(frozen=True, slots=True)
+class PathVerificationDetail:
+    report: RoutingReport
+    conflicting_net_ids: tuple[str, ...] = ()
+    component_conflicts: tuple[tuple[int, str, int], ...] = ()
+
+
+@dataclass(frozen=True, slots=True)
+class DynamicCongestionDetail:
+    soft_overlap_count: int = 0
+    hits_frozen_net: bool = False
+
+
 class RoutingWorld:
     """
     Internal spatial state for collision detection, congestion, and verification.
@@ -96,14 +141,34 @@ class RoutingWorld:
     def _ensure_dynamic_grid(self) -> None:
         self._dynamic_paths.ensure_grid()
 
-    def add_path(self, net_id: str, geometry: Sequence[Polygon], dilated_geometry: Sequence[Polygon]) -> None:
-        self._dynamic_paths.add_path(net_id, geometry, dilated_geometry=dilated_geometry)
+    def add_path(
+        self,
+        net_id: str,
+        geometry: Sequence[Polygon],
+        dilated_geometry: Sequence[Polygon],
+        component_indexes: Sequence[int] | None = None,
+    ) -> None:
+        self._dynamic_paths.add_path(
+            net_id,
+            geometry,
+            dilated_geometry=dilated_geometry,
+            component_indexes=component_indexes,
+        )
 
     def remove_path(self, net_id: str) -> None:
         self._dynamic_paths.remove_path(net_id)
 
+    def has_dynamic_paths(self) -> bool:
+        return bool(self._dynamic_paths.geometries)
+
     def check_move_straight_static(self, start_port: Port, length: float, net_width: float) -> bool:
-        reach = self.ray_cast(start_port, start_port.r, max_dist=length + 0.01, net_width=net_width)
+        reach = self.ray_cast(
+            start_port,
+            start_port.r,
+            max_dist=length + 0.01,
+            net_width=net_width,
+            caller="straight_static",
+        )
         return reach < length - 0.001
 
     def _is_in_safety_zone_fast(self, idx: int, start_port: Port | None, end_port: Port | None) -> bool:
@@ -229,105 +294,390 @@ class RoutingWorld:
 
         return False
 
-    def _check_real_congestion(self, result: ComponentResult, net_id: str) -> int:
+    def _check_real_congestion(
+        self,
+        result: ComponentResult,
+        candidates_by_net: dict[str, dict[int, tuple[int, ...]]],
+        frozen_net_ids: frozenset[str] = frozenset(),
+    ) -> DynamicCongestionDetail:
+        if not candidates_by_net:
+            return DynamicCongestionDetail()
+
         dynamic_paths = self._dynamic_paths
-        self._ensure_dynamic_tree()
-        if dynamic_paths.tree is None:
-            return 0
-
-        total_bounds = result.total_dilated_bounds
-        dynamic_bounds = dynamic_paths.bounds_array
-        possible_total = (
-            (total_bounds[0] < dynamic_bounds[:, 2])
-            & (total_bounds[2] > dynamic_bounds[:, 0])
-            & (total_bounds[1] < dynamic_bounds[:, 3])
-            & (total_bounds[3] > dynamic_bounds[:, 1])
-        )
-
-        valid_hits_mask = dynamic_paths.net_ids_array != net_id
-        if not numpy.any(possible_total & valid_hits_mask):
-            return 0
-
         geometries_to_test = result.dilated_collision_geometry
-        res_indices, tree_indices = dynamic_paths.tree.query(geometries_to_test, predicate="intersects")
-        if tree_indices.size == 0:
-            return 0
 
-        hit_net_ids = numpy.take(dynamic_paths.net_ids_array, tree_indices)
-        unique_other_nets = numpy.unique(hit_net_ids[hit_net_ids != net_id])
-        if unique_other_nets.size == 0:
-            return 0
-
-        tree_geometries = dynamic_paths.tree.geometries
         real_hits_count = 0
-        for other_net_id in unique_other_nets:
-            other_mask = hit_net_ids == other_net_id
-            sub_tree_indices = tree_indices[other_mask]
-            sub_res_indices = res_indices[other_mask]
-
+        for other_net_id, other_obj_ids in candidates_by_net.items():
             found_real = False
-            for index in range(len(sub_tree_indices)):
+            for obj_id, test_geometry_indexes in other_obj_ids.items():
+                tree_geometry = dynamic_paths.dilated[obj_id]
+                for test_geometry_index in test_geometry_indexes:
+                    test_geometry = geometries_to_test[test_geometry_index]
                     if self.metrics is not None:
                         self.metrics.total_congestion_exact_pair_checks += 1
-                test_geometry = geometries_to_test[sub_res_indices[index]]
-                tree_geometry = tree_geometries[sub_tree_indices[index]]
-                if not test_geometry.touches(tree_geometry) and test_geometry.intersection(tree_geometry).area > 1e-7:
+                    if _has_non_touching_overlap(test_geometry, tree_geometry):
                         found_real = True
                         break
-
                 if found_real:
+                    break
+            if found_real:
+                if other_net_id in frozen_net_ids:
+                    return DynamicCongestionDetail(
+                        soft_overlap_count=real_hits_count,
+                        hits_frozen_net=True,
+                    )
                 real_hits_count += 1
 
-        return real_hits_count
+        return DynamicCongestionDetail(soft_overlap_count=real_hits_count)
 
-    def check_move_congestion(self, result: ComponentResult, net_id: str) -> int:
+    def _collect_congestion_candidates(
+        self,
+        result: ComponentResult,
+        net_id: str,
+        net_envelope_cache: dict[tuple[str, int, int, int, int], tuple[str, ...]] | None = None,
+        grid_net_cache: dict[tuple[str, int, int, int, int], tuple[str, ...]] | None = None,
+        broad_phase_cache: dict[tuple[str, int, int, int, int], dict[str, tuple[int, ...]]] | None = None,
+    ) -> dict[str, dict[int, tuple[int, ...]]]:
+        dynamic_paths = self._dynamic_paths
+        if not dynamic_paths.dilated:
+            return {}
+
+        self._ensure_dynamic_grid()
+        if not dynamic_paths.grid:
+            return {}
+
+        candidates_by_net: dict[str, dict[int, set[int]]] = {}
+        for test_geometry_index, test_bounds in enumerate(result.dilated_bounds):
+            if not self._has_possible_congestion_in_grid(test_bounds, net_id):
+                continue
+            envelope_net_ids = self._get_net_envelope_candidates(
+                test_bounds,
+                net_id,
+                net_envelope_cache,
+            )
+            if not envelope_net_ids:
+                continue
+            grid_net_ids = self._get_grid_span_net_candidates(
+                test_bounds,
+                net_id,
+                grid_net_cache,
+            )
+            if not grid_net_ids:
+                continue
+            candidate_net_ids = tuple(sorted(set(envelope_net_ids) & set(grid_net_ids)))
+            if not candidate_net_ids:
+                continue
+            grid_candidates = self._get_grid_span_candidates(
+                test_bounds,
+                net_id,
+                broad_phase_cache,
+            )
+            for other_net_id in candidate_net_ids:
+                if self.metrics is not None:
+                    self.metrics.total_congestion_candidate_nets += 1
+                obj_ids = grid_candidates.get(other_net_id)
+                if not obj_ids:
+                    continue
+                for obj_id in obj_ids:
+                    if not _bounds_overlap(test_bounds, dynamic_paths.dilated_bounds[obj_id]):
+                        continue
+                    if self.metrics is not None:
+                        self.metrics.total_congestion_candidate_ids += 1
+                    candidate_indexes = candidates_by_net.setdefault(other_net_id, {}).setdefault(obj_id, set())
+                    candidate_indexes.add(test_geometry_index)
+
+        return {
+            other_net_id: {
+                obj_id: tuple(sorted(test_geometry_indexes))
+                for obj_id, test_geometry_indexes in sorted(obj_ids.items())
+            }
+            for other_net_id, obj_ids in candidates_by_net.items()
+        }
+
+    def _has_possible_congestion_in_grid(
+        self,
+        bounds: tuple[float, float, float, float],
+        net_id: str,
+    ) -> bool:
+        gx_min, gy_min, gx_max, gy_max = grid_cell_span(bounds, self.grid_cell_size)
+        dynamic_paths = self._dynamic_paths
+
+        if gx_min == gx_max and gy_min == gy_max:
+            net_counts = dynamic_paths.grid_net_counts.get((gx_min, gy_min))
+            return bool(net_counts and (len(net_counts) > 1 or net_id not in net_counts))
+
+        for gx in range(gx_min, gx_max + 1):
+            for gy in range(gy_min, gy_max + 1):
+                net_counts = dynamic_paths.grid_net_counts.get((gx, gy))
+                if net_counts and (len(net_counts) > 1 or net_id not in net_counts):
+                    return True
+        return False
+
+    def has_possible_move_congestion(
+        self,
+        result: ComponentResult,
+        net_id: str,
+        presence_cache: dict[tuple[str, int, int, int, int], bool] | None = None,
+    ) -> bool:
+        dynamic_paths = self._dynamic_paths
+        if not dynamic_paths.dilated:
+            return False
+
+        self._ensure_dynamic_grid()
+        if not dynamic_paths.grid:
+            return False
+
+        for bounds in result.dilated_bounds:
+            gx_min, gy_min, gx_max, gy_max = grid_cell_span(bounds, self.grid_cell_size)
+            cache_key = (net_id, gx_min, gy_min, gx_max, gy_max)
+            if presence_cache is not None and cache_key in presence_cache:
+                if self.metrics is not None:
+                    self.metrics.total_congestion_presence_cache_hits += 1
+                has_possible = presence_cache[cache_key]
+            else:
+                if self.metrics is not None:
+                    self.metrics.total_congestion_presence_cache_misses += 1
+                has_possible = self._has_possible_congestion_in_grid(bounds, net_id)
+                if presence_cache is not None:
+                    presence_cache[cache_key] = has_possible
+            if has_possible:
+                return True
+        return False
+
+    def has_candidate_move_congestion(
+        self,
+        result: ComponentResult,
+        net_id: str,
+        candidate_precheck_cache: dict[tuple[str, int, int, int, int], bool] | None = None,
+        net_envelope_cache: dict[tuple[str, int, int, int, int], tuple[str, ...]] | None = None,
+        grid_net_cache: dict[tuple[str, int, int, int, int], tuple[str, ...]] | None = None,
+    ) -> bool:
+        dynamic_paths = self._dynamic_paths
+        if not dynamic_paths.dilated:
+            return False
+
+        for bounds in result.dilated_bounds:
+            gx_min, gy_min, gx_max, gy_max = grid_cell_span(bounds, self.grid_cell_size)
+            cache_key = (net_id, gx_min, gy_min, gx_max, gy_max)
+            if candidate_precheck_cache is not None and cache_key in candidate_precheck_cache:
+                if self.metrics is not None:
+                    self.metrics.total_congestion_candidate_precheck_hits += 1
+                has_candidates = candidate_precheck_cache[cache_key]
+            else:
+                if self.metrics is not None:
+                    self.metrics.total_congestion_candidate_precheck_misses += 1
+                span_bounds = _span_to_bounds(gx_min, gy_min, gx_max, gy_max, self.grid_cell_size)
+                envelope_net_ids = self._get_net_envelope_candidates(
+                    span_bounds,
+                    net_id,
+                    net_envelope_cache,
+                )
+                if not envelope_net_ids:
+                    has_candidates = False
+                else:
+                    grid_net_ids = self._get_grid_span_net_candidates(
+                        span_bounds,
+                        net_id,
+                        grid_net_cache,
+                    )
+                    if not grid_net_ids:
+                        has_candidates = False
+                    else:
+                        has_candidates = bool(set(envelope_net_ids) & set(grid_net_ids))
+                if candidate_precheck_cache is not None:
+                    candidate_precheck_cache[cache_key] = has_candidates
+            if has_candidates:
+                return True
+        return False
+
+    def _get_grid_span_candidates(
+        self,
+        bounds: tuple[float, float, float, float],
+        net_id: str,
+        broad_phase_cache: dict[tuple[str, int, int, int, int], dict[str, tuple[int, ...]]] | None,
+    ) -> dict[str, tuple[int, ...]]:
+        gx_min, gy_min, gx_max, gy_max = grid_cell_span(bounds, self.grid_cell_size)
+        cache_key = (net_id, gx_min, gy_min, gx_max, gy_max)
+        if broad_phase_cache is not None and cache_key in broad_phase_cache:
+            if self.metrics is not None:
+                self.metrics.total_congestion_grid_span_cache_hits += 1
+            return broad_phase_cache[cache_key]
+
+        if self.metrics is not None:
+            self.metrics.total_congestion_grid_span_cache_misses += 1
+
+        dynamic_paths = self._dynamic_paths
+        candidates_by_net: dict[str, set[int]] = {}
+        for gx in range(gx_min, gx_max + 1):
+            for gy in range(gy_min, gy_max + 1):
+                for other_net_id, obj_ids in dynamic_paths.grid_net_obj_ids.get((gx, gy), {}).items():
+                    if other_net_id == net_id:
+                        continue
+                    candidates_by_net.setdefault(other_net_id, set()).update(obj_ids)
+
+        frozen = {
+            other_net_id: tuple(sorted(obj_ids))
+            for other_net_id, obj_ids in sorted(candidates_by_net.items())
+        }
+        if broad_phase_cache is not None:
+            broad_phase_cache[cache_key] = frozen
+        return frozen
+
+    def _get_grid_span_net_candidates(
+        self,
+        bounds: tuple[float, float, float, float],
+        net_id: str,
+        grid_net_cache: dict[tuple[str, int, int, int, int], tuple[str, ...]] | None,
+    ) -> tuple[str, ...]:
+        gx_min, gy_min, gx_max, gy_max = grid_cell_span(bounds, self.grid_cell_size)
+        cache_key = (net_id, gx_min, gy_min, gx_max, gy_max)
+        if grid_net_cache is not None and cache_key in grid_net_cache:
+            if self.metrics is not None:
+                self.metrics.total_congestion_grid_net_cache_hits += 1
+            return grid_net_cache[cache_key]
+
+        if self.metrics is not None:
+            self.metrics.total_congestion_grid_net_cache_misses += 1
+
+        dynamic_paths = self._dynamic_paths
+        candidate_net_ids: set[str] = set()
+        for gx in range(gx_min, gx_max + 1):
+            for gy in range(gy_min, gy_max + 1):
+                for other_net_id in dynamic_paths.grid_net_obj_ids.get((gx, gy), {}):
+                    if other_net_id != net_id:
+                        candidate_net_ids.add(other_net_id)
+
+        frozen = tuple(sorted(candidate_net_ids))
+        if grid_net_cache is not None:
+            grid_net_cache[cache_key] = frozen
+        return frozen
+
+    def _get_net_envelope_candidates(
+        self,
+        bounds: tuple[float, float, float, float],
+        net_id: str,
+        net_envelope_cache: dict[tuple[str, int, int, int, int], tuple[str, ...]] | None,
+    ) -> tuple[str, ...]:
+        dynamic_paths = self._dynamic_paths
+        gx_min, gy_min, gx_max, gy_max = grid_cell_span(bounds, self.grid_cell_size)
+        cache_key = (net_id, gx_min, gy_min, gx_max, gy_max)
+        if net_envelope_cache is not None and cache_key in net_envelope_cache:
+            if self.metrics is not None:
+                self.metrics.total_congestion_net_envelope_cache_hits += 1
+            cached_net_ids = net_envelope_cache[cache_key]
+        else:
+            if self.metrics is not None:
+                self.metrics.total_congestion_net_envelope_cache_misses += 1
+            span_bounds = _span_to_bounds(gx_min, gy_min, gx_max, gy_max, self.grid_cell_size)
+            cached_net_ids = tuple(
+                sorted(
+                    dynamic_paths.net_envelope_obj_to_net[obj_id]
+                    for obj_id in dynamic_paths.net_envelope_index.intersection(span_bounds)
+                    if dynamic_paths.net_envelope_obj_to_net[obj_id] != net_id
+                )
+            )
+            if net_envelope_cache is not None:
+                net_envelope_cache[cache_key] = cached_net_ids
+
+        return tuple(
+            other_net_id
+            for other_net_id in cached_net_ids
+            if _bounds_overlap(bounds, dynamic_paths.net_envelopes[other_net_id])
+        )
+
+    def _get_verify_net_envelope_candidates(
+        self,
+        bounds: tuple[float, float, float, float],
+        net_id: str,
+    ) -> tuple[str, ...]:
+        dynamic_paths = self._dynamic_paths
+        candidate_net_ids: list[str] = []
+        for obj_id in dynamic_paths.net_envelope_index.intersection(bounds):
+            other_net_id = dynamic_paths.net_envelope_obj_to_net[obj_id]
+            if other_net_id == net_id:
+                continue
+            if not _bounds_overlap(bounds, dynamic_paths.net_envelopes[other_net_id]):
+                continue
+            candidate_net_ids.append(other_net_id)
+        if self.metrics is not None:
+            self.metrics.total_verify_dynamic_candidate_nets += len(candidate_net_ids)
+        return tuple(candidate_net_ids)
+
+    def _get_verify_grid_span_obj_ids(
+        self,
+        bounds: tuple[float, float, float, float],
+        other_net_id: str,
+    ) -> tuple[int, ...]:
+        dynamic_paths = self._dynamic_paths
+        gx_min, gy_min, gx_max, gy_max = grid_cell_span(bounds, self.grid_cell_size)
+        obj_ids: set[int] = set()
+        for gx in range(gx_min, gx_max + 1):
+            for gy in range(gy_min, gy_max + 1):
+                obj_ids.update(dynamic_paths.grid_net_obj_ids.get((gx, gy), {}).get(other_net_id, ()))
+        return tuple(sorted(obj_ids))
+
+    def check_move_congestion(
+        self,
+        result: ComponentResult,
+        net_id: str,
+        net_envelope_cache: dict[tuple[str, int, int, int, int], tuple[str, ...]] | None = None,
+        grid_net_cache: dict[tuple[str, int, int, int, int], tuple[str, ...]] | None = None,
+        broad_phase_cache: dict[tuple[str, int, int, int, int], dict[str, tuple[int, ...]]] | None = None,
+    ) -> int:
+        return self.check_move_congestion_detail(
+            result,
+            net_id,
+            net_envelope_cache=net_envelope_cache,
+            grid_net_cache=grid_net_cache,
+            broad_phase_cache=broad_phase_cache,
+        ).soft_overlap_count
+
+    def check_move_congestion_detail(
+        self,
+        result: ComponentResult,
+        net_id: str,
+        *,
+        frozen_net_ids: frozenset[str] = frozenset(),
+        net_envelope_cache: dict[tuple[str, int, int, int, int], tuple[str, ...]] | None = None,
+        grid_net_cache: dict[tuple[str, int, int, int, int], tuple[str, ...]] | None = None,
+        broad_phase_cache: dict[tuple[str, int, int, int, int], dict[str, tuple[int, ...]]] | None = None,
+    ) -> DynamicCongestionDetail:
         if self.metrics is not None:
             self.metrics.total_congestion_check_calls += 1
         dynamic_paths = self._dynamic_paths
         if not dynamic_paths.geometries:
-            return 0
+            return DynamicCongestionDetail()
 
-        total_bounds = result.total_dilated_bounds
-        self._ensure_dynamic_grid()
-        dynamic_grid = dynamic_paths.grid
-        if not dynamic_grid:
-            return 0
+        candidates_by_net = self._collect_congestion_candidates(
+            result,
+            net_id,
+            net_envelope_cache,
+            grid_net_cache,
+            broad_phase_cache,
+        )
+        if not candidates_by_net:
+            return DynamicCongestionDetail()
+        return self._check_real_congestion(
+            result,
+            candidates_by_net,
+            frozen_net_ids=frozen_net_ids,
+        )
 
-        gx_min, gy_min, gx_max, gy_max = grid_cell_span(total_bounds, self.grid_cell_size)
-
-        if gx_min == gx_max and gy_min == gy_max:
-            cell = (gx_min, gy_min)
-            if cell in dynamic_grid:
-                for obj_id in dynamic_grid[cell]:
-                    if dynamic_paths.geometries[obj_id][0] != net_id:
-                        return self._check_real_congestion(result, net_id)
-            return 0
-
-        any_possible = False
-        for gx in range(gx_min, gx_max + 1):
-            for gy in range(gy_min, gy_max + 1):
-                cell = (gx, gy)
-                if cell in dynamic_grid:
-                    for obj_id in dynamic_grid[cell]:
-                        if dynamic_paths.geometries[obj_id][0] != net_id:
-                            any_possible = True
-                            break
-                if any_possible:
-                    break
-            if any_possible:
-                break
-
-        if not any_possible:
-            return 0
-        return self._check_real_congestion(result, net_id)
-
-    def verify_path_report(self, net_id: str, components: Sequence[ComponentResult]) -> RoutingReport:
+    def verify_path_details(
+        self,
+        net_id: str,
+        components: Sequence[ComponentResult],
+        *,
+        capture_component_conflicts: bool = False,
+    ) -> PathVerificationDetail:
         if self.metrics is not None:
             self.metrics.total_verify_path_report_calls += 1
         static_collision_count = 0
         dynamic_collision_count = 0
         self_collision_count = 0
         total_length = sum(component.length for component in components)
+        conflicting_net_ids: set[str] = set()
+        component_conflicts: set[tuple[int, str, int]] = set()
 
         static_obstacles = self._static_obstacles
         dynamic_paths = self._dynamic_paths
@@ -350,52 +700,78 @@ class RoutingWorld:
                         if not self._is_in_safety_zone(polygon, obj_id, None, None):
                             static_collision_count += 1
 
-        self._ensure_dynamic_tree()
-        if dynamic_paths.tree is not None:
-            tree_geometries = dynamic_paths.tree.geometries
-            for component in components:
+        if dynamic_paths.dilated:
+            for component_index, component in enumerate(components):
                 test_geometries = component.dilated_physical_geometry
-                res_indices, tree_indices = dynamic_paths.tree.query(test_geometries, predicate="intersects")
-                if tree_indices.size == 0:
-                    continue
-
-                hit_net_ids = numpy.take(dynamic_paths.net_ids_array, tree_indices)
                 component_hits = []
-                for index in range(len(tree_indices)):
-                    if hit_net_ids[index] == str(net_id):
+                for new_geometry in test_geometries:
+                    for hit_net_id in self._get_verify_net_envelope_candidates(new_geometry.bounds, str(net_id)):
+                        for obj_id in self._get_verify_grid_span_obj_ids(new_geometry.bounds, hit_net_id):
+                            if not _bounds_overlap(new_geometry.bounds, dynamic_paths.dilated_bounds[obj_id]):
                                 continue
-
                             if self.metrics is not None:
                                 self.metrics.total_verify_dynamic_exact_pair_checks += 1
-                    new_geometry = test_geometries[res_indices[index]]
-                    tree_geometry = tree_geometries[tree_indices[index]]
-                    if not new_geometry.touches(tree_geometry) and new_geometry.intersection(tree_geometry).area > 1e-7:
-                        component_hits.append(hit_net_ids[index])
+                            tree_geometry = dynamic_paths.dilated[obj_id]
+                            if _has_non_touching_overlap(new_geometry, tree_geometry):
+                                component_hits.append(hit_net_id)
+                                if capture_component_conflicts:
+                                    component_conflicts.add(
+                                        (
+                                            component_index,
+                                            hit_net_id,
+                                            dynamic_paths.component_indexes[obj_id],
+                                        )
+                                    )
+                                break
 
                 if component_hits:
-                    dynamic_collision_count += len(numpy.unique(component_hits))
+                    unique_hits = tuple(sorted(set(component_hits)))
+                    dynamic_collision_count += len(unique_hits)
+                    conflicting_net_ids.update(unique_hits)
 
         for index, component in enumerate(components):
             for other_index in range(index + 2, len(components)):
                 if components_overlap(component, components[other_index], prefer_actual=True):
                     self_collision_count += 1
 
-        return RoutingReport(
+        return PathVerificationDetail(
+            report=RoutingReport(
                 static_collision_count=static_collision_count,
                 dynamic_collision_count=dynamic_collision_count,
                 self_collision_count=self_collision_count,
                 total_length=total_length,
+            ),
+            conflicting_net_ids=tuple(sorted(conflicting_net_ids)),
+            component_conflicts=tuple(sorted(component_conflicts)),
         )
 
+    def verify_path_report(self, net_id: str, components: Sequence[ComponentResult]) -> RoutingReport:
+        return self.verify_path_details(net_id, components).report
+
     def ray_cast(
         self,
         origin: Port,
         angle_deg: float,
         max_dist: float = 2000.0,
         net_width: float | None = None,
+        caller: str = "other",
     ) -> float:
         if self.metrics is not None:
             self.metrics.total_ray_cast_calls += 1
+            if caller == "straight_static":
+                self.metrics.total_ray_cast_calls_straight_static += 1
+            elif caller == "expand_snap":
+                self.metrics.total_ray_cast_calls_expand_snap += 1
+            elif caller == "expand_forward":
+                self.metrics.total_ray_cast_calls_expand_forward += 1
+            elif caller == "visibility_build":
+                self.metrics.total_ray_cast_calls_visibility_build += 1
+            elif caller == "visibility_query":
+                self.metrics.total_ray_cast_calls_visibility_query += 1
+            elif caller == "visibility_tangent":
+                self.metrics.total_ray_cast_calls_visibility_tangent += 1
+            else:
+                self.metrics.total_ray_cast_calls_other += 1
         static_obstacles = self._static_obstacles
         tree: STRtree | None
         is_rect_array: numpy.ndarray | None

inire/geometry/dynamic_path_index.py

@@ -21,10 +21,20 @@ class DynamicPathIndex:
         "engine",
         "index",
         "geometries",
+        "component_indexes",
         "dilated",
+        "dilated_bounds",
+        "net_envelope_index",
+        "net_envelopes",
+        "net_envelope_obj_ids",
+        "net_envelope_obj_to_net",
         "tree",
         "obj_ids",
         "grid",
+        "grid_net_obj_ids",
+        "grid_net_counts",
+        "obj_cells",
+        "net_to_obj_ids",
         "id_counter",
         "net_ids_array",
         "bounds_array",
@@ -34,17 +44,71 @@ class DynamicPathIndex:
         self.engine = engine
         self.index = rtree.index.Index()
         self.geometries: dict[int, tuple[str, Polygon]] = {}
+        self.component_indexes: dict[int, int] = {}
         self.dilated: dict[int, Polygon] = {}
+        self.dilated_bounds: dict[int, tuple[float, float, float, float]] = {}
+        self.net_envelope_index = rtree.index.Index()
+        self.net_envelopes: dict[str, tuple[float, float, float, float]] = {}
+        self.net_envelope_obj_ids: dict[str, int] = {}
+        self.net_envelope_obj_to_net: dict[int, str] = {}
         self.tree: STRtree | None = None
         self.obj_ids: numpy.ndarray = numpy.array([], dtype=numpy.int32)
         self.grid: dict[tuple[int, int], list[int]] = {}
+        self.grid_net_obj_ids: dict[tuple[int, int], dict[str, set[int]]] = {}
+        self.grid_net_counts: dict[tuple[int, int], dict[str, int]] = {}
+        self.obj_cells: dict[int, tuple[tuple[int, int], ...]] = {}
+        self.net_to_obj_ids: dict[str, set[int]] = {}
         self.id_counter = 0
         self.net_ids_array = numpy.array([], dtype=object)
         self.bounds_array = numpy.array([], dtype=numpy.float64).reshape(0, 4)
 
+    def _combine_net_bounds(self, obj_ids: set[int]) -> tuple[float, float, float, float]:
+        first_obj_id = next(iter(obj_ids))
+        minx, miny, maxx, maxy = self.dilated_bounds[first_obj_id]
+        for obj_id in obj_ids:
+            bounds = self.dilated_bounds[obj_id]
+            minx = min(minx, bounds[0])
+            miny = min(miny, bounds[1])
+            maxx = max(maxx, bounds[2])
+            maxy = max(maxy, bounds[3])
+        return (minx, miny, maxx, maxy)
+
+    def _set_net_envelope(self, net_id: str, bounds: tuple[float, float, float, float]) -> None:
+        old_bounds = self.net_envelopes.get(net_id)
+        if old_bounds is not None:
+            obj_id = self.net_envelope_obj_ids[net_id]
+            self.net_envelope_index.delete(obj_id, old_bounds)
+        else:
+            obj_id = len(self.net_envelope_obj_ids)
+            while obj_id in self.net_envelope_obj_to_net:
+                obj_id += 1
+            self.net_envelope_obj_ids[net_id] = obj_id
+            self.net_envelope_obj_to_net[obj_id] = net_id
+
+        self.net_envelopes[net_id] = bounds
+        self.net_envelope_index.insert(self.net_envelope_obj_ids[net_id], bounds)
+
+    def _clear_net_envelope(self, net_id: str) -> None:
+        old_bounds = self.net_envelopes.pop(net_id, None)
+        obj_id = self.net_envelope_obj_ids.pop(net_id, None)
+        if old_bounds is None or obj_id is None:
+            return
+        self.net_envelope_index.delete(obj_id, old_bounds)
+        self.net_envelope_obj_to_net.pop(obj_id, None)
+
+    def _refresh_net_envelope(self, net_id: str) -> None:
+        obj_ids = self.net_to_obj_ids.get(net_id)
+        if not obj_ids:
+            self._clear_net_envelope(net_id)
+            return
+        self._set_net_envelope(net_id, self._combine_net_bounds(obj_ids))
+
     def invalidate_queries(self) -> None:
         self.tree = None
         self.grid = {}
+        self.grid_net_obj_ids = {}
+        self.grid_net_counts = {}
+        self.obj_cells = {}
 
     def ensure_tree(self) -> None:
         if self.tree is None and self.dilated:
@@ -65,33 +129,105 @@ class DynamicPathIndex:
             self.engine.metrics.total_dynamic_grid_rebuilds += 1
         cell_size = self.engine.grid_cell_size
         for obj_id, polygon in self.dilated.items():
-            for cell in iter_grid_cells(polygon.bounds, cell_size):
-                self.grid.setdefault(cell, []).append(obj_id)
+            self._register_grid_membership(obj_id, self.geometries[obj_id][0], polygon.bounds, cell_size=cell_size)
 
-    def add_path(self, net_id: str, geometry: Sequence[Polygon], dilated_geometry: Sequence[Polygon]) -> None:
-        self.invalidate_queries()
+    def _register_grid_membership(
+        self,
+        obj_id: int,
+        net_id: str,
+        bounds: tuple[float, float, float, float],
+        *,
+        cell_size: float,
+    ) -> None:
+        cells = tuple(iter_grid_cells(bounds, cell_size))
+        self.obj_cells[obj_id] = cells
+        for cell in cells:
+            self.grid.setdefault(cell, []).append(obj_id)
+            net_obj_ids = self.grid_net_obj_ids.setdefault(cell, {})
+            net_obj_ids.setdefault(net_id, set()).add(obj_id)
+            net_counts = self.grid_net_counts.setdefault(cell, {})
+            net_counts[net_id] = net_counts.get(net_id, 0) + 1
+
+    def _unregister_grid_membership(self, obj_id: int, net_id: str) -> None:
+        cells = self.obj_cells.pop(obj_id, ())
+        for cell in cells:
+            obj_ids = self.grid.get(cell)
+            if obj_ids is not None:
+                try:
+                    obj_ids.remove(obj_id)
+                except ValueError:
+                    pass
+                if not obj_ids:
+                    self.grid.pop(cell, None)
+            net_obj_ids = self.grid_net_obj_ids.get(cell)
+            if net_obj_ids is not None:
+                member_ids = net_obj_ids.get(net_id)
+                if member_ids is not None:
+                    member_ids.discard(obj_id)
+                    if not member_ids:
+                        net_obj_ids.pop(net_id, None)
+                if not net_obj_ids:
+                    self.grid_net_obj_ids.pop(cell, None)
+            net_counts = self.grid_net_counts.get(cell)
+            if net_counts is not None:
+                remaining = net_counts.get(net_id, 0) - 1
+                if remaining > 0:
+                    net_counts[net_id] = remaining
+                else:
+                    net_counts.pop(net_id, None)
+                if not net_counts:
+                    self.grid_net_counts.pop(cell, None)
+
+    def add_path(
+        self,
+        net_id: str,
+        geometry: Sequence[Polygon],
+        dilated_geometry: Sequence[Polygon],
+        component_indexes: Sequence[int] | None = None,
+    ) -> None:
         if self.engine.metrics is not None:
             self.engine.metrics.total_dynamic_path_objects_added += len(geometry)
+        cell_size = self.engine.grid_cell_size
         for index, polygon in enumerate(geometry):
             obj_id = self.id_counter
             self.id_counter += 1
             dilated = dilated_geometry[index]
+            dilated_bounds = dilated.bounds
             self.geometries[obj_id] = (net_id, polygon)
+            self.component_indexes[obj_id] = index if component_indexes is None else component_indexes[index]
             self.dilated[obj_id] = dilated
-            self.index.insert(obj_id, dilated.bounds)
+            self.dilated_bounds[obj_id] = dilated_bounds
+            self.index.insert(obj_id, dilated_bounds)
+            self.net_to_obj_ids.setdefault(net_id, set()).add(obj_id)
+            self._register_grid_membership(obj_id, net_id, dilated_bounds, cell_size=cell_size)
+        self._refresh_net_envelope(net_id)
+        self.tree = None
 
     def remove_path(self, net_id: str) -> None:
-        to_remove = [obj_id for obj_id, (existing_net_id, _) in self.geometries.items() if existing_net_id == net_id]
+        to_remove = list(self.net_to_obj_ids.get(net_id, ()))
         self.remove_obj_ids(to_remove)
 
     def remove_obj_ids(self, obj_ids: list[int]) -> None:
         if not obj_ids:
             return
 
-        self.invalidate_queries()
         if self.engine.metrics is not None:
             self.engine.metrics.total_dynamic_path_objects_removed += len(obj_ids)
+        affected_nets: set[str] = set()
         for obj_id in obj_ids:
-            self.index.delete(obj_id, self.dilated[obj_id].bounds)
+            net_id, _ = self.geometries[obj_id]
+            affected_nets.add(net_id)
+            self._unregister_grid_membership(obj_id, net_id)
+            self.index.delete(obj_id, self.dilated_bounds[obj_id])
             del self.geometries[obj_id]
+            del self.component_indexes[obj_id]
             del self.dilated[obj_id]
+            del self.dilated_bounds[obj_id]
+            obj_id_set = self.net_to_obj_ids.get(net_id)
+            if obj_id_set is not None:
+                obj_id_set.discard(obj_id)
+                if not obj_id_set:
+                    self.net_to_obj_ids.pop(net_id, None)
+        for net_id in affected_nets:
+            self._refresh_net_envelope(net_id)
+        self.tree = None

inire/model.py

@@ -105,6 +105,8 @@ class RefinementOptions:
 @dataclass(frozen=True, slots=True)
 class DiagnosticsOptions:
     capture_expanded: bool = False
+    capture_conflict_trace: bool = False
+    capture_frontier_trace: bool = False
 
 
 @dataclass(frozen=True, slots=True)

inire/results.py

@@ -12,6 +12,8 @@ if TYPE_CHECKING:
 
 
 RoutingOutcome = Literal["completed", "colliding", "partial", "unroutable"]
+ConflictTraceStage = Literal["iteration", "restored_best", "final"]
+FrontierTraceReason = Literal["closed_set", "hard_collision", "self_collision", "cost"]
 
 
 @dataclass(frozen=True, slots=True)
@@ -30,6 +32,52 @@ class RoutingReport:
         return self.collision_count == 0
 
 
+@dataclass(frozen=True, slots=True)
+class ComponentConflictTrace:
+    other_net_id: str
+    self_component_index: int
+    other_component_index: int
+
+
+@dataclass(frozen=True, slots=True)
+class NetConflictTrace:
+    net_id: str
+    outcome: RoutingOutcome
+    reached_target: bool
+    report: RoutingReport
+    conflicting_net_ids: tuple[str, ...] = ()
+    component_conflicts: tuple[ComponentConflictTrace, ...] = ()
+
+
+@dataclass(frozen=True, slots=True)
+class ConflictTraceEntry:
+    stage: ConflictTraceStage
+    iteration: int | None
+    completed_net_ids: tuple[str, ...]
+    conflict_edges: tuple[tuple[str, str], ...]
+    nets: tuple[NetConflictTrace, ...]
+
+
+@dataclass(frozen=True, slots=True)
+class FrontierPruneSample:
+    reason: FrontierTraceReason
+    move_type: str
+    hotspot_index: int
+    parent_state: tuple[int, int, int]
+    end_state: tuple[int, int, int]
+
+
+@dataclass(frozen=True, slots=True)
+class NetFrontierTrace:
+    net_id: str
+    hotspot_bounds: tuple[tuple[float, float, float, float], ...]
+    pruned_closed_set: int
+    pruned_hard_collision: int
+    pruned_self_collision: int
+    pruned_cost: int
+    samples: tuple[FrontierPruneSample, ...] = ()
+
+
 @dataclass(frozen=True, slots=True)
 class RouteMetrics:
     nodes_expanded: int
@@ -45,14 +93,48 @@ class RouteMetrics:
     warm_start_paths_used: int
     refine_path_calls: int
     timeout_events: int
+    iteration_reverify_calls: int
+    iteration_reverified_nets: int
+    iteration_conflicting_nets: int
+    iteration_conflict_edges: int
+    nets_carried_forward: int
+    score_component_calls: int
+    score_component_total_ns: int
+    path_cost_calls: int
+    danger_map_lookup_calls: int
+    danger_map_cache_hits: int
+    danger_map_cache_misses: int
+    danger_map_query_calls: int
+    danger_map_total_ns: int
     move_cache_abs_hits: int
     move_cache_abs_misses: int
     move_cache_rel_hits: int
     move_cache_rel_misses: int
+    guidance_match_moves: int
+    guidance_match_moves_straight: int
+    guidance_match_moves_bend90: int
+    guidance_match_moves_sbend: int
+    guidance_bonus_applied: float
+    guidance_bonus_applied_straight: float
+    guidance_bonus_applied_bend90: float
+    guidance_bonus_applied_sbend: float
     static_safe_cache_hits: int
     hard_collision_cache_hits: int
     congestion_cache_hits: int
     congestion_cache_misses: int
+    congestion_presence_cache_hits: int
+    congestion_presence_cache_misses: int
+    congestion_presence_skips: int
+    congestion_candidate_precheck_hits: int
+    congestion_candidate_precheck_misses: int
+    congestion_candidate_precheck_skips: int
+    congestion_grid_net_cache_hits: int
+    congestion_grid_net_cache_misses: int
+    congestion_grid_span_cache_hits: int
+    congestion_grid_span_cache_misses: int
+    congestion_candidate_nets: int
+    congestion_net_envelope_cache_hits: int
+    congestion_net_envelope_cache_misses: int
     dynamic_path_objects_added: int
     dynamic_path_objects_removed: int
     dynamic_tree_rebuilds: int
@@ -60,21 +142,47 @@ class RouteMetrics:
     static_tree_rebuilds: int
     static_raw_tree_rebuilds: int
     static_net_tree_rebuilds: int
+    visibility_corner_index_builds: int
     visibility_builds: int
     visibility_corner_pairs_checked: int
-    visibility_corner_queries: int
-    visibility_corner_hits: int
+    visibility_corner_queries_exact: int
+    visibility_corner_hits_exact: int
     visibility_point_queries: int
     visibility_point_cache_hits: int
     visibility_point_cache_misses: int
+    visibility_tangent_candidate_scans: int
+    visibility_tangent_candidate_corner_checks: int
+    visibility_tangent_candidate_ray_tests: int
     ray_cast_calls: int
+    ray_cast_calls_straight_static: int
+    ray_cast_calls_expand_snap: int
+    ray_cast_calls_expand_forward: int
+    ray_cast_calls_visibility_build: int
+    ray_cast_calls_visibility_query: int
+    ray_cast_calls_visibility_tangent: int
+    ray_cast_calls_other: int
     ray_cast_candidate_bounds: int
     ray_cast_exact_geometry_checks: int
     congestion_check_calls: int
+    congestion_lazy_resolutions: int
+    congestion_lazy_requeues: int
+    congestion_candidate_ids: int
     congestion_exact_pair_checks: int
     verify_path_report_calls: int
     verify_static_buffer_ops: int
+    verify_dynamic_candidate_nets: int
     verify_dynamic_exact_pair_checks: int
+    refinement_windows_considered: int
+    refinement_static_bounds_checked: int
+    refinement_dynamic_bounds_checked: int
+    refinement_candidate_side_extents: int
+    refinement_candidates_built: int
+    refinement_candidates_verified: int
+    refinement_candidates_accepted: int
+    pair_local_search_pairs_considered: int
+    pair_local_search_attempts: int
+    pair_local_search_accepts: int
+    pair_local_search_nodes_expanded: int
 
 
 @dataclass(frozen=True, slots=True)
@@ -121,3 +229,5 @@ class RoutingRunResult:
     results_by_net: dict[str, RoutingResult]
     metrics: RouteMetrics
     expanded_nodes: tuple[tuple[int, int, int], ...] = ()
+    conflict_trace: tuple[ConflictTraceEntry, ...] = ()
+    frontier_trace: tuple[NetFrontierTrace, ...] = ()

inire/router/_astar_admission.py

@@ -26,6 +26,11 @@ def process_move(
     context: AStarContext,
     metrics: AStarMetrics,
     congestion_cache: dict[tuple, int],
+    congestion_presence_cache: dict[tuple[str, int, int, int, int], bool],
+    congestion_candidate_precheck_cache: dict[tuple[str, int, int, int, int], bool],
+    congestion_net_envelope_cache: dict[tuple[str, int, int, int, int], tuple[str, ...]],
+    congestion_grid_net_cache: dict[tuple[str, int, int, int, int], tuple[str, ...]],
+    congestion_grid_span_cache: dict[tuple[str, int, int, int, int], dict[str, tuple[int, ...]]],
     config: SearchRunConfig,
     move_class: MoveKind,
     params: tuple,
@@ -109,6 +114,11 @@ def process_move(
         context,
         metrics,
         congestion_cache,
+        congestion_presence_cache,
+        congestion_candidate_precheck_cache,
+        congestion_net_envelope_cache,
+        congestion_grid_net_cache,
+        congestion_grid_span_cache,
         config,
         move_class,
         abs_key,
@@ -126,15 +136,23 @@ def add_node(
     context: AStarContext,
     metrics: AStarMetrics,
     congestion_cache: dict[tuple, int],
+    congestion_presence_cache: dict[tuple[str, int, int, int, int], bool],
+    congestion_candidate_precheck_cache: dict[tuple[str, int, int, int, int], bool],
+    congestion_net_envelope_cache: dict[tuple[str, int, int, int, int], tuple[str, ...]],
+    congestion_grid_net_cache: dict[tuple[str, int, int, int, int], tuple[str, ...]],
+    congestion_grid_span_cache: dict[tuple[str, int, int, int, int], dict[str, tuple[int, ...]]],
     config: SearchRunConfig,
     move_type: MoveKind,
     cache_key: tuple,
 ) -> None:
+    frontier_trace = config.frontier_trace
     metrics.moves_generated += 1
     metrics.total_moves_generated += 1
     state = result.end_port.as_tuple()
     new_lower_bound_g = parent.g_cost + result.length
     if state in closed_set and closed_set[state] <= new_lower_bound_g + TOLERANCE_LINEAR:
+        if frontier_trace is not None:
+            frontier_trace.record("closed_set", move_type, parent.port.as_tuple(), state, result.total_dilated_bounds)
         metrics.pruned_closed_set += 1
         metrics.total_pruned_closed_set += 1
         return
@@ -143,6 +161,8 @@ def add_node(
     end_p = result.end_port
 
     if cache_key in context.hard_collision_set:
+        if frontier_trace is not None:
+            frontier_trace.record("hard_collision", move_type, parent.port.as_tuple(), state, result.total_dilated_bounds)
         context.metrics.total_hard_collision_cache_hits += 1
         metrics.pruned_hard_collision += 1
         metrics.total_pruned_hard_collision += 1
@@ -159,41 +179,100 @@ def add_node(
             collision_found = ce.check_move_static(result, start_port=parent_p, end_port=end_p)
         if collision_found:
             context.hard_collision_set.add(cache_key)
+            if frontier_trace is not None:
+                frontier_trace.record("hard_collision", move_type, parent.port.as_tuple(), state, result.total_dilated_bounds)
             metrics.pruned_hard_collision += 1
             metrics.total_pruned_hard_collision += 1
             return
         context.static_safe_cache.add(cache_key)
 
-    total_overlaps = 0
-    if not config.skip_congestion:
-        if cache_key in congestion_cache:
-            context.metrics.total_congestion_cache_hits += 1
-            total_overlaps = congestion_cache[cache_key]
-        else:
-            context.metrics.total_congestion_cache_misses += 1
-            total_overlaps = context.cost_evaluator.collision_engine.check_move_congestion(result, net_id)
-            congestion_cache[cache_key] = total_overlaps
-
     if config.self_collision_check and component_hits_ancestor_chain(result, parent):
+        if frontier_trace is not None:
+            frontier_trace.record("self_collision", move_type, parent.port.as_tuple(), state, result.total_dilated_bounds)
         return
 
     move_cost = context.cost_evaluator.score_component(
         result,
         start_port=parent_p,
     )
-    move_cost += total_overlaps * context.congestion_penalty
+    next_seed_index = None
+    if (
+        config.guidance_seed is not None
+        and parent.seed_index is not None
+        and parent.seed_index < len(config.guidance_seed)
+        and result.move_spec == config.guidance_seed[parent.seed_index]
+    ):
+        context.metrics.total_guidance_match_moves += 1
+        if result.move_type == "straight":
+            context.metrics.total_guidance_match_moves_straight += 1
+            applied_bonus = config.guidance_bonus
+            context.metrics.total_guidance_bonus_applied_straight += applied_bonus
+        elif result.move_type == "bend90":
+            context.metrics.total_guidance_match_moves_bend90 += 1
+            applied_bonus = config.guidance_bonus
+            context.metrics.total_guidance_bonus_applied_bend90 += applied_bonus
+        else:
+            context.metrics.total_guidance_match_moves_sbend += 1
+            applied_bonus = config.guidance_bonus
+            context.metrics.total_guidance_bonus_applied_sbend += applied_bonus
+        context.metrics.total_guidance_bonus_applied += applied_bonus
+        move_cost = max(0.001, move_cost - applied_bonus)
+        next_seed_index = parent.seed_index + 1
 
     if config.max_cost is not None and parent.g_cost + move_cost > config.max_cost:
+        if frontier_trace is not None:
+            frontier_trace.record("cost", move_type, parent.port.as_tuple(), state, result.total_dilated_bounds)
         metrics.pruned_cost += 1
         metrics.total_pruned_cost += 1
         return
     if move_cost > 1e12:
+        if frontier_trace is not None:
+            frontier_trace.record("cost", move_type, parent.port.as_tuple(), state, result.total_dilated_bounds)
         metrics.pruned_cost += 1
         metrics.total_pruned_cost += 1
         return
 
+    if state in closed_set and closed_set[state] <= parent.g_cost + move_cost + TOLERANCE_LINEAR:
+        if frontier_trace is not None:
+            frontier_trace.record("closed_set", move_type, parent.port.as_tuple(), state, result.total_dilated_bounds)
+        metrics.pruned_closed_set += 1
+        metrics.total_pruned_closed_set += 1
+        return
+
+    total_overlaps = 0
+    if not config.skip_congestion and context.cost_evaluator.collision_engine.has_dynamic_paths():
+        ce = context.cost_evaluator.collision_engine
+        if ce.has_possible_move_congestion(result, net_id, congestion_presence_cache):
+            if ce.has_candidate_move_congestion(
+                result,
+                net_id,
+                congestion_candidate_precheck_cache,
+                congestion_net_envelope_cache,
+                congestion_grid_net_cache,
+            ):
+                if cache_key in congestion_cache:
+                    context.metrics.total_congestion_cache_hits += 1
+                    total_overlaps = congestion_cache[cache_key]
+                else:
+                    context.metrics.total_congestion_cache_misses += 1
+                    total_overlaps = ce.check_move_congestion(
+                        result,
+                        net_id,
+                        net_envelope_cache=congestion_net_envelope_cache,
+                        grid_net_cache=congestion_grid_net_cache,
+                        broad_phase_cache=congestion_grid_span_cache,
+                    )
+                    congestion_cache[cache_key] = total_overlaps
+            else:
+                context.metrics.total_congestion_candidate_precheck_skips += 1
+        else:
+            context.metrics.total_congestion_presence_skips += 1
+    move_cost += total_overlaps * context.congestion_penalty
+
     g_cost = parent.g_cost + move_cost
     if state in closed_set and closed_set[state] <= g_cost + TOLERANCE_LINEAR:
+        if frontier_trace is not None:
+            frontier_trace.record("closed_set", move_type, parent.port.as_tuple(), state, result.total_dilated_bounds)
         metrics.pruned_closed_set += 1
         metrics.total_pruned_closed_set += 1
         return
@@ -203,6 +282,16 @@ def add_node(
         target,
         min_bend_radius=context.min_bend_radius,
     )
-    heapq.heappush(open_set, AStarNode(result.end_port, g_cost, h_cost, parent, result))
+    heapq.heappush(
+        open_set,
+        AStarNode(
+            result.end_port,
+            g_cost,
+            h_cost,
+            parent,
+            result,
+            seed_index=next_seed_index,
+        ),
+    )
     metrics.moves_added += 1
     metrics.total_moves_added += 1

inire/router/_astar_moves.py

@@ -18,6 +18,27 @@ def _quantized_lengths(values: list[float], max_reach: float) -> list[int]:
     return sorted((v for v in out if v > 0), reverse=True)
 
 
+def _distance_to_bounds_in_heading(
+    current: Port,
+    bounds: tuple[float, float, float, float],
+) -> float:
+    min_x, min_y, max_x, max_y = bounds
+    if current.r == 0:
+        return max(0.0, max_x - current.x)
+    if current.r == 90:
+        return max(0.0, max_y - current.y)
+    if current.r == 180:
+        return max(0.0, current.x - min_x)
+    return max(0.0, current.y - min_y)
+
+
+def _should_cap_straights_to_bounds(context: AStarContext) -> bool:
+    return (
+        not context.options.congestion.warm_start_enabled
+        and len(context.problem.nets) >= 8
+    )
+
+
 def _sbend_forward_span(offset: float, radius: float) -> float | None:
     abs_offset = abs(offset)
     if abs_offset <= TOLERANCE_LINEAR or radius <= 0 or abs_offset >= 2.0 * radius:
@@ -63,19 +84,25 @@ def _visible_straight_candidates(
         return []
 
     visibility_manager = context.visibility_manager
-    visibility_manager._ensure_current()
+    visibility_manager.ensure_corner_index_current()
+    context.metrics.total_visibility_tangent_candidate_scans += 1
     max_bend_radius = max(search_options.bend_radii, default=0.0)
     if max_bend_radius <= 0 or not visibility_manager.corners:
         return []
 
     reach = max_reach + max_bend_radius
-    bounds = (current.x - reach, current.y - reach, current.x + reach, current.y + reach)
-    candidate_ids = list(visibility_manager.corner_index.intersection(bounds))
+    candidate_ids = visibility_manager.get_tangent_corner_candidates(
+        current,
+        min_forward=search_options.min_straight_length,
+        max_forward=reach,
+        radii=search_options.bend_radii,
+    )
     if not candidate_ids:
         return []
 
     scored: list[tuple[float, float, float, float, float]] = []
     for idx in candidate_ids:
+        context.metrics.total_visibility_tangent_candidate_corner_checks += 1
         cx, cy = visibility_manager.corners[idx]
         dx = cx - current.x
         dy = cy - current.y
@@ -101,8 +128,15 @@ def _visible_straight_candidates(
     collision_engine = context.cost_evaluator.collision_engine
     tangent_candidates: set[int] = set()
     for _, dist, length, dx, dy in sorted(scored)[:4]:
+        context.metrics.total_visibility_tangent_candidate_ray_tests += 1
         angle = math.degrees(math.atan2(dy, dx))
-        corner_reach = collision_engine.ray_cast(current, angle, max_dist=dist + 0.05, net_width=net_width)
+        corner_reach = collision_engine.ray_cast(
+            current,
+            angle,
+            max_dist=dist + 0.05,
+            net_width=net_width,
+            caller="visibility_tangent",
+        )
         if corner_reach < dist - 0.01:
             continue
         qlen = int(round(length))
@@ -132,6 +166,11 @@ def expand_moves(
     context: AStarContext,
     metrics: AStarMetrics,
     congestion_cache: dict[tuple, int],
+    congestion_presence_cache: dict[tuple[str, int, int, int, int], bool],
+    congestion_candidate_precheck_cache: dict[tuple[str, int, int, int, int], bool],
+    congestion_net_envelope_cache: dict[tuple[str, int, int, int, int], tuple[str, ...]],
+    congestion_grid_net_cache: dict[tuple[str, int, int, int, int], tuple[str, ...]],
+    congestion_grid_span_cache: dict[tuple[str, int, int, int, int], dict[str, tuple[int, ...]]],
     config: SearchRunConfig,
 ) -> None:
     search_options = context.options.search
@@ -156,7 +195,13 @@ def expand_moves(
     dy_local = perp_t
 
     if proj_t > 0 and abs(perp_t) < 1e-6 and cp.r == target.r:
-        max_reach = context.cost_evaluator.collision_engine.ray_cast(cp, cp.r, proj_t + 1.0, net_width=net_width)
+        max_reach = context.cost_evaluator.collision_engine.ray_cast(
+            cp,
+            cp.r,
+            proj_t + 1.0,
+            net_width=net_width,
+            caller="expand_snap",
+        )
         if max_reach >= proj_t - 0.01 and (
             prev_straight_length is None or proj_t < prev_straight_length - TOLERANCE_LINEAR
         ):
@@ -170,12 +215,25 @@ def expand_moves(
                 context,
                 metrics,
                 congestion_cache,
+                congestion_presence_cache,
+                congestion_candidate_precheck_cache,
+                congestion_net_envelope_cache,
+                congestion_grid_net_cache,
+                congestion_grid_span_cache,
                 config,
                 "straight",
                 (int(round(proj_t)),),
             )
 
-    max_reach = context.cost_evaluator.collision_engine.ray_cast(cp, cp.r, search_options.max_straight_length, net_width=net_width)
+    max_reach = context.cost_evaluator.collision_engine.ray_cast(
+        cp,
+        cp.r,
+        search_options.max_straight_length,
+        net_width=net_width,
+        caller="expand_forward",
+    )
+    if _should_cap_straights_to_bounds(context):
+        max_reach = min(max_reach, _distance_to_bounds_in_heading(cp, context.problem.bounds))
     candidate_lengths = [
         search_options.min_straight_length,
         max_reach,
@@ -221,6 +279,11 @@ def expand_moves(
             context,
             metrics,
             congestion_cache,
+            congestion_presence_cache,
+            congestion_candidate_precheck_cache,
+            congestion_net_envelope_cache,
+            congestion_grid_net_cache,
+            congestion_grid_span_cache,
             config,
             "straight",
             (length,),
@@ -249,6 +312,11 @@ def expand_moves(
                 context,
                 metrics,
                 congestion_cache,
+                congestion_presence_cache,
+                congestion_candidate_precheck_cache,
+                congestion_net_envelope_cache,
+                congestion_grid_net_cache,
+                congestion_grid_span_cache,
                 config,
                 "bend90",
                 (radius, direction),
@@ -283,6 +351,11 @@ def expand_moves(
                 context,
                 metrics,
                 congestion_cache,
+                congestion_presence_cache,
+                congestion_candidate_precheck_cache,
+                congestion_net_envelope_cache,
+                congestion_grid_net_cache,
+                congestion_grid_span_cache,
                 config,
                 "sbend",
                 (offset, radius),

inire/router/_astar_types.py

@@ -1,6 +1,6 @@
 from __future__ import annotations
 
-from dataclasses import dataclass
+from dataclasses import dataclass, field
 from typing import TYPE_CHECKING
 
 from inire.model import resolve_bend_geometry
@@ -12,6 +12,51 @@ if TYPE_CHECKING:
     from inire.geometry.primitives import Port
     from inire.model import RoutingOptions, RoutingProblem
     from inire.router.cost import CostEvaluator
+    from inire.seeds import PathSegmentSeed
+
+
+@dataclass(slots=True)
+class FrontierTraceCollector:
+    hotspot_bounds: tuple[tuple[float, float, float, float], ...]
+    sample_limit: int = 64
+    pruned_closed_set: int = 0
+    pruned_hard_collision: int = 0
+    pruned_self_collision: int = 0
+    pruned_cost: int = 0
+    samples: list[tuple[str, str, int, tuple[int, int, int], tuple[int, int, int]]] = field(default_factory=list)
+
+    def _matching_hotspot_index(self, bounds: tuple[float, float, float, float]) -> int | None:
+        for idx, hotspot_bounds in enumerate(self.hotspot_bounds):
+            if (
+                bounds[0] < hotspot_bounds[2]
+                and bounds[2] > hotspot_bounds[0]
+                and bounds[1] < hotspot_bounds[3]
+                and bounds[3] > hotspot_bounds[1]
+            ):
+                return idx
+        return None
+
+    def record(
+        self,
+        reason: str,
+        move_type: str,
+        parent_state: tuple[int, int, int],
+        end_state: tuple[int, int, int],
+        bounds: tuple[float, float, float, float],
+    ) -> None:
+        hotspot_index = self._matching_hotspot_index(bounds)
+        if hotspot_index is None:
+            return
+        if reason == "closed_set":
+            self.pruned_closed_set += 1
+        elif reason == "hard_collision":
+            self.pruned_hard_collision += 1
+        elif reason == "self_collision":
+            self.pruned_self_collision += 1
+        else:
+            self.pruned_cost += 1
+        if len(self.samples) < self.sample_limit:
+            self.samples.append((reason, move_type, hotspot_index, parent_state, end_state))
 
 
 @dataclass(frozen=True, slots=True)
@@ -20,6 +65,9 @@ class SearchRunConfig:
     bend_physical_geometry: BendPhysicalGeometry
     bend_clip_margin: float | None
     node_limit: int
+    guidance_seed: tuple[PathSegmentSeed, ...] | None = None
+    guidance_bonus: float = 0.0
+    frontier_trace: FrontierTraceCollector | None = None
     return_partial: bool = False
     store_expanded: bool = False
     skip_congestion: bool = False
@@ -33,6 +81,9 @@ class SearchRunConfig:
         *,
         bend_collision_type: BendCollisionModel | None = None,
         node_limit: int | None = None,
+        guidance_seed: tuple[PathSegmentSeed, ...] | None = None,
+        guidance_bonus: float = 0.0,
+        frontier_trace: FrontierTraceCollector | None = None,
         return_partial: bool = False,
         store_expanded: bool = False,
         skip_congestion: bool = False,
@@ -49,6 +100,9 @@ class SearchRunConfig:
             bend_physical_geometry=bend_physical_geometry,
             bend_clip_margin=search.bend_clip_margin,
             node_limit=search.node_limit if node_limit is None else node_limit,
+            guidance_seed=guidance_seed,
+            guidance_bonus=float(guidance_bonus),
+            frontier_trace=frontier_trace,
             return_partial=return_partial,
             store_expanded=store_expanded,
             skip_congestion=skip_congestion,
@@ -58,7 +112,18 @@ class SearchRunConfig:
 
 
 class AStarNode:
-    __slots__ = ("port", "g_cost", "h_cost", "fh_cost", "parent", "component_result")
+    __slots__ = (
+        "port",
+        "g_cost",
+        "h_cost",
+        "fh_cost",
+        "parent",
+        "component_result",
+        "base_move_cost",
+        "cache_key",
+        "seed_index",
+        "congestion_resolved",
+    )
 
     def __init__(
         self,
@@ -67,6 +132,11 @@ class AStarNode:
         h_cost: float,
         parent: AStarNode | None = None,
         component_result: ComponentResult | None = None,
+        *,
+        base_move_cost: float = 0.0,
+        cache_key: tuple | None = None,
+        seed_index: int | None = None,
+        congestion_resolved: bool = True,
     ) -> None:
         self.port = port
         self.g_cost = g_cost
@@ -74,6 +144,10 @@ class AStarNode:
         self.fh_cost = (g_cost + h_cost, h_cost)
         self.parent = parent
         self.component_result = component_result
+        self.base_move_cost = base_move_cost
+        self.cache_key = cache_key
+        self.seed_index = seed_index
+        self.congestion_resolved = congestion_resolved
 
     def __lt__(self, other: AStarNode) -> bool:
         return self.fh_cost < other.fh_cost
@@ -94,14 +168,48 @@ class AStarMetrics:
         "total_warm_start_paths_used",
         "total_refine_path_calls",
         "total_timeout_events",
+        "total_iteration_reverify_calls",
+        "total_iteration_reverified_nets",
+        "total_iteration_conflicting_nets",
+        "total_iteration_conflict_edges",
+        "total_nets_carried_forward",
+        "total_score_component_calls",
+        "total_score_component_total_ns",
+        "total_path_cost_calls",
+        "total_danger_map_lookup_calls",
+        "total_danger_map_cache_hits",
+        "total_danger_map_cache_misses",
+        "total_danger_map_query_calls",
+        "total_danger_map_total_ns",
         "total_move_cache_abs_hits",
         "total_move_cache_abs_misses",
         "total_move_cache_rel_hits",
         "total_move_cache_rel_misses",
+        "total_guidance_match_moves",
+        "total_guidance_match_moves_straight",
+        "total_guidance_match_moves_bend90",
+        "total_guidance_match_moves_sbend",
+        "total_guidance_bonus_applied",
+        "total_guidance_bonus_applied_straight",
+        "total_guidance_bonus_applied_bend90",
+        "total_guidance_bonus_applied_sbend",
         "total_static_safe_cache_hits",
         "total_hard_collision_cache_hits",
         "total_congestion_cache_hits",
         "total_congestion_cache_misses",
+        "total_congestion_presence_cache_hits",
+        "total_congestion_presence_cache_misses",
+        "total_congestion_presence_skips",
+        "total_congestion_candidate_precheck_hits",
+        "total_congestion_candidate_precheck_misses",
+        "total_congestion_candidate_precheck_skips",
+        "total_congestion_grid_net_cache_hits",
+        "total_congestion_grid_net_cache_misses",
+        "total_congestion_grid_span_cache_hits",
+        "total_congestion_grid_span_cache_misses",
+        "total_congestion_candidate_nets",
+        "total_congestion_net_envelope_cache_hits",
+        "total_congestion_net_envelope_cache_misses",
         "total_dynamic_path_objects_added",
         "total_dynamic_path_objects_removed",
         "total_dynamic_tree_rebuilds",
@@ -109,21 +217,47 @@ class AStarMetrics:
         "total_static_tree_rebuilds",
         "total_static_raw_tree_rebuilds",
         "total_static_net_tree_rebuilds",
+        "total_visibility_corner_index_builds",
         "total_visibility_builds",
         "total_visibility_corner_pairs_checked",
-        "total_visibility_corner_queries",
-        "total_visibility_corner_hits",
+        "total_visibility_corner_queries_exact",
+        "total_visibility_corner_hits_exact",
         "total_visibility_point_queries",
         "total_visibility_point_cache_hits",
         "total_visibility_point_cache_misses",
+        "total_visibility_tangent_candidate_scans",
+        "total_visibility_tangent_candidate_corner_checks",
+        "total_visibility_tangent_candidate_ray_tests",
         "total_ray_cast_calls",
+        "total_ray_cast_calls_straight_static",
+        "total_ray_cast_calls_expand_snap",
+        "total_ray_cast_calls_expand_forward",
+        "total_ray_cast_calls_visibility_build",
+        "total_ray_cast_calls_visibility_query",
+        "total_ray_cast_calls_visibility_tangent",
+        "total_ray_cast_calls_other",
         "total_ray_cast_candidate_bounds",
         "total_ray_cast_exact_geometry_checks",
         "total_congestion_check_calls",
+        "total_congestion_lazy_resolutions",
+        "total_congestion_lazy_requeues",
+        "total_congestion_candidate_ids",
         "total_congestion_exact_pair_checks",
         "total_verify_path_report_calls",
         "total_verify_static_buffer_ops",
+        "total_verify_dynamic_candidate_nets",
         "total_verify_dynamic_exact_pair_checks",
+        "total_refinement_windows_considered",
+        "total_refinement_static_bounds_checked",
+        "total_refinement_dynamic_bounds_checked",
+        "total_refinement_candidate_side_extents",
+        "total_refinement_candidates_built",
+        "total_refinement_candidates_verified",
+        "total_refinement_candidates_accepted",
+        "total_pair_local_search_pairs_considered",
+        "total_pair_local_search_attempts",
+        "total_pair_local_search_accepts",
+        "total_pair_local_search_nodes_expanded",
         "last_expanded_nodes",
         "nodes_expanded",
         "moves_generated",
@@ -147,14 +281,48 @@ class AStarMetrics:
         self.total_warm_start_paths_used = 0
         self.total_refine_path_calls = 0
         self.total_timeout_events = 0
+        self.total_iteration_reverify_calls = 0
+        self.total_iteration_reverified_nets = 0
+        self.total_iteration_conflicting_nets = 0
+        self.total_iteration_conflict_edges = 0
+        self.total_nets_carried_forward = 0
+        self.total_score_component_calls = 0
+        self.total_score_component_total_ns = 0
+        self.total_path_cost_calls = 0
+        self.total_danger_map_lookup_calls = 0
+        self.total_danger_map_cache_hits = 0
+        self.total_danger_map_cache_misses = 0
+        self.total_danger_map_query_calls = 0
+        self.total_danger_map_total_ns = 0
         self.total_move_cache_abs_hits = 0
         self.total_move_cache_abs_misses = 0
         self.total_move_cache_rel_hits = 0
         self.total_move_cache_rel_misses = 0
+        self.total_guidance_match_moves = 0
+        self.total_guidance_match_moves_straight = 0
+        self.total_guidance_match_moves_bend90 = 0
+        self.total_guidance_match_moves_sbend = 0
+        self.total_guidance_bonus_applied = 0.0
+        self.total_guidance_bonus_applied_straight = 0.0
+        self.total_guidance_bonus_applied_bend90 = 0.0
+        self.total_guidance_bonus_applied_sbend = 0.0
         self.total_static_safe_cache_hits = 0
         self.total_hard_collision_cache_hits = 0
         self.total_congestion_cache_hits = 0
         self.total_congestion_cache_misses = 0
+        self.total_congestion_presence_cache_hits = 0
+        self.total_congestion_presence_cache_misses = 0
+        self.total_congestion_presence_skips = 0
+        self.total_congestion_candidate_precheck_hits = 0
+        self.total_congestion_candidate_precheck_misses = 0
+        self.total_congestion_candidate_precheck_skips = 0
+        self.total_congestion_grid_net_cache_hits = 0
+        self.total_congestion_grid_net_cache_misses = 0
+        self.total_congestion_grid_span_cache_hits = 0
+        self.total_congestion_grid_span_cache_misses = 0
+        self.total_congestion_candidate_nets = 0
+        self.total_congestion_net_envelope_cache_hits = 0
+        self.total_congestion_net_envelope_cache_misses = 0
         self.total_dynamic_path_objects_added = 0
         self.total_dynamic_path_objects_removed = 0
         self.total_dynamic_tree_rebuilds = 0
@@ -162,21 +330,47 @@ class AStarMetrics:
         self.total_static_tree_rebuilds = 0
         self.total_static_raw_tree_rebuilds = 0
         self.total_static_net_tree_rebuilds = 0
+        self.total_visibility_corner_index_builds = 0
         self.total_visibility_builds = 0
         self.total_visibility_corner_pairs_checked = 0
-        self.total_visibility_corner_queries = 0
-        self.total_visibility_corner_hits = 0
+        self.total_visibility_corner_queries_exact = 0
+        self.total_visibility_corner_hits_exact = 0
         self.total_visibility_point_queries = 0
         self.total_visibility_point_cache_hits = 0
         self.total_visibility_point_cache_misses = 0
+        self.total_visibility_tangent_candidate_scans = 0
+        self.total_visibility_tangent_candidate_corner_checks = 0
+        self.total_visibility_tangent_candidate_ray_tests = 0
         self.total_ray_cast_calls = 0
+        self.total_ray_cast_calls_straight_static = 0
+        self.total_ray_cast_calls_expand_snap = 0
+        self.total_ray_cast_calls_expand_forward = 0
+        self.total_ray_cast_calls_visibility_build = 0
+        self.total_ray_cast_calls_visibility_query = 0
+        self.total_ray_cast_calls_visibility_tangent = 0
+        self.total_ray_cast_calls_other = 0
         self.total_ray_cast_candidate_bounds = 0
         self.total_ray_cast_exact_geometry_checks = 0
         self.total_congestion_check_calls = 0
+        self.total_congestion_lazy_resolutions = 0
+        self.total_congestion_lazy_requeues = 0
+        self.total_congestion_candidate_ids = 0
         self.total_congestion_exact_pair_checks = 0
         self.total_verify_path_report_calls = 0
         self.total_verify_static_buffer_ops = 0
+        self.total_verify_dynamic_candidate_nets = 0
         self.total_verify_dynamic_exact_pair_checks = 0
+        self.total_refinement_windows_considered = 0
+        self.total_refinement_static_bounds_checked = 0
+        self.total_refinement_dynamic_bounds_checked = 0
+        self.total_refinement_candidate_side_extents = 0
+        self.total_refinement_candidates_built = 0
+        self.total_refinement_candidates_verified = 0
+        self.total_refinement_candidates_accepted = 0
+        self.total_pair_local_search_pairs_considered = 0
+        self.total_pair_local_search_attempts = 0
+        self.total_pair_local_search_accepts = 0
+        self.total_pair_local_search_nodes_expanded = 0
         self.last_expanded_nodes: list[tuple[int, int, int]] = []
         self.nodes_expanded = 0
         self.moves_generated = 0
@@ -199,14 +393,48 @@ class AStarMetrics:
         self.total_warm_start_paths_used = 0
         self.total_refine_path_calls = 0
         self.total_timeout_events = 0
+        self.total_iteration_reverify_calls = 0
+        self.total_iteration_reverified_nets = 0
+        self.total_iteration_conflicting_nets = 0
+        self.total_iteration_conflict_edges = 0
+        self.total_nets_carried_forward = 0
+        self.total_score_component_calls = 0
+        self.total_score_component_total_ns = 0
+        self.total_path_cost_calls = 0
+        self.total_danger_map_lookup_calls = 0
+        self.total_danger_map_cache_hits = 0
+        self.total_danger_map_cache_misses = 0
+        self.total_danger_map_query_calls = 0
+        self.total_danger_map_total_ns = 0
         self.total_move_cache_abs_hits = 0
         self.total_move_cache_abs_misses = 0
         self.total_move_cache_rel_hits = 0
         self.total_move_cache_rel_misses = 0
+        self.total_guidance_match_moves = 0
+        self.total_guidance_match_moves_straight = 0
+        self.total_guidance_match_moves_bend90 = 0
+        self.total_guidance_match_moves_sbend = 0
+        self.total_guidance_bonus_applied = 0.0
+        self.total_guidance_bonus_applied_straight = 0.0
+        self.total_guidance_bonus_applied_bend90 = 0.0
+        self.total_guidance_bonus_applied_sbend = 0.0
         self.total_static_safe_cache_hits = 0
         self.total_hard_collision_cache_hits = 0
         self.total_congestion_cache_hits = 0
         self.total_congestion_cache_misses = 0
+        self.total_congestion_presence_cache_hits = 0
+        self.total_congestion_presence_cache_misses = 0
+        self.total_congestion_presence_skips = 0
+        self.total_congestion_candidate_precheck_hits = 0
+        self.total_congestion_candidate_precheck_misses = 0
+        self.total_congestion_candidate_precheck_skips = 0
+        self.total_congestion_grid_net_cache_hits = 0
+        self.total_congestion_grid_net_cache_misses = 0
+        self.total_congestion_grid_span_cache_hits = 0
+        self.total_congestion_grid_span_cache_misses = 0
+        self.total_congestion_candidate_nets = 0
+        self.total_congestion_net_envelope_cache_hits = 0
+        self.total_congestion_net_envelope_cache_misses = 0
         self.total_dynamic_path_objects_added = 0
         self.total_dynamic_path_objects_removed = 0
         self.total_dynamic_tree_rebuilds = 0
@@ -214,21 +442,47 @@ class AStarMetrics:
         self.total_static_tree_rebuilds = 0
         self.total_static_raw_tree_rebuilds = 0
         self.total_static_net_tree_rebuilds = 0
+        self.total_visibility_corner_index_builds = 0
         self.total_visibility_builds = 0
         self.total_visibility_corner_pairs_checked = 0
-        self.total_visibility_corner_queries = 0
-        self.total_visibility_corner_hits = 0
+        self.total_visibility_corner_queries_exact = 0
+        self.total_visibility_corner_hits_exact = 0
         self.total_visibility_point_queries = 0
         self.total_visibility_point_cache_hits = 0
         self.total_visibility_point_cache_misses = 0
+        self.total_visibility_tangent_candidate_scans = 0
+        self.total_visibility_tangent_candidate_corner_checks = 0
+        self.total_visibility_tangent_candidate_ray_tests = 0
         self.total_ray_cast_calls = 0
+        self.total_ray_cast_calls_straight_static = 0
+        self.total_ray_cast_calls_expand_snap = 0
+        self.total_ray_cast_calls_expand_forward = 0
+        self.total_ray_cast_calls_visibility_build = 0
+        self.total_ray_cast_calls_visibility_query = 0
+        self.total_ray_cast_calls_visibility_tangent = 0
+        self.total_ray_cast_calls_other = 0
         self.total_ray_cast_candidate_bounds = 0
         self.total_ray_cast_exact_geometry_checks = 0
         self.total_congestion_check_calls = 0
+        self.total_congestion_lazy_resolutions = 0
+        self.total_congestion_lazy_requeues = 0
+        self.total_congestion_candidate_ids = 0
         self.total_congestion_exact_pair_checks = 0
         self.total_verify_path_report_calls = 0
         self.total_verify_static_buffer_ops = 0
+        self.total_verify_dynamic_candidate_nets = 0
         self.total_verify_dynamic_exact_pair_checks = 0
+        self.total_refinement_windows_considered = 0
+        self.total_refinement_static_bounds_checked = 0
+        self.total_refinement_dynamic_bounds_checked = 0
+        self.total_refinement_candidate_side_extents = 0
+        self.total_refinement_candidates_built = 0
+        self.total_refinement_candidates_verified = 0
+        self.total_refinement_candidates_accepted = 0
+        self.total_pair_local_search_pairs_considered = 0
+        self.total_pair_local_search_attempts = 0
+        self.total_pair_local_search_accepts = 0
+        self.total_pair_local_search_nodes_expanded = 0
 
     def reset_per_route(self) -> None:
         self.nodes_expanded = 0
@@ -254,14 +508,48 @@ class AStarMetrics:
             warm_start_paths_used=self.total_warm_start_paths_used,
             refine_path_calls=self.total_refine_path_calls,
             timeout_events=self.total_timeout_events,
+            iteration_reverify_calls=self.total_iteration_reverify_calls,
+            iteration_reverified_nets=self.total_iteration_reverified_nets,
+            iteration_conflicting_nets=self.total_iteration_conflicting_nets,
+            iteration_conflict_edges=self.total_iteration_conflict_edges,
+            nets_carried_forward=self.total_nets_carried_forward,
+            score_component_calls=self.total_score_component_calls,
+            score_component_total_ns=self.total_score_component_total_ns,
+            path_cost_calls=self.total_path_cost_calls,
+            danger_map_lookup_calls=self.total_danger_map_lookup_calls,
+            danger_map_cache_hits=self.total_danger_map_cache_hits,
+            danger_map_cache_misses=self.total_danger_map_cache_misses,
+            danger_map_query_calls=self.total_danger_map_query_calls,
+            danger_map_total_ns=self.total_danger_map_total_ns,
             move_cache_abs_hits=self.total_move_cache_abs_hits,
             move_cache_abs_misses=self.total_move_cache_abs_misses,
             move_cache_rel_hits=self.total_move_cache_rel_hits,
             move_cache_rel_misses=self.total_move_cache_rel_misses,
+            guidance_match_moves=self.total_guidance_match_moves,
+            guidance_match_moves_straight=self.total_guidance_match_moves_straight,
+            guidance_match_moves_bend90=self.total_guidance_match_moves_bend90,
+            guidance_match_moves_sbend=self.total_guidance_match_moves_sbend,
+            guidance_bonus_applied=self.total_guidance_bonus_applied,
+            guidance_bonus_applied_straight=self.total_guidance_bonus_applied_straight,
+            guidance_bonus_applied_bend90=self.total_guidance_bonus_applied_bend90,
+            guidance_bonus_applied_sbend=self.total_guidance_bonus_applied_sbend,
             static_safe_cache_hits=self.total_static_safe_cache_hits,
             hard_collision_cache_hits=self.total_hard_collision_cache_hits,
             congestion_cache_hits=self.total_congestion_cache_hits,
             congestion_cache_misses=self.total_congestion_cache_misses,
+            congestion_presence_cache_hits=self.total_congestion_presence_cache_hits,
+            congestion_presence_cache_misses=self.total_congestion_presence_cache_misses,
+            congestion_presence_skips=self.total_congestion_presence_skips,
+            congestion_candidate_precheck_hits=self.total_congestion_candidate_precheck_hits,
+            congestion_candidate_precheck_misses=self.total_congestion_candidate_precheck_misses,
+            congestion_candidate_precheck_skips=self.total_congestion_candidate_precheck_skips,
+            congestion_grid_net_cache_hits=self.total_congestion_grid_net_cache_hits,
+            congestion_grid_net_cache_misses=self.total_congestion_grid_net_cache_misses,
+            congestion_grid_span_cache_hits=self.total_congestion_grid_span_cache_hits,
+            congestion_grid_span_cache_misses=self.total_congestion_grid_span_cache_misses,
+            congestion_candidate_nets=self.total_congestion_candidate_nets,
+            congestion_net_envelope_cache_hits=self.total_congestion_net_envelope_cache_hits,
+            congestion_net_envelope_cache_misses=self.total_congestion_net_envelope_cache_misses,
             dynamic_path_objects_added=self.total_dynamic_path_objects_added,
             dynamic_path_objects_removed=self.total_dynamic_path_objects_removed,
             dynamic_tree_rebuilds=self.total_dynamic_tree_rebuilds,
@@ -269,21 +557,47 @@ class AStarMetrics:
             static_tree_rebuilds=self.total_static_tree_rebuilds,
             static_raw_tree_rebuilds=self.total_static_raw_tree_rebuilds,
             static_net_tree_rebuilds=self.total_static_net_tree_rebuilds,
+            visibility_corner_index_builds=self.total_visibility_corner_index_builds,
             visibility_builds=self.total_visibility_builds,
             visibility_corner_pairs_checked=self.total_visibility_corner_pairs_checked,
-            visibility_corner_queries=self.total_visibility_corner_queries,
-            visibility_corner_hits=self.total_visibility_corner_hits,
+            visibility_corner_queries_exact=self.total_visibility_corner_queries_exact,
+            visibility_corner_hits_exact=self.total_visibility_corner_hits_exact,
             visibility_point_queries=self.total_visibility_point_queries,
             visibility_point_cache_hits=self.total_visibility_point_cache_hits,
             visibility_point_cache_misses=self.total_visibility_point_cache_misses,
+            visibility_tangent_candidate_scans=self.total_visibility_tangent_candidate_scans,
+            visibility_tangent_candidate_corner_checks=self.total_visibility_tangent_candidate_corner_checks,
+            visibility_tangent_candidate_ray_tests=self.total_visibility_tangent_candidate_ray_tests,
             ray_cast_calls=self.total_ray_cast_calls,
+            ray_cast_calls_straight_static=self.total_ray_cast_calls_straight_static,
+            ray_cast_calls_expand_snap=self.total_ray_cast_calls_expand_snap,
+            ray_cast_calls_expand_forward=self.total_ray_cast_calls_expand_forward,
+            ray_cast_calls_visibility_build=self.total_ray_cast_calls_visibility_build,
+            ray_cast_calls_visibility_query=self.total_ray_cast_calls_visibility_query,
+            ray_cast_calls_visibility_tangent=self.total_ray_cast_calls_visibility_tangent,
+            ray_cast_calls_other=self.total_ray_cast_calls_other,
             ray_cast_candidate_bounds=self.total_ray_cast_candidate_bounds,
             ray_cast_exact_geometry_checks=self.total_ray_cast_exact_geometry_checks,
             congestion_check_calls=self.total_congestion_check_calls,
+            congestion_lazy_resolutions=self.total_congestion_lazy_resolutions,
+            congestion_lazy_requeues=self.total_congestion_lazy_requeues,
+            congestion_candidate_ids=self.total_congestion_candidate_ids,
             congestion_exact_pair_checks=self.total_congestion_exact_pair_checks,
             verify_path_report_calls=self.total_verify_path_report_calls,
             verify_static_buffer_ops=self.total_verify_static_buffer_ops,
+            verify_dynamic_candidate_nets=self.total_verify_dynamic_candidate_nets,
             verify_dynamic_exact_pair_checks=self.total_verify_dynamic_exact_pair_checks,
+            refinement_windows_considered=self.total_refinement_windows_considered,
+            refinement_static_bounds_checked=self.total_refinement_static_bounds_checked,
+            refinement_dynamic_bounds_checked=self.total_refinement_dynamic_bounds_checked,
+            refinement_candidate_side_extents=self.total_refinement_candidate_side_extents,
+            refinement_candidates_built=self.total_refinement_candidates_built,
+            refinement_candidates_verified=self.total_refinement_candidates_verified,
+            refinement_candidates_accepted=self.total_refinement_candidates_accepted,
+            pair_local_search_pairs_considered=self.total_pair_local_search_pairs_considered,
+            pair_local_search_attempts=self.total_pair_local_search_attempts,
+            pair_local_search_accepts=self.total_pair_local_search_accepts,
+            pair_local_search_nodes_expanded=self.total_pair_local_search_nodes_expanded,
         )
 
 

inire/router/_router.py

@@ -5,11 +5,23 @@ import time
 from dataclasses import dataclass
 from typing import TYPE_CHECKING
 
-from inire.model import NetOrder, NetSpec, resolve_bend_geometry
-from inire.results import RoutingOutcome, RoutingReport, RoutingResult
-from inire.router._astar_types import AStarContext, AStarMetrics, SearchRunConfig
+from inire.geometry.collision import RoutingWorld
+from inire.model import NetOrder, NetSpec, RoutingProblem, resolve_bend_geometry
+from inire.results import (
+    ComponentConflictTrace,
+    ConflictTraceEntry,
+    FrontierPruneSample,
+    NetConflictTrace,
+    NetFrontierTrace,
+    RoutingOutcome,
+    RoutingReport,
+    RoutingResult,
+)
+from inire.router._astar_types import AStarContext, AStarMetrics, FrontierTraceCollector, SearchRunConfig
 from inire.router._search import route_astar
 from inire.router._seed_materialization import materialize_path_seed
+from inire.router.cost import CostEvaluator
+from inire.router.danger_map import DangerMap
 from inire.router.refiner import PathRefiner
 
 if TYPE_CHECKING:
@@ -17,6 +29,7 @@ if TYPE_CHECKING:
 
     from shapely.geometry import Polygon
 
+    from inire.geometry.collision import PathVerificationDetail
     from inire.geometry.components import ComponentResult
 
 
@@ -30,6 +43,27 @@ class _RoutingState:
     timeout_s: float
     initial_paths: dict[str, tuple[ComponentResult, ...]] | None
     accumulated_expanded_nodes: list[tuple[int, int, int]]
+    best_results: dict[str, RoutingResult]
+    best_completed_nets: int
+    best_conflict_edges: int
+    best_dynamic_collisions: int
+    last_conflict_signature: tuple[tuple[str, str], ...]
+    last_conflict_edge_count: int
+    repeated_conflict_count: int
+
+
+@dataclass(slots=True)
+class _IterationReview:
+    conflicting_nets: set[str]
+    conflict_edges: set[tuple[str, str]]
+    completed_net_ids: set[str]
+    total_dynamic_collisions: int
+
+
+@dataclass(frozen=True, slots=True)
+class _PairLocalTarget:
+    net_ids: tuple[str, str]
+
 
 class PathFinder:
     __slots__ = (
@@ -37,6 +71,8 @@ class PathFinder:
         "metrics",
         "refiner",
         "accumulated_expanded_nodes",
+        "conflict_trace",
+        "frontier_trace",
     )
 
     def __init__(
@@ -48,16 +84,27 @@ class PathFinder:
         self.metrics = self.context.metrics if metrics is None else metrics
         self.context.metrics = self.metrics
         self.context.cost_evaluator.collision_engine.metrics = self.metrics
+        if self.context.cost_evaluator.danger_map is not None:
+            self.context.cost_evaluator.danger_map.metrics = self.metrics
         self.refiner = PathRefiner(self.context)
         self.accumulated_expanded_nodes: list[tuple[int, int, int]] = []
+        self.conflict_trace: list[ConflictTraceEntry] = []
+        self.frontier_trace: list[NetFrontierTrace] = []
 
     def _install_path(self, net_id: str, path: Sequence[ComponentResult]) -> None:
         all_geoms: list[Polygon] = []
         all_dilated: list[Polygon] = []
-        for result in path:
+        component_indexes: list[int] = []
+        for component_index, result in enumerate(path):
             all_geoms.extend(result.collision_geometry)
             all_dilated.extend(result.dilated_collision_geometry)
-        self.context.cost_evaluator.collision_engine.add_path(net_id, all_geoms, dilated_geometry=all_dilated)
+            component_indexes.extend([component_index] * len(result.collision_geometry))
+        self.context.cost_evaluator.collision_engine.add_path(
+            net_id,
+            all_geoms,
+            dilated_geometry=all_dilated,
+            component_indexes=component_indexes,
+        )
 
     def _routing_order(
         self,
@@ -134,6 +181,13 @@ class PathFinder:
             timeout_s=max(60.0, 10.0 * num_nets * congestion.max_iterations),
             initial_paths=initial_paths,
             accumulated_expanded_nodes=[],
+            best_results={},
+            best_completed_nets=-1,
+            best_conflict_edges=10**9,
+            best_dynamic_collisions=10**9,
+            last_conflict_signature=(),
+            last_conflict_edge_count=0,
+            repeated_conflict_count=0,
         )
         if state.initial_paths is None and congestion.warm_start_enabled:
             state.initial_paths = self._build_greedy_warm_start_paths(net_specs, congestion.net_order)
@@ -163,6 +217,533 @@ class PathFinder:
         )
         return initial_paths
 
+    def _replace_installed_paths(self, state: _RoutingState, results: dict[str, RoutingResult]) -> None:
+        for net_id in state.ordered_net_ids:
+            self.context.cost_evaluator.collision_engine.remove_path(net_id)
+        for net_id in state.ordered_net_ids:
+            result = results.get(net_id)
+            if result and result.path:
+                self._install_path(net_id, result.path)
+
+    def _update_best_iteration(self, state: _RoutingState, review: _IterationReview) -> bool:
+        completed_nets = len(review.completed_net_ids)
+        conflict_edges = len(review.conflict_edges)
+        dynamic_collisions = review.total_dynamic_collisions
+        is_better = (
+            completed_nets > state.best_completed_nets
+            or (
+                completed_nets == state.best_completed_nets
+                and (
+                    conflict_edges < state.best_conflict_edges
+                    or (
+                        conflict_edges == state.best_conflict_edges
+                        and dynamic_collisions < state.best_dynamic_collisions
+                    )
+                )
+            )
+        )
+        if not is_better:
+            return False
+
+        state.best_results = dict(state.results)
+        state.best_completed_nets = completed_nets
+        state.best_conflict_edges = conflict_edges
+        state.best_dynamic_collisions = dynamic_collisions
+        return True
+
+    def _restore_best_iteration(self, state: _RoutingState) -> None:
+        if not state.best_results:
+            return
+        state.results = dict(state.best_results)
+        self._replace_installed_paths(state, state.results)
+
+    def _capture_conflict_trace_entry(
+        self,
+        state: _RoutingState,
+        *,
+        stage: str,
+        iteration: int | None,
+        results: dict[str, RoutingResult],
+        details_by_net: dict[str, PathVerificationDetail],
+        review: _IterationReview,
+    ) -> None:
+        if not self.context.options.diagnostics.capture_conflict_trace:
+            return
+
+        nets = []
+        for net_id in state.ordered_net_ids:
+            result = results.get(net_id)
+            if result is None:
+                result = RoutingResult(net_id=net_id, path=(), reached_target=False)
+            detail = details_by_net.get(net_id)
+            component_conflicts = ()
+            conflicting_net_ids = ()
+            if detail is not None:
+                conflicting_net_ids = detail.conflicting_net_ids
+                component_conflicts = tuple(
+                    ComponentConflictTrace(
+                        other_net_id=other_net_id,
+                        self_component_index=self_component_index,
+                        other_component_index=other_component_index,
+                    )
+                    for self_component_index, other_net_id, other_component_index in detail.component_conflicts
+                )
+            nets.append(
+                NetConflictTrace(
+                    net_id=net_id,
+                    outcome=result.outcome,
+                    reached_target=result.reached_target,
+                    report=result.report,
+                    conflicting_net_ids=tuple(conflicting_net_ids),
+                    component_conflicts=component_conflicts,
+                )
+            )
+
+        self.conflict_trace.append(
+            ConflictTraceEntry(
+                stage=stage,  # type: ignore[arg-type]
+                iteration=iteration,
+                completed_net_ids=tuple(sorted(review.completed_net_ids)),
+                conflict_edges=tuple(sorted(review.conflict_edges)),
+                nets=tuple(nets),
+            )
+        )
+
+    def _build_frontier_hotspot_bounds(
+        self,
+        state: _RoutingState,
+        net_id: str,
+        details_by_net: dict[str, PathVerificationDetail],
+    ) -> tuple[tuple[float, float, float, float], ...]:
+        result = state.results.get(net_id)
+        detail = details_by_net.get(net_id)
+        if result is None or detail is None or not result.path:
+            return ()
+
+        hotspot_bounds: list[tuple[float, float, float, float]] = []
+        seen: set[tuple[float, float, float, float]] = set()
+        margin = max(5.0, self.context.cost_evaluator.collision_engine.clearance * 2.0)
+
+        for self_component_index, other_net_id, other_component_index in detail.component_conflicts:
+            other_result = state.results.get(other_net_id)
+            if other_result is None or not other_result.path:
+                continue
+            if self_component_index >= len(result.path) or other_component_index >= len(other_result.path):
+                continue
+            left_component = result.path[self_component_index]
+            right_component = other_result.path[other_component_index]
+            overlap_found = False
+            for left_poly in left_component.dilated_physical_geometry:
+                for right_poly in right_component.dilated_physical_geometry:
+                    if not left_poly.intersects(right_poly) or left_poly.touches(right_poly):
+                        continue
+                    overlap = left_poly.intersection(right_poly)
+                    if overlap.is_empty:
+                        continue
+                    buffered = overlap.buffer(margin, join_style="mitre").bounds
+                    if buffered not in seen:
+                        seen.add(buffered)
+                        hotspot_bounds.append(buffered)
+                    overlap_found = True
+            if overlap_found:
+                continue
+
+            left_bounds = left_component.total_dilated_bounds
+            right_bounds = right_component.total_dilated_bounds
+            if (
+                left_bounds[0] < right_bounds[2]
+                and left_bounds[2] > right_bounds[0]
+                and left_bounds[1] < right_bounds[3]
+                and left_bounds[3] > right_bounds[1]
+            ):
+                buffered = (
+                    max(left_bounds[0], right_bounds[0]) - margin,
+                    max(left_bounds[1], right_bounds[1]) - margin,
+                    min(left_bounds[2], right_bounds[2]) + margin,
+                    min(left_bounds[3], right_bounds[3]) + margin,
+                )
+                if buffered not in seen:
+                    seen.add(buffered)
+                    hotspot_bounds.append(buffered)
+
+        return tuple(hotspot_bounds)
+
+    def _analyze_results(
+        self,
+        ordered_net_ids: Sequence[str],
+        results: dict[str, RoutingResult],
+        *,
+        capture_component_conflicts: bool,
+        count_iteration_metrics: bool,
+    ) -> tuple[dict[str, RoutingResult], dict[str, PathVerificationDetail], _IterationReview]:
+        if count_iteration_metrics:
+            self.metrics.total_iteration_reverify_calls += 1
+        conflict_edges: set[tuple[str, str]] = set()
+        conflicting_nets: set[str] = set()
+        completed_net_ids: set[str] = set()
+        total_dynamic_collisions = 0
+        analyzed_results = dict(results)
+        details_by_net: dict[str, PathVerificationDetail] = {}
+
+        for net_id in ordered_net_ids:
+            result = results.get(net_id)
+            if not result or not result.path or not result.reached_target:
+                continue
+
+            if count_iteration_metrics:
+                self.metrics.total_iteration_reverified_nets += 1
+            detail = self.context.cost_evaluator.collision_engine.verify_path_details(
+                net_id,
+                result.path,
+                capture_component_conflicts=capture_component_conflicts,
+            )
+            details_by_net[net_id] = detail
+            analyzed_results[net_id] = RoutingResult(
+                net_id=net_id,
+                path=result.path,
+                reached_target=result.reached_target,
+                report=detail.report,
+            )
+            total_dynamic_collisions += detail.report.dynamic_collision_count
+            if analyzed_results[net_id].outcome == "completed":
+                completed_net_ids.add(net_id)
+            if not detail.conflicting_net_ids:
+                continue
+            conflicting_nets.add(net_id)
+            for other_net_id in detail.conflicting_net_ids:
+                conflicting_nets.add(other_net_id)
+                if other_net_id == net_id:
+                    continue
+                conflict_edges.add(tuple(sorted((net_id, other_net_id))))
+
+        if count_iteration_metrics:
+            self.metrics.total_iteration_conflicting_nets += len(conflicting_nets)
+            self.metrics.total_iteration_conflict_edges += len(conflict_edges)
+        return (
+            analyzed_results,
+            details_by_net,
+            _IterationReview(
+                conflicting_nets=conflicting_nets,
+                conflict_edges=conflict_edges,
+                completed_net_ids=completed_net_ids,
+                total_dynamic_collisions=total_dynamic_collisions,
+            ),
+        )
+
+    def _capture_frontier_trace(
+        self,
+        state: _RoutingState,
+        final_results: dict[str, RoutingResult],
+    ) -> None:
+        if not self.context.options.diagnostics.capture_frontier_trace:
+            return
+
+        self.frontier_trace = []
+        state.results = dict(final_results)
+        state.results, details_by_net, _ = self._analyze_results(
+            state.ordered_net_ids,
+            state.results,
+            capture_component_conflicts=True,
+            count_iteration_metrics=False,
+        )
+
+        original_metrics = self.metrics
+        original_context_metrics = self.context.metrics
+        original_engine_metrics = self.context.cost_evaluator.collision_engine.metrics
+        original_danger_metrics = None
+        if self.context.cost_evaluator.danger_map is not None:
+            original_danger_metrics = self.context.cost_evaluator.danger_map.metrics
+
+        try:
+            for net_id in state.ordered_net_ids:
+                result = state.results.get(net_id)
+                detail = details_by_net.get(net_id)
+                if result is None or detail is None or not result.reached_target:
+                    continue
+                if detail.report.dynamic_collision_count == 0 or not detail.component_conflicts:
+                    continue
+
+                hotspot_bounds = self._build_frontier_hotspot_bounds(state, net_id, details_by_net)
+                if not hotspot_bounds:
+                    continue
+
+                scratch_metrics = AStarMetrics()
+                self.context.metrics = scratch_metrics
+                self.context.cost_evaluator.collision_engine.metrics = scratch_metrics
+                if self.context.cost_evaluator.danger_map is not None:
+                    self.context.cost_evaluator.danger_map.metrics = scratch_metrics
+
+                guidance_seed = result.as_seed().segments if result.path else None
+                guidance_bonus = 0.0
+                if guidance_seed:
+                    guidance_bonus = max(10.0, self.context.options.objective.bend_penalty * 0.25)
+                collector = FrontierTraceCollector(hotspot_bounds=hotspot_bounds)
+                run_config = SearchRunConfig.from_options(
+                    self.context.options,
+                    return_partial=True,
+                    store_expanded=False,
+                    guidance_seed=guidance_seed,
+                    guidance_bonus=guidance_bonus,
+                    frontier_trace=collector,
+                    self_collision_check=(net_id in state.needs_self_collision_check),
+                    node_limit=self.context.options.search.node_limit,
+                )
+
+                self.context.cost_evaluator.collision_engine.remove_path(net_id)
+                try:
+                    route_astar(
+                        state.net_specs[net_id].start,
+                        state.net_specs[net_id].target,
+                        state.net_specs[net_id].width,
+                        context=self.context,
+                        metrics=scratch_metrics,
+                        net_id=net_id,
+                        config=run_config,
+                    )
+                finally:
+                    if result.path:
+                        self._install_path(net_id, result.path)
+
+                self.frontier_trace.append(
+                    NetFrontierTrace(
+                        net_id=net_id,
+                        hotspot_bounds=hotspot_bounds,
+                        pruned_closed_set=collector.pruned_closed_set,
+                        pruned_hard_collision=collector.pruned_hard_collision,
+                        pruned_self_collision=collector.pruned_self_collision,
+                        pruned_cost=collector.pruned_cost,
+                        samples=tuple(
+                            FrontierPruneSample(
+                                reason=reason,  # type: ignore[arg-type]
+                                move_type=move_type,
+                                hotspot_index=hotspot_index,
+                                parent_state=parent_state,
+                                end_state=end_state,
+                            )
+                            for reason, move_type, hotspot_index, parent_state, end_state in collector.samples
+                        ),
+                    )
+                )
+        finally:
+            self.metrics = original_metrics
+            self.context.metrics = original_context_metrics
+            self.context.cost_evaluator.collision_engine.metrics = original_engine_metrics
+            if self.context.cost_evaluator.danger_map is not None:
+                self.context.cost_evaluator.danger_map.metrics = original_danger_metrics
+
+    def _whole_set_is_better(
+        self,
+        candidate_results: dict[str, RoutingResult],
+        candidate_review: _IterationReview,
+        incumbent_results: dict[str, RoutingResult],
+        incumbent_review: _IterationReview,
+    ) -> bool:
+        candidate_completed = len(candidate_review.completed_net_ids)
+        incumbent_completed = len(incumbent_review.completed_net_ids)
+        if candidate_completed != incumbent_completed:
+            return candidate_completed > incumbent_completed
+
+        candidate_edges = len(candidate_review.conflict_edges)
+        incumbent_edges = len(incumbent_review.conflict_edges)
+        if candidate_edges != incumbent_edges:
+            return candidate_edges < incumbent_edges
+
+        if candidate_review.total_dynamic_collisions != incumbent_review.total_dynamic_collisions:
+            return candidate_review.total_dynamic_collisions < incumbent_review.total_dynamic_collisions
+
+        candidate_length = sum(
+            result.report.total_length
+            for result in candidate_results.values()
+            if result.reached_target
+        )
+        incumbent_length = sum(
+            result.report.total_length
+            for result in incumbent_results.values()
+            if result.reached_target
+        )
+        if abs(candidate_length - incumbent_length) > 1e-6:
+            return candidate_length < incumbent_length
+        return False
+
+    def _collect_pair_local_targets(
+        self,
+        state: _RoutingState,
+        results: dict[str, RoutingResult],
+        review: _IterationReview,
+    ) -> list[_PairLocalTarget]:
+        if not review.conflict_edges:
+            return []
+        order_index = {net_id: idx for idx, net_id in enumerate(state.ordered_net_ids)}
+        seen_net_ids: set[str] = set()
+        targets: list[_PairLocalTarget] = []
+        for left_net_id, right_net_id in sorted(review.conflict_edges):
+            if left_net_id in seen_net_ids or right_net_id in seen_net_ids:
+                return []
+            left_result = results.get(left_net_id)
+            right_result = results.get(right_net_id)
+            if (
+                left_result is None
+                or right_result is None
+                or not left_result.reached_target
+                or not right_result.reached_target
+            ):
+                continue
+            seen_net_ids.update((left_net_id, right_net_id))
+            targets.append(_PairLocalTarget(net_ids=(left_net_id, right_net_id)))
+        targets.sort(key=lambda target: min(order_index[target.net_ids[0]], order_index[target.net_ids[1]]))
+        return targets
+
+    def _build_pair_local_context(
+        self,
+        state: _RoutingState,
+        incumbent_results: dict[str, RoutingResult],
+        pair_net_ids: tuple[str, str],
+    ) -> AStarContext:
+        problem = self.context.problem
+        objective = self.context.options.objective
+        static_obstacles = tuple(self.context.cost_evaluator.collision_engine._static_obstacles.geometries.values())
+        engine = RoutingWorld(
+            clearance=self.context.cost_evaluator.collision_engine.clearance,
+            safety_zone_radius=self.context.cost_evaluator.collision_engine.safety_zone_radius,
+        )
+        for obstacle in static_obstacles:
+            engine.add_static_obstacle(obstacle)
+        for net_id in state.ordered_net_ids:
+            if net_id in pair_net_ids:
+                continue
+            result = incumbent_results.get(net_id)
+            if result is None or not result.path:
+                continue
+            for component in result.path:
+                for polygon in component.physical_geometry:
+                    engine.add_static_obstacle(polygon)
+
+        danger_map = DangerMap(bounds=problem.bounds)
+        danger_map.precompute(list(static_obstacles))
+        evaluator = CostEvaluator(
+            engine,
+            danger_map,
+            unit_length_cost=objective.unit_length_cost,
+            greedy_h_weight=self.context.cost_evaluator.greedy_h_weight,
+            bend_penalty=objective.bend_penalty,
+            sbend_penalty=objective.sbend_penalty,
+            danger_weight=objective.danger_weight,
+        )
+        return AStarContext(
+            evaluator,
+            RoutingProblem(
+                bounds=problem.bounds,
+                nets=tuple(state.net_specs[net_id] for net_id in state.ordered_net_ids),
+                static_obstacles=static_obstacles,
+                clearance=problem.clearance,
+                safety_zone_radius=problem.safety_zone_radius,
+            ),
+            self.context.options,
+            metrics=AStarMetrics(),
+        )
+
+    def _run_pair_local_attempt(
+        self,
+        state: _RoutingState,
+        incumbent_results: dict[str, RoutingResult],
+        pair_order: tuple[str, str],
+    ) -> tuple[dict[str, RoutingResult], int] | None:
+        local_context = self._build_pair_local_context(state, incumbent_results, pair_order)
+        local_results = dict(incumbent_results)
+
+        for net_id in pair_order:
+            net = state.net_specs[net_id]
+            guidance_result = incumbent_results.get(net_id)
+            guidance_seed = None
+            guidance_bonus = 0.0
+            if guidance_result and guidance_result.reached_target and guidance_result.path:
+                guidance_seed = guidance_result.as_seed().segments
+                guidance_bonus = max(10.0, self.context.options.objective.bend_penalty * 0.25)
+
+            run_config = SearchRunConfig.from_options(
+                self.context.options,
+                return_partial=False,
+                skip_congestion=True,
+                self_collision_check=(net_id in state.needs_self_collision_check),
+                guidance_seed=guidance_seed,
+                guidance_bonus=guidance_bonus,
+                node_limit=self.context.options.search.node_limit,
+            )
+            path = route_astar(
+                net.start,
+                net.target,
+                net.width,
+                context=local_context,
+                metrics=local_context.metrics,
+                net_id=net_id,
+                config=run_config,
+            )
+            if not path or path[-1].end_port != net.target:
+                return None
+
+            report = local_context.cost_evaluator.collision_engine.verify_path_report(net_id, path)
+            if not report.is_valid:
+                return None
+            local_results[net_id] = RoutingResult(
+                net_id=net_id,
+                path=tuple(path),
+                reached_target=True,
+                report=report,
+            )
+            for component in path:
+                for polygon in component.physical_geometry:
+                    local_context.cost_evaluator.collision_engine.add_static_obstacle(polygon)
+            local_context.clear_static_caches()
+
+        return local_results, local_context.metrics.total_nodes_expanded
+
+    def _run_pair_local_search(self, state: _RoutingState) -> None:
+        state.results, _details_by_net, review = self._analyze_results(
+            state.ordered_net_ids,
+            state.results,
+            capture_component_conflicts=True,
+            count_iteration_metrics=False,
+        )
+        targets = self._collect_pair_local_targets(state, state.results, review)
+        if not targets:
+            return
+
+        for target in targets[:2]:
+            self.metrics.total_pair_local_search_pairs_considered += 1
+            incumbent_results = dict(state.results)
+            incumbent_review = review
+            accepted = False
+            for pair_order in (target.net_ids, target.net_ids[::-1]):
+                self.metrics.total_pair_local_search_attempts += 1
+                candidate = self._run_pair_local_attempt(state, incumbent_results, pair_order)
+                if candidate is None:
+                    continue
+                candidate_results, nodes_expanded = candidate
+                self.metrics.total_pair_local_search_nodes_expanded += nodes_expanded
+                self._replace_installed_paths(state, candidate_results)
+                candidate_results, _candidate_details_by_net, candidate_review = self._analyze_results(
+                    state.ordered_net_ids,
+                    candidate_results,
+                    capture_component_conflicts=True,
+                    count_iteration_metrics=False,
+                )
+                if self._whole_set_is_better(
+                    candidate_results,
+                    candidate_review,
+                    incumbent_results,
+                    incumbent_review,
+                ):
+                    self.metrics.total_pair_local_search_accepts += 1
+                    state.results = candidate_results
+                    review = candidate_review
+                    accepted = True
+                    break
+                self._replace_installed_paths(state, incumbent_results)
+
+            if not accepted:
+                state.results = incumbent_results
+                self._replace_installed_paths(state, incumbent_results)
+
     def _route_net_once(
         self,
         state: _RoutingState,
@@ -182,16 +763,25 @@ class PathFinder:
         else:
             coll_model, _ = resolve_bend_geometry(search)
             skip_congestion = False
+            guidance_seed = None
+            guidance_bonus = 0.0
             if congestion.use_tiered_strategy and iteration == 0:
                 skip_congestion = True
                 if coll_model == "arc":
                     coll_model = "clipped_bbox"
+            elif iteration > 0:
+                guidance_result = state.results.get(net_id)
+                if guidance_result and guidance_result.reached_target and guidance_result.path:
+                    guidance_seed = guidance_result.as_seed().segments
+                    guidance_bonus = max(10.0, self.context.options.objective.bend_penalty * 0.25)
 
             run_config = SearchRunConfig.from_options(
                 self.context.options,
                 bend_collision_type=coll_model,
                 return_partial=True,
                 store_expanded=diagnostics.capture_expanded,
+                guidance_seed=guidance_seed,
+                guidance_bonus=guidance_bonus,
                 skip_congestion=skip_congestion,
                 self_collision_check=(net_id in state.needs_self_collision_check),
                 node_limit=search.node_limit,
@@ -233,29 +823,49 @@ class PathFinder:
         self,
         state: _RoutingState,
         iteration: int,
+        reroute_net_ids: set[str],
         iteration_callback: Callable[[int, dict[str, RoutingResult]], None] | None,
-    ) -> dict[str, RoutingOutcome] | None:
-        outcomes: dict[str, RoutingOutcome] = {}
+    ) -> _IterationReview | None:
         congestion = self.context.options.congestion
         self.metrics.total_route_iterations += 1
         self.metrics.reset_per_route()
-
         if congestion.shuffle_nets and (iteration > 0 or state.initial_paths is None):
             iteration_seed = (congestion.seed + iteration) if congestion.seed is not None else None
             random.Random(iteration_seed).shuffle(state.ordered_net_ids)
 
-        for net_id in state.ordered_net_ids:
+        routed_net_ids = [net_id for net_id in state.ordered_net_ids if net_id in reroute_net_ids]
+        self.metrics.total_nets_carried_forward += len(state.ordered_net_ids) - len(routed_net_ids)
+
+        for net_id in routed_net_ids:
             if time.monotonic() - state.start_time > state.timeout_s:
                 self.metrics.total_timeout_events += 1
                 return None
 
             result = self._route_net_once(state, iteration, net_id)
             state.results[net_id] = result
-            outcomes[net_id] = result.outcome
+
+        review = self._reverify_iteration_results(state)
 
         if iteration_callback:
             iteration_callback(iteration, state.results)
-        return outcomes
+        return review
+
+    def _reverify_iteration_results(self, state: _RoutingState) -> _IterationReview:
+        state.results, details_by_net, review = self._analyze_results(
+            state.ordered_net_ids,
+            state.results,
+            capture_component_conflicts=self.context.options.diagnostics.capture_conflict_trace,
+            count_iteration_metrics=True,
+        )
+        self._capture_conflict_trace_entry(
+            state,
+            stage="iteration",
+            iteration=self.metrics.total_route_iterations - 1,
+            results=state.results,
+            details_by_net=details_by_net,
+            review=review,
+        )
+        return review
 
     def _run_iterations(
         self,
@@ -264,10 +874,33 @@ class PathFinder:
     ) -> bool:
         congestion = self.context.options.congestion
         for iteration in range(congestion.max_iterations):
-            outcomes = self._run_iteration(state, iteration, iteration_callback)
-            if outcomes is None:
+            review = self._run_iteration(
+                state,
+                iteration,
+                set(state.ordered_net_ids),
+                iteration_callback,
+            )
+            if review is None:
                 return True
-            if not any(outcome in {"colliding", "partial", "unroutable"} for outcome in outcomes.values()):
+            self._update_best_iteration(state, review)
+            if not any(
+                result.outcome in {"colliding", "partial", "unroutable"}
+                for result in state.results.values()
+            ):
+                return False
+
+            current_signature = tuple(sorted(review.conflict_edges))
+            repeated = (
+                bool(current_signature)
+                and (
+                    current_signature == state.last_conflict_signature
+                    or len(current_signature) == state.last_conflict_edge_count
+                )
+            )
+            state.repeated_conflict_count = state.repeated_conflict_count + 1 if repeated else 0
+            state.last_conflict_signature = current_signature
+            state.last_conflict_edge_count = len(current_signature)
+            if state.repeated_conflict_count >= 2:
                 return False
             self.context.congestion_penalty *= congestion.multiplier
         return False
@@ -285,27 +918,49 @@ class PathFinder:
             self.context.cost_evaluator.collision_engine.remove_path(net_id)
             refined_path = self.refiner.refine_path(net_id, net.start, net.width, result.path)
             self._install_path(net_id, refined_path)
-            report = self.context.cost_evaluator.collision_engine.verify_path_report(net_id, refined_path)
+            # Defer full verification until _verify_results() so we do not
+            # verify the same refined path twice in one route_all() call.
             state.results[net_id] = RoutingResult(
                 net_id=net_id,
                 path=tuple(refined_path),
                 reached_target=result.reached_target,
-                report=report,
+                report=result.report,
             )
 
     def _verify_results(self, state: _RoutingState) -> dict[str, RoutingResult]:
         final_results: dict[str, RoutingResult] = {}
+        details_by_net: dict[str, PathVerificationDetail] = {}
         for net in self.context.problem.nets:
             result = state.results.get(net.net_id)
             if not result or not result.path:
                 final_results[net.net_id] = RoutingResult(net_id=net.net_id, path=(), reached_target=False)
                 continue
-            report = self.context.cost_evaluator.collision_engine.verify_path_report(net.net_id, result.path)
+            detail = self.context.cost_evaluator.collision_engine.verify_path_details(
+                net.net_id,
+                result.path,
+                capture_component_conflicts=self.context.options.diagnostics.capture_conflict_trace,
+            )
+            details_by_net[net.net_id] = detail
             final_results[net.net_id] = RoutingResult(
                 net_id=net.net_id,
                 path=result.path,
                 reached_target=result.reached_target,
-                report=report,
+                report=detail.report,
+            )
+        if self.context.options.diagnostics.capture_conflict_trace:
+            _, _, review = self._analyze_results(
+                state.ordered_net_ids,
+                final_results,
+                capture_component_conflicts=True,
+                count_iteration_metrics=False,
+            )
+            self._capture_conflict_trace_entry(
+                state,
+                stage="final",
+                iteration=None,
+                results=final_results,
+                details_by_net=details_by_net,
+                review=review,
             )
         return final_results
 
@@ -316,15 +971,38 @@ class PathFinder:
     ) -> dict[str, RoutingResult]:
         self.context.congestion_penalty = self.context.options.congestion.base_penalty
         self.accumulated_expanded_nodes = []
+        self.conflict_trace = []
+        self.frontier_trace = []
         self.metrics.reset_totals()
         self.metrics.reset_per_route()
 
         state = self._prepare_state()
         timed_out = self._run_iterations(state, iteration_callback)
         self.accumulated_expanded_nodes = list(state.accumulated_expanded_nodes)
+        self._restore_best_iteration(state)
+        if self.context.options.diagnostics.capture_conflict_trace:
+            state.results, details_by_net, review = self._analyze_results(
+                state.ordered_net_ids,
+                state.results,
+                capture_component_conflicts=True,
+                count_iteration_metrics=False,
+            )
+            self._capture_conflict_trace_entry(
+                state,
+                stage="restored_best",
+                iteration=None,
+                results=state.results,
+                details_by_net=details_by_net,
+                review=review,
+            )
 
         if timed_out:
-            return self._verify_results(state)
+            final_results = self._verify_results(state)
+            self._capture_frontier_trace(state, final_results)
+            return final_results
 
+        self._run_pair_local_search(state)
         self._refine_results(state)
-        return self._verify_results(state)
+        final_results = self._verify_results(state)
+        self._capture_frontier_trace(state, final_results)
+        return final_results

inire/router/_search.py

@@ -41,11 +41,17 @@ def route_astar(
     open_set: list[_AStarNode] = []
     closed_set: dict[tuple[int, int, int], float] = {}
     congestion_cache: dict[tuple, int] = {}
+    congestion_presence_cache: dict[tuple[str, int, int, int, int], bool] = {}
+    congestion_candidate_precheck_cache: dict[tuple[str, int, int, int, int], bool] = {}
+    congestion_net_envelope_cache: dict[tuple[str, int, int, int, int], tuple[str, ...]] = {}
+    congestion_grid_net_cache: dict[tuple[str, int, int, int, int], tuple[str, ...]] = {}
+    congestion_grid_span_cache: dict[tuple[str, int, int, int, int], dict[str, tuple[int, ...]]] = {}
 
     start_node = _AStarNode(
         start,
         0.0,
         context.cost_evaluator.h_manhattan(start, target, min_bend_radius=context.min_bend_radius),
+        seed_index=0 if config.guidance_seed else None,
     )
     heapq.heappush(open_set, start_node)
     best_node = start_node
@@ -89,6 +95,11 @@ def route_astar(
             context,
             metrics,
             congestion_cache,
+            congestion_presence_cache,
+            congestion_candidate_precheck_cache,
+            congestion_net_envelope_cache,
+            congestion_grid_net_cache,
+            congestion_grid_span_cache,
             config=config,
         )
 

inire/router/cost.py

@@ -1,5 +1,6 @@
 from __future__ import annotations
 
+from time import perf_counter_ns
 from typing import TYPE_CHECKING
 
 import numpy as np
@@ -130,10 +131,16 @@ class CostEvaluator:
         start_port: Port | None = None,
         weights: ObjectiveWeights | None = None,
     ) -> float:
+        metrics = self.collision_engine.metrics
+        if metrics is not None:
+            metrics.total_score_component_calls += 1
+        start_ns = perf_counter_ns()
         active_weights = self._resolve_weights(weights)
         danger_map = self.danger_map
         end_port = component.end_port
         if danger_map is not None and not danger_map.is_within_bounds(end_port.x, end_port.y):
+            if metrics is not None:
+                metrics.total_score_component_total_ns += perf_counter_ns() - start_ns
             return 1e15
 
         move_radius = None
@@ -145,7 +152,8 @@ class CostEvaluator:
             weights=active_weights,
         )
 
-        if danger_map is not None and active_weights.danger_weight:
+        # Skip danger sampling entirely when there are no static obstacles in the KD-tree.
+        if danger_map is not None and active_weights.danger_weight and danger_map.tree is not None:
             cost_s = danger_map.get_cost(start_port.x, start_port.y) if start_port else 0.0
             cost_e = danger_map.get_cost(end_port.x, end_port.y)
             if start_port:
@@ -155,6 +163,8 @@ class CostEvaluator:
                 total_cost += component.length * active_weights.danger_weight * (cost_s + cost_m + cost_e) / 3.0
             else:
                 total_cost += component.length * active_weights.danger_weight * cost_e
+        if metrics is not None:
+            metrics.total_score_component_total_ns += perf_counter_ns() - start_ns
         return total_cost
 
     def component_penalty(
@@ -181,6 +191,9 @@ class CostEvaluator:
         *,
         weights: ObjectiveWeights | None = None,
     ) -> float:
+        metrics = self.collision_engine.metrics
+        if metrics is not None:
+            metrics.total_path_cost_calls += 1
         active_weights = self._resolve_weights(weights)
         total = 0.0
         current_port = start_port

inire/router/danger_map.py

@@ -1,6 +1,7 @@
 from __future__ import annotations
 
 from collections import OrderedDict
+from time import perf_counter_ns
 from typing import TYPE_CHECKING
 
 import numpy
@@ -8,6 +9,7 @@ from scipy.spatial import cKDTree
 
 if TYPE_CHECKING:
     from shapely.geometry import Polygon
+    from inire.router._astar_types import AStarMetrics
 
 
 _COST_CACHE_SIZE = 100000
@@ -18,7 +20,7 @@ class DangerMap:
     A proximity cost evaluator using a KD-Tree of obstacle boundary points.
     Scales with obstacle perimeter rather than design area.
     """
-    __slots__ = ('minx', 'miny', 'maxx', 'maxy', 'resolution', 'safety_threshold', 'k', 'tree', '_cost_cache')
+    __slots__ = ('minx', 'miny', 'maxx', 'maxy', 'resolution', 'safety_threshold', 'k', 'tree', '_cost_cache', 'metrics')
 
     def __init__(
             self,
@@ -42,6 +44,7 @@ class DangerMap:
         self.k = k
         self.tree: cKDTree | None = None
         self._cost_cache: OrderedDict[tuple[int, int], float] = OrderedDict()
+        self.metrics: AStarMetrics | None = None
 
     def precompute(self, obstacles: list[Polygon]) -> None:
         """
@@ -82,17 +85,28 @@ class DangerMap:
         Get the proximity cost at a specific coordinate using the KD-Tree.
         Coordinates are quantized to 1nm to improve cache performance.
         """
+        metrics = self.metrics
+        if metrics is not None:
+            metrics.total_danger_map_lookup_calls += 1
+        start_ns = perf_counter_ns()
         qx_milli = int(round(x * 1000))
         qy_milli = int(round(y * 1000))
         key = (qx_milli, qy_milli)
         if key in self._cost_cache:
+            if metrics is not None:
+                metrics.total_danger_map_cache_hits += 1
+                metrics.total_danger_map_total_ns += perf_counter_ns() - start_ns
             self._cost_cache.move_to_end(key)
             return self._cost_cache[key]
 
+        if metrics is not None:
+            metrics.total_danger_map_cache_misses += 1
         cost = self._compute_cost_quantized(qx_milli, qy_milli)
         self._cost_cache[key] = cost
         if len(self._cost_cache) > _COST_CACHE_SIZE:
             self._cost_cache.popitem(last=False)
+        if metrics is not None:
+            metrics.total_danger_map_total_ns += perf_counter_ns() - start_ns
         return cost
 
     def _compute_cost_quantized(self, qx_milli: int, qy_milli: int) -> float:
@@ -102,6 +116,8 @@ class DangerMap:
             return 1e15
         if self.tree is None:
             return 0.0
+        if self.metrics is not None:
+            self.metrics.total_danger_map_query_calls += 1
         dist, _ = self.tree.query([qx, qy], distance_upper_bound=self.safety_threshold)
         if dist >= self.safety_threshold:
             return 0.0

inire/router/refiner.py

@@ -128,6 +128,7 @@ class PathRefiner:
         x_max = max(0.0, float(local_dx)) + 0.01
 
         for bounds in self.collision_engine.iter_static_obstacle_bounds(query_bounds):
+            self.context.metrics.total_refinement_static_bounds_checked += 1
             local_corners = (
                 self._to_local_xy(start, bounds[0], bounds[1]),
                 self._to_local_xy(start, bounds[0], bounds[3]),
@@ -144,6 +145,7 @@ class PathRefiner:
             negative_anchors.add(obs_min_y)
 
         for bounds in self.collision_engine.iter_dynamic_path_bounds(query_bounds):
+            self.context.metrics.total_refinement_dynamic_bounds_checked += 1
             local_corners = (
                 self._to_local_xy(start, bounds[0], bounds[1]),
                 self._to_local_xy(start, bounds[0], bounds[3]),
@@ -166,6 +168,7 @@ class PathRefiner:
             if anchor < min(0.0, float(local_dy)) + 0.01:
                 direct_extents.add(anchor - pad)
 
+        self.context.metrics.total_refinement_candidate_side_extents += len(direct_extents)
         return sorted(direct_extents, key=lambda value: (abs(value), value))
 
     def _build_same_orientation_dogleg(
@@ -243,6 +246,7 @@ class PathRefiner:
                 local_dx, _ = self._to_local(window_start, window_end)
                 if local_dx < 4.0 * min_radius - 0.01:
                     continue
+                self.context.metrics.total_refinement_windows_considered += 1
                 windows.append((start_idx, end_idx))
         return windows
 
@@ -270,12 +274,15 @@ class PathRefiner:
                 replacement = self._build_same_orientation_dogleg(window_start, window_end, net_width, radius, side_extent)
                 if replacement is None:
                     continue
+                self.context.metrics.total_refinement_candidates_built += 1
                 candidate_path = path[:start_idx] + replacement + path[end_idx:]
+                self.context.metrics.total_refinement_candidates_verified += 1
                 report = self.collision_engine.verify_path_report(net_id, candidate_path)
                 if not report.is_valid:
                     continue
                 candidate_cost = self.path_cost(candidate_path)
                 if candidate_cost + 1e-6 < best_candidate_cost:
+                    self.context.metrics.total_refinement_candidates_accepted += 1
                     best_candidate_cost = candidate_cost
                     best_path = candidate_path
 

inire/router/visibility.py

@@ -16,7 +16,15 @@ class VisibilityManager:
     """
     Manages corners of static obstacles for sparse A* / Visibility Graph jumps.
     """
-    __slots__ = ("collision_engine", "corners", "corner_index", "_corner_graph", "_point_visibility_cache", "_built_static_version")
+    __slots__ = (
+        "collision_engine",
+        "corners",
+        "corner_index",
+        "_corner_graph",
+        "_point_visibility_cache",
+        "_corner_index_version",
+        "_corner_graph_version",
+    )
 
     def __init__(self, collision_engine: RoutingWorld) -> None:
         self.collision_engine = collision_engine
@@ -24,8 +32,8 @@ class VisibilityManager:
         self.corner_index = rtree.index.Index()
         self._corner_graph: dict[int, list[tuple[float, float, float]]] = {}
         self._point_visibility_cache: dict[tuple[int, int, int], list[tuple[float, float, float]]] = {}
-        self._built_static_version = -1
-        self._build()
+        self._corner_index_version = -1
+        self._corner_graph_version = -1
 
     def clear_cache(self) -> None:
         """
@@ -35,19 +43,31 @@ class VisibilityManager:
         self.corner_index = rtree.index.Index()
         self._corner_graph = {}
         self._point_visibility_cache = {}
-        self._build()
+        self._corner_index_version = -1
+        self._corner_graph_version = -1
+
+    def ensure_corner_index_current(self) -> None:
+        if self._corner_index_version != self.collision_engine.get_static_version():
+            self._build_corner_index()
+
+    def ensure_corner_graph_current(self) -> None:
+        self.ensure_corner_index_current()
+        static_version = self.collision_engine.get_static_version()
+        if self._corner_graph_version != static_version:
+            self._build_corner_graph()
 
     def _ensure_current(self) -> None:
-        if self._built_static_version != self.collision_engine.get_static_version():
-            self.clear_cache()
+        self.ensure_corner_graph_current()
 
-    def _build(self) -> None:
-        """
-        Extract corners and pre-compute corner-to-corner visibility.
-        """
+    def _build_corner_index(self) -> None:
         if self.collision_engine.metrics is not None:
-            self.collision_engine.metrics.total_visibility_builds += 1
-        self._built_static_version = self.collision_engine.get_static_version()
+            self.collision_engine.metrics.total_visibility_corner_index_builds += 1
+        self.corners = []
+        self.corner_index = rtree.index.Index()
+        self._corner_graph = {}
+        self._point_visibility_cache = {}
+        self._corner_graph_version = -1
+        self._corner_index_version = self.collision_engine.get_static_version()
         raw_corners = []
         for poly in self.collision_engine.iter_static_dilated_geometries():
             coords = list(poly.exterior.coords)
@@ -63,7 +83,6 @@ class VisibilityManager:
         if not raw_corners:
             return
 
-        # Deduplicate repeated corner coordinates
         seen = set()
         for x, y in raw_corners:
             sx, sy = round(x, 3), round(y, 3)
@@ -71,10 +90,22 @@ class VisibilityManager:
                 seen.add((sx, sy))
                 self.corners.append((sx, sy))
 
-        # Build spatial index for corners
         for i, (x, y) in enumerate(self.corners):
             self.corner_index.insert(i, (x, y, x, y))
 
+    def _build_corner_graph(self) -> None:
+        """
+        Pre-compute corner-to-corner visibility from the current corner index.
+        """
+        self.ensure_corner_index_current()
+        if self.collision_engine.metrics is not None:
+            self.collision_engine.metrics.total_visibility_builds += 1
+        self._corner_graph = {}
+        self._corner_graph_version = self.collision_engine.get_static_version()
+
+        if not self.corners:
+            return
+
         # Pre-compute visibility graph between corners
         num_corners = len(self.corners)
         if num_corners > 200:
@@ -93,11 +124,12 @@ class VisibilityManager:
                 dx, dy = cx - p1.x, cy - p1.y
                 dist = numpy.sqrt(dx**2 + dy**2)
                 angle = numpy.degrees(numpy.arctan2(dy, dx))
-                reach = self.collision_engine.ray_cast(p1, angle, max_dist=dist + 0.05)
+                reach = self.collision_engine.ray_cast(p1, angle, max_dist=dist + 0.05, caller="visibility_build")
                 if reach >= dist - 0.01:
                     self._corner_graph[i].append((cx, cy, dist))
 
     def _corner_idx_at(self, origin: Port) -> int | None:
+        self.ensure_corner_index_current()
         ox, oy = round(origin.x, 3), round(origin.y, 3)
         nearby = list(self.corner_index.intersection((ox - 0.001, oy - 0.001, ox + 0.001, oy + 0.001)))
         for idx in nearby:
@@ -106,6 +138,49 @@ class VisibilityManager:
                 return idx
         return None
 
+    def get_tangent_corner_candidates(
+        self,
+        origin: Port,
+        *,
+        min_forward: float,
+        max_forward: float,
+        radii: tuple[float, ...],
+        tolerance: float = 2.0,
+    ) -> list[int]:
+        self.ensure_corner_index_current()
+        if max_forward <= min_forward or not radii or not self.corners:
+            return []
+
+        candidate_ids: set[int] = set()
+        x0 = float(origin.x)
+        y0 = float(origin.y)
+
+        def _add_hits(bounds: tuple[float, float, float, float]) -> None:
+            min_x, min_y, max_x, max_y = bounds
+            if min_x > max_x or min_y > max_y:
+                return
+            candidate_ids.update(self.corner_index.intersection(bounds))
+
+        for radius in radii:
+            if origin.r == 0:
+                x_bounds = (x0 + min_forward, x0 + max_forward)
+                _add_hits((x_bounds[0], y0 + radius - tolerance, x_bounds[1], y0 + radius + tolerance))
+                _add_hits((x_bounds[0], y0 - radius - tolerance, x_bounds[1], y0 - radius + tolerance))
+            elif origin.r == 180:
+                x_bounds = (x0 - max_forward, x0 - min_forward)
+                _add_hits((x_bounds[0], y0 + radius - tolerance, x_bounds[1], y0 + radius + tolerance))
+                _add_hits((x_bounds[0], y0 - radius - tolerance, x_bounds[1], y0 - radius + tolerance))
+            elif origin.r == 90:
+                y_bounds = (y0 + min_forward, y0 + max_forward)
+                _add_hits((x0 + radius - tolerance, y_bounds[0], x0 + radius + tolerance, y_bounds[1]))
+                _add_hits((x0 - radius - tolerance, y_bounds[0], x0 - radius + tolerance, y_bounds[1]))
+            else:
+                y_bounds = (y0 - max_forward, y0 - min_forward)
+                _add_hits((x0 + radius - tolerance, y_bounds[0], x0 + radius + tolerance, y_bounds[1]))
+                _add_hits((x0 - radius - tolerance, y_bounds[0], x0 - radius + tolerance, y_bounds[1]))
+
+        return sorted(candidate_ids)
+
     def get_point_visibility(self, origin: Port, max_dist: float = 1000.0) -> list[tuple[float, float, float]]:
         """
         Find visible corners from an arbitrary point.
@@ -113,11 +188,13 @@ class VisibilityManager:
         """
         if self.collision_engine.metrics is not None:
             self.collision_engine.metrics.total_visibility_point_queries += 1
-        self._ensure_current()
+        self.ensure_corner_index_current()
         if max_dist < 0:
             return []
 
         corner_idx = self._corner_idx_at(origin)
+        if corner_idx is not None:
+            self.ensure_corner_graph_current()
         if corner_idx is not None and corner_idx in self._corner_graph:
             return [corner for corner in self._corner_graph[corner_idx] if corner[2] <= max_dist]
 
@@ -143,7 +220,7 @@ class VisibilityManager:
                 continue
 
             angle = numpy.degrees(numpy.arctan2(dy, dx))
-            reach = self.collision_engine.ray_cast(origin, angle, max_dist=dist + 0.05)
+            reach = self.collision_engine.ray_cast(origin, angle, max_dist=dist + 0.05, caller="visibility_query")
             if reach >= dist - 0.01:
                 visible.append((cx, cy, dist))
 
@@ -156,14 +233,14 @@ class VisibilityManager:
         This avoids the expensive arbitrary-point visibility scan in hot search paths.
         """
         if self.collision_engine.metrics is not None:
-            self.collision_engine.metrics.total_visibility_corner_queries += 1
-        self._ensure_current()
+            self.collision_engine.metrics.total_visibility_corner_queries_exact += 1
+        self.ensure_corner_graph_current()
         if max_dist < 0:
             return []
 
         corner_idx = self._corner_idx_at(origin)
         if corner_idx is not None and corner_idx in self._corner_graph:
             if self.collision_engine.metrics is not None:
-                self.collision_engine.metrics.total_visibility_corner_hits += 1
+                self.collision_engine.metrics.total_visibility_corner_hits_exact += 1
             return [corner for corner in self._corner_graph[corner_idx] if corner[2] <= max_dist]
         return []

inire/tests/example_scenarios.py

@@ -15,6 +15,7 @@ from inire import (
     RoutingOptions,
     RoutingProblem,
     RoutingResult,
+    RoutingRunResult,
     SearchOptions,
 )
 from inire.geometry.collision import RoutingWorld
@@ -34,6 +35,7 @@ _OBJECTIVE_FIELDS = set(ObjectiveWeights.__dataclass_fields__)
 ScenarioOutcome = tuple[float, int, int, int]
 ScenarioRun = Callable[[], ScenarioOutcome]
 ScenarioSnapshotRun = Callable[[], "ScenarioSnapshot"]
+TraceScenarioRun = Callable[[], RoutingRunResult]
 
 
 @dataclass(frozen=True, slots=True)
@@ -79,6 +81,19 @@ def _make_snapshot(
     )
 
 
+def _make_run_result(
+    results: dict[str, RoutingResult],
+    pathfinder: PathFinder,
+) -> RoutingRunResult:
+    return RoutingRunResult(
+        results_by_net=results,
+        metrics=pathfinder.metrics.snapshot(),
+        expanded_nodes=tuple(pathfinder.accumulated_expanded_nodes),
+        conflict_trace=tuple(pathfinder.conflict_trace),
+        frontier_trace=tuple(pathfinder.frontier_trace),
+    )
+
+
 def _sum_metrics(metrics_list: tuple[RouteMetrics, ...]) -> RouteMetrics:
     metric_names = RouteMetrics.__dataclass_fields__
     return RouteMetrics(
@@ -318,6 +333,24 @@ def run_example_05() -> ScenarioOutcome:
     return snapshot_example_05().as_outcome()
 
 
+def trace_example_05() -> RoutingRunResult:
+    netlist = {
+        "u_turn": (Port(50, 50, 0), Port(50, 70, 180)),
+        "loop": (Port(100, 100, 90), Port(100, 80, 270)),
+        "zig_zag": (Port(20, 150, 0), Port(180, 150, 0)),
+    }
+    widths = dict.fromkeys(netlist, 2.0)
+    _, _, _, pathfinder = _build_routing_stack(
+        bounds=(0, 0, 200, 200),
+        netlist=netlist,
+        widths=widths,
+        evaluator_kwargs={"bend_penalty": 50.0},
+        request_kwargs={"bend_radii": [20.0], "capture_conflict_trace": True, "capture_frontier_trace": True},
+    )
+    results = pathfinder.route_all()
+    return _make_run_result(results, pathfinder)
+
+
 def snapshot_example_06() -> ScenarioSnapshot:
     bounds = (-20, -20, 170, 170)
     obstacles = [
@@ -378,12 +411,40 @@ def run_example_06() -> ScenarioOutcome:
 
 
 def snapshot_example_07() -> ScenarioSnapshot:
+    return _snapshot_example_07_variant(
+        "example_07_large_scale_routing",
+        warm_start_enabled=True,
+    )
+
+
+def snapshot_example_07_no_warm_start() -> ScenarioSnapshot:
+    return _snapshot_example_07_variant(
+        "example_07_large_scale_routing_no_warm_start",
+        warm_start_enabled=False,
+    )
+
+
+def trace_example_07() -> RoutingRunResult:
+    return _trace_example_07_variant(warm_start_enabled=True)
+
+
+def trace_example_07_no_warm_start() -> RoutingRunResult:
+    return _trace_example_07_variant(warm_start_enabled=False)
+
+
+def _build_example_07_variant_stack(
+    *,
+    num_nets: int,
+    seed: int,
+    warm_start_enabled: bool,
+    capture_conflict_trace: bool = False,
+    capture_frontier_trace: bool = False,
+) -> tuple[CostEvaluator, AStarMetrics, PathFinder]:
     bounds = (0, 0, 1000, 1000)
     obstacles = [
         box(450, 0, 550, 400),
         box(450, 600, 550, 1000),
     ]
-    num_nets = 10
     start_x = 50
     start_y_base = 500 - (num_nets * 10.0) / 2.0
     end_x = 950
@@ -418,10 +479,31 @@ def snapshot_example_07() -> ScenarioSnapshot:
             "multiplier": 1.4,
             "net_order": "shortest",
             "capture_expanded": True,
+            "capture_conflict_trace": capture_conflict_trace,
+            "capture_frontier_trace": capture_frontier_trace,
             "shuffle_nets": True,
-            "seed": 42,
+            "seed": seed,
+            "warm_start_enabled": warm_start_enabled,
         },
     )
+    return evaluator, metrics, pathfinder
+
+
+def _run_example_07_variant(
+    *,
+    num_nets: int,
+    seed: int,
+    warm_start_enabled: bool,
+    capture_conflict_trace: bool = False,
+    capture_frontier_trace: bool = False,
+) -> RoutingRunResult:
+    evaluator, metrics, pathfinder = _build_example_07_variant_stack(
+        num_nets=num_nets,
+        seed=seed,
+        warm_start_enabled=warm_start_enabled,
+        capture_conflict_trace=capture_conflict_trace,
+        capture_frontier_trace=capture_frontier_trace,
+    )
 
     def iteration_callback(idx: int, current_results: dict[str, RoutingResult]) -> None:
         _ = current_results
@@ -429,10 +511,36 @@ def snapshot_example_07() -> ScenarioSnapshot:
         evaluator.greedy_h_weight = new_greedy
         metrics.reset_per_route()
 
-    t0 = perf_counter()
     results = pathfinder.route_all(iteration_callback=iteration_callback)
+    return _make_run_result(results, pathfinder)
+
+
+def _snapshot_example_07_variant(
+    name: str,
+    *,
+    warm_start_enabled: bool,
+) -> ScenarioSnapshot:
+    t0 = perf_counter()
+    run = _run_example_07_variant(
+        num_nets=10,
+        seed=42,
+        warm_start_enabled=warm_start_enabled,
+    )
     t1 = perf_counter()
-    return _make_snapshot("example_07_large_scale_routing", results, t1 - t0, pathfinder.metrics.snapshot())
+    return _make_snapshot(name, run.results_by_net, t1 - t0, run.metrics)
+
+
+def _trace_example_07_variant(
+    *,
+    warm_start_enabled: bool,
+) -> RoutingRunResult:
+    return _run_example_07_variant(
+        num_nets=10,
+        seed=42,
+        warm_start_enabled=warm_start_enabled,
+        capture_conflict_trace=True,
+        capture_frontier_trace=True,
+    )
 
 
 def run_example_07() -> ScenarioOutcome:
@@ -534,6 +642,19 @@ SCENARIO_SNAPSHOTS: tuple[tuple[str, ScenarioSnapshotRun], ...] = (
     ("example_09_unroutable_best_effort", snapshot_example_09),
 )
 
+PERFORMANCE_SCENARIO_SNAPSHOTS: tuple[tuple[str, ScenarioSnapshotRun], ...] = (
+    ("example_07_large_scale_routing_no_warm_start", snapshot_example_07_no_warm_start),
+)
+
+TRACE_SCENARIO_RUNS: tuple[tuple[str, TraceScenarioRun], ...] = (
+    ("example_05_orientation_stress", trace_example_05),
+    ("example_07_large_scale_routing", trace_example_07),
+)
+
+TRACE_PERFORMANCE_SCENARIO_RUNS: tuple[tuple[str, TraceScenarioRun], ...] = (
+    ("example_07_large_scale_routing_no_warm_start", trace_example_07_no_warm_start),
+)
+
 
 def capture_all_scenario_snapshots() -> tuple[ScenarioSnapshot, ...]:
     return tuple(run() for _, run in SCENARIO_SNAPSHOTS)

inire/tests/test_api.py

@@ -16,8 +16,12 @@ from inire import (
     route,
 )
 from inire.geometry.components import Straight
-
-
+from inire.geometry.collision import RoutingWorld
+from inire.results import RoutingReport, RoutingResult
+from inire.router._astar_types import AStarContext
+from inire.router._router import PathFinder, _IterationReview
+from inire.router.cost import CostEvaluator
+from inire.router.danger_map import DangerMap
 def test_root_module_exports_only_stable_surface() -> None:
     import inire
 
@@ -48,6 +52,8 @@ def test_route_problem_smoke() -> None:
 
     assert set(run.results_by_net) == {"net1"}
     assert run.results_by_net["net1"].is_valid
+    assert run.conflict_trace == ()
+    assert run.frontier_trace == ()
 
 
 def test_route_problem_supports_configs_and_debug_data() -> None:
@@ -77,14 +83,353 @@ def test_route_problem_supports_configs_and_debug_data() -> None:
     assert run.expanded_nodes
     assert run.metrics.nodes_expanded > 0
     assert run.metrics.route_iterations >= 1
+    assert run.metrics.iteration_reverify_calls >= 1
+    assert run.metrics.iteration_reverified_nets >= 0
+    assert run.metrics.iteration_conflicting_nets >= 0
+    assert run.metrics.iteration_conflict_edges >= 0
+    assert run.metrics.nets_carried_forward >= 0
     assert run.metrics.nets_routed >= 1
     assert run.metrics.move_cache_abs_misses >= 0
     assert run.metrics.ray_cast_calls >= 0
     assert run.metrics.dynamic_tree_rebuilds >= 0
+    assert run.metrics.visibility_corner_index_builds >= 0
     assert run.metrics.visibility_builds >= 0
+    assert run.metrics.guidance_match_moves >= 0
+    assert run.metrics.guidance_match_moves_straight >= 0
+    assert run.metrics.guidance_match_moves_bend90 >= 0
+    assert run.metrics.guidance_match_moves_sbend >= 0
+    assert run.metrics.guidance_bonus_applied >= 0.0
+    assert run.metrics.guidance_bonus_applied_straight >= 0.0
+    assert run.metrics.guidance_bonus_applied_bend90 >= 0.0
+    assert run.metrics.guidance_bonus_applied_sbend >= 0.0
+    assert run.metrics.congestion_grid_span_cache_hits >= 0
+    assert run.metrics.congestion_grid_span_cache_misses >= 0
+    assert run.metrics.congestion_presence_cache_hits >= 0
+    assert run.metrics.congestion_presence_cache_misses >= 0
+    assert run.metrics.congestion_presence_skips >= 0
+    assert run.metrics.congestion_candidate_precheck_hits >= 0
+    assert run.metrics.congestion_candidate_precheck_misses >= 0
+    assert run.metrics.congestion_candidate_precheck_skips >= 0
+    assert run.metrics.congestion_candidate_nets >= 0
+    assert run.metrics.congestion_net_envelope_cache_hits >= 0
+    assert run.metrics.congestion_net_envelope_cache_misses >= 0
+    assert run.metrics.congestion_grid_net_cache_hits >= 0
+    assert run.metrics.congestion_grid_net_cache_misses >= 0
+    assert run.metrics.congestion_lazy_resolutions >= 0
+    assert run.metrics.congestion_lazy_requeues >= 0
+    assert run.metrics.congestion_candidate_ids >= 0
+    assert run.metrics.verify_dynamic_candidate_nets >= 0
     assert run.metrics.verify_path_report_calls >= 0
+    assert run.metrics.pair_local_search_pairs_considered >= 0
+    assert run.metrics.pair_local_search_attempts >= 0
+    assert run.metrics.pair_local_search_accepts >= 0
+    assert run.metrics.pair_local_search_nodes_expanded >= 0
 
 
+def test_iteration_callback_observes_reverified_conflicts() -> None:
+    problem = RoutingProblem(
+        bounds=(0, 0, 100, 100),
+        nets=(
+            NetSpec("horizontal", Port(10, 50, 0), Port(90, 50, 0), width=2.0),
+            NetSpec("vertical", Port(50, 10, 90), Port(50, 90, 90), width=2.0),
+        ),
+    )
+    options = RoutingOptions(
+        congestion=CongestionOptions(max_iterations=1, warm_start_enabled=False),
+        refinement=RefinementOptions(enabled=False),
+    )
+    evaluator = CostEvaluator(RoutingWorld(clearance=2.0), DangerMap(bounds=problem.bounds))
+    pathfinder = PathFinder(AStarContext(evaluator, problem, options))
+    snapshots: list[dict[str, str]] = []
+
+    def callback(iteration: int, current_results: dict[str, object]) -> None:
+        _ = iteration
+        snapshots.append({net_id: result.outcome for net_id, result in current_results.items()})
+
+    results = pathfinder.route_all(iteration_callback=callback)
+
+    assert snapshots == [{"horizontal": "colliding", "vertical": "colliding"}]
+    assert results["horizontal"].outcome == "colliding"
+    assert results["vertical"].outcome == "colliding"
+
+
+def test_capture_conflict_trace_preserves_route_outputs() -> None:
+    problem = RoutingProblem(
+        bounds=(0, 0, 100, 100),
+        nets=(
+            NetSpec("horizontal", Port(10, 50, 0), Port(90, 50, 0), width=2.0),
+            NetSpec("vertical", Port(50, 10, 90), Port(50, 90, 90), width=2.0),
+        ),
+    )
+    base_options = RoutingOptions(
+        congestion=CongestionOptions(max_iterations=1, warm_start_enabled=False),
+        refinement=RefinementOptions(enabled=False),
+    )
+
+    run_without_trace = route(problem, options=base_options)
+    run_with_trace = route(
+        problem,
+        options=RoutingOptions(
+            congestion=base_options.congestion,
+            refinement=base_options.refinement,
+            diagnostics=DiagnosticsOptions(capture_conflict_trace=True),
+        ),
+    )
+
+    assert {net_id: result.outcome for net_id, result in run_without_trace.results_by_net.items()} == {
+        net_id: result.outcome for net_id, result in run_with_trace.results_by_net.items()
+    }
+    assert [entry.stage for entry in run_with_trace.conflict_trace] == ["iteration", "restored_best", "final"]
+
+
+def test_capture_conflict_trace_records_component_pairs() -> None:
+    problem = RoutingProblem(
+        bounds=(0, 0, 100, 100),
+        nets=(
+            NetSpec("horizontal", Port(10, 50, 0), Port(90, 50, 0), width=2.0),
+            NetSpec("vertical", Port(50, 10, 90), Port(50, 90, 90), width=2.0),
+        ),
+    )
+    options = RoutingOptions(
+        congestion=CongestionOptions(max_iterations=1, warm_start_enabled=False),
+        refinement=RefinementOptions(enabled=False),
+        diagnostics=DiagnosticsOptions(capture_conflict_trace=True),
+    )
+
+    run = route(problem, options=options)
+    final_entry = run.conflict_trace[-1]
+    trace_by_net = {net.net_id: net for net in final_entry.nets}
+
+    assert final_entry.stage == "final"
+    assert final_entry.conflict_edges == (("horizontal", "vertical"),)
+    assert trace_by_net["horizontal"].component_conflicts[0].other_net_id == "vertical"
+    assert trace_by_net["horizontal"].component_conflicts[0].self_component_index == 0
+    assert trace_by_net["horizontal"].component_conflicts[0].other_component_index == 0
+    assert trace_by_net["vertical"].component_conflicts[0].other_net_id == "horizontal"
+    assert trace_by_net["vertical"].component_conflicts[0].self_component_index == 0
+    assert trace_by_net["vertical"].component_conflicts[0].other_component_index == 0
+
+
+def test_capture_frontier_trace_preserves_route_outputs() -> None:
+    problem = RoutingProblem(
+        bounds=(0, 0, 100, 100),
+        nets=(
+            NetSpec("horizontal", Port(10, 50, 0), Port(90, 50, 0), width=2.0),
+            NetSpec("vertical", Port(50, 10, 90), Port(50, 90, 90), width=2.0),
+        ),
+    )
+    base_options = RoutingOptions(
+        congestion=CongestionOptions(max_iterations=1, warm_start_enabled=False),
+        refinement=RefinementOptions(enabled=False),
+    )
+
+    run_without_trace = route(problem, options=base_options)
+    run_with_trace = route(
+        problem,
+        options=RoutingOptions(
+            congestion=base_options.congestion,
+            refinement=base_options.refinement,
+            diagnostics=DiagnosticsOptions(capture_frontier_trace=True),
+        ),
+    )
+
+    assert {net_id: result.outcome for net_id, result in run_without_trace.results_by_net.items()} == {
+        net_id: result.outcome for net_id, result in run_with_trace.results_by_net.items()
+    }
+    assert {trace.net_id for trace in run_with_trace.frontier_trace} == {"horizontal", "vertical"}
+
+
+def test_capture_frontier_trace_records_prune_reasons() -> None:
+    problem = RoutingProblem(
+        bounds=(0, 0, 100, 100),
+        nets=(
+            NetSpec("horizontal", Port(10, 50, 0), Port(90, 50, 0), width=2.0),
+            NetSpec("vertical", Port(50, 10, 90), Port(50, 90, 90), width=2.0),
+        ),
+    )
+    run = route(
+        problem,
+        options=RoutingOptions(
+            congestion=CongestionOptions(max_iterations=1, warm_start_enabled=False),
+            refinement=RefinementOptions(enabled=False),
+            diagnostics=DiagnosticsOptions(capture_frontier_trace=True),
+        ),
+    )
+
+    trace_by_net = {entry.net_id: entry for entry in run.frontier_trace}
+    assert trace_by_net["horizontal"].hotspot_bounds
+    assert (
+        trace_by_net["horizontal"].pruned_closed_set
+        + trace_by_net["horizontal"].pruned_hard_collision
+        + trace_by_net["horizontal"].pruned_self_collision
+        + trace_by_net["horizontal"].pruned_cost
+    ) > 0
+    assert trace_by_net["horizontal"].samples
+
+
+def test_reverify_iterations_stop_early_on_stalled_conflict_graph() -> None:
+    problem = RoutingProblem(
+        bounds=(0, 0, 100, 100),
+        nets=(
+            NetSpec("horizontal", Port(10, 50, 0), Port(90, 50, 0), width=2.0),
+            NetSpec("vertical", Port(50, 10, 90), Port(50, 90, 90), width=2.0),
+        ),
+    )
+    options = RoutingOptions(
+        congestion=CongestionOptions(max_iterations=10, warm_start_enabled=False),
+        refinement=RefinementOptions(enabled=False),
+    )
+
+    run = route(problem, options=options)
+
+    assert run.metrics.route_iterations < 10
+
+
+def test_route_all_restores_best_iteration_snapshot(monkeypatch: pytest.MonkeyPatch) -> None:
+    problem = RoutingProblem(
+        bounds=(0, 0, 100, 100),
+        nets=(
+            NetSpec("netA", Port(10, 50, 0), Port(90, 50, 0), width=2.0),
+            NetSpec("netB", Port(50, 10, 90), Port(50, 90, 90), width=2.0),
+        ),
+    )
+    options = RoutingOptions(
+        congestion=CongestionOptions(max_iterations=2, warm_start_enabled=False),
+        refinement=RefinementOptions(enabled=False),
+    )
+    evaluator = CostEvaluator(RoutingWorld(clearance=2.0), DangerMap(bounds=problem.bounds))
+    pathfinder = PathFinder(AStarContext(evaluator, problem, options))
+    best_result = RoutingResult(
+        net_id="netA",
+        path=(Straight.generate(Port(10, 50, 0), 80.0, 2.0, dilation=1.0),),
+        reached_target=True,
+        report=RoutingReport(),
+    )
+    missing_result = RoutingResult(net_id="netA", path=(), reached_target=False)
+    unroutable_b = RoutingResult(net_id="netB", path=(), reached_target=False)
+
+    def fake_run_iteration(self, state, iteration, reroute_net_ids, iteration_callback):
+        _ = self
+        _ = reroute_net_ids
+        _ = iteration_callback
+        if iteration == 0:
+            state.results = {"netA": best_result, "netB": unroutable_b}
+            return _IterationReview(
+                conflicting_nets={"netA", "netB"},
+                conflict_edges={("netA", "netB")},
+                completed_net_ids={"netA"},
+                total_dynamic_collisions=1,
+            )
+        state.results = {"netA": missing_result, "netB": unroutable_b}
+        return _IterationReview(
+            conflicting_nets={"netA", "netB"},
+            conflict_edges={("netA", "netB")},
+            completed_net_ids=set(),
+            total_dynamic_collisions=2,
+        )
+
+    monkeypatch.setattr(PathFinder, "_run_iteration", fake_run_iteration)
+    monkeypatch.setattr(PathFinder, "_verify_results", lambda self, state: dict(state.results))
+
+    results = pathfinder.route_all()
+
+    assert results["netA"].outcome == "completed"
+    assert results["netB"].outcome == "unroutable"
+
+
+def test_route_all_restores_best_iteration_snapshot_on_timeout(monkeypatch: pytest.MonkeyPatch) -> None:
+    problem = RoutingProblem(
+        bounds=(0, 0, 100, 100),
+        nets=(NetSpec("netA", Port(10, 50, 0), Port(90, 50, 0), width=2.0),),
+    )
+    options = RoutingOptions(
+        congestion=CongestionOptions(max_iterations=2, warm_start_enabled=False),
+        refinement=RefinementOptions(enabled=False),
+    )
+    evaluator = CostEvaluator(RoutingWorld(clearance=2.0), DangerMap(bounds=problem.bounds))
+    pathfinder = PathFinder(AStarContext(evaluator, problem, options))
+    best_result = RoutingResult(
+        net_id="netA",
+        path=(Straight.generate(Port(10, 50, 0), 80.0, 2.0, dilation=1.0),),
+        reached_target=True,
+        report=RoutingReport(),
+    )
+    worse_result = RoutingResult(net_id="netA", path=(), reached_target=False)
+
+    def fake_run_iterations(self, state, iteration_callback):
+        _ = iteration_callback
+        _ = self
+        state.results = {"netA": best_result}
+        pathfinder._update_best_iteration(
+            state,
+            _IterationReview(
+                conflicting_nets=set(),
+                conflict_edges=set(),
+                completed_net_ids={"netA"},
+                total_dynamic_collisions=0,
+            ),
+        )
+        state.results = {"netA": worse_result}
+        return True
+
+    monkeypatch.setattr(PathFinder, "_run_iterations", fake_run_iterations)
+    monkeypatch.setattr(PathFinder, "_verify_results", lambda self, state: dict(state.results))
+
+    results = pathfinder.route_all()
+
+    assert results["netA"].outcome == "completed"
+
+
+def test_capture_conflict_trace_records_restored_best_stage(monkeypatch: pytest.MonkeyPatch) -> None:
+    problem = RoutingProblem(
+        bounds=(0, 0, 100, 100),
+        nets=(NetSpec("netA", Port(10, 50, 0), Port(90, 50, 0), width=2.0),),
+    )
+    options = RoutingOptions(
+        congestion=CongestionOptions(max_iterations=2, warm_start_enabled=False),
+        refinement=RefinementOptions(enabled=False),
+        diagnostics=DiagnosticsOptions(capture_conflict_trace=True),
+    )
+    evaluator = CostEvaluator(RoutingWorld(clearance=2.0), DangerMap(bounds=problem.bounds))
+    pathfinder = PathFinder(AStarContext(evaluator, problem, options))
+    best_result = RoutingResult(
+        net_id="netA",
+        path=(Straight.generate(Port(10, 50, 0), 80.0, 2.0, dilation=1.0),),
+        reached_target=True,
+        report=RoutingReport(),
+    )
+    worse_result = RoutingResult(net_id="netA", path=(), reached_target=False)
+
+    def fake_run_iteration(self, state, iteration, reroute_net_ids, iteration_callback):
+        _ = self
+        _ = reroute_net_ids
+        _ = iteration_callback
+        if iteration == 0:
+            state.results = {"netA": best_result}
+            return _IterationReview(
+                conflicting_nets=set(),
+                conflict_edges=set(),
+                completed_net_ids={"netA"},
+                total_dynamic_collisions=0,
+            )
+        state.results = {"netA": worse_result}
+        return _IterationReview(
+            conflicting_nets=set(),
+            conflict_edges=set(),
+            completed_net_ids=set(),
+            total_dynamic_collisions=0,
+        )
+
+    monkeypatch.setattr(PathFinder, "_run_iteration", fake_run_iteration)
+
+    pathfinder.route_all()
+
+    assert [entry.stage for entry in pathfinder.conflict_trace] == [
+        "restored_best",
+        "final",
+    ]
+    restored_entry = pathfinder.conflict_trace[0]
+    assert restored_entry.nets[0].outcome == "completed"
 def test_route_problem_locked_routes_become_static_obstacles() -> None:
     locked = (Straight.generate(Port(10, 50, 0), 80.0, 2.0, dilation=1.0),)
     problem = RoutingProblem(

inire/tests/test_astar.py

@@ -1,16 +1,22 @@
 import math
-
 import pytest
 from shapely.geometry import Polygon
 
-from inire import RoutingProblem, RoutingOptions, RoutingResult, SearchOptions
+from inire import CongestionOptions, NetSpec, RoutingProblem, RoutingOptions, RoutingResult, SearchOptions
 from inire.geometry.components import Bend90, Straight
 from inire.geometry.collision import RoutingWorld
 from inire.geometry.primitives import Port
-from inire.router._astar_types import AStarContext, SearchRunConfig
+from inire.router._astar_types import AStarContext, AStarNode, SearchRunConfig
+from inire.router._astar_admission import add_node
+from inire.router._astar_moves import (
+    _distance_to_bounds_in_heading,
+    _should_cap_straights_to_bounds,
+)
+from inire.router._router import PathFinder, _RoutingState
 from inire.router._search import route_astar
 from inire.router.cost import CostEvaluator
 from inire.router.danger_map import DangerMap
+from inire.seeds import StraightSeed
 
 BOUNDS = (0, -50, 150, 150)
 
@@ -214,6 +220,84 @@ def test_astar_context_keeps_evaluator_weights_separate(basic_evaluator: CostEva
     assert basic_evaluator.h_manhattan(Port(0, 0, 0), Port(10, 10, 0)) > 0.0
 
 
+def test_distance_to_bounds_in_heading_is_directional() -> None:
+    bounds = (0, 0, 100, 200)
+
+    assert _distance_to_bounds_in_heading(Port(20, 30, 0), bounds) == pytest.approx(80.0)
+    assert _distance_to_bounds_in_heading(Port(20, 30, 90), bounds) == pytest.approx(170.0)
+    assert _distance_to_bounds_in_heading(Port(20, 30, 180), bounds) == pytest.approx(20.0)
+    assert _distance_to_bounds_in_heading(Port(20, 30, 270), bounds) == pytest.approx(30.0)
+
+
+def test_should_cap_straights_to_bounds_only_for_large_no_warm_runs(basic_evaluator: CostEvaluator) -> None:
+    large_context = AStarContext(
+        basic_evaluator,
+        RoutingProblem(
+            bounds=(0, 0, 1000, 1000),
+            nets=tuple(
+                NetSpec(f"net{i}", Port(0, i * 10, 0), Port(10, i * 10, 0), width=2.0)
+                for i in range(8)
+            ),
+        ),
+        RoutingOptions(
+            congestion=CongestionOptions(warm_start_enabled=False),
+        ),
+    )
+    small_context = _build_context(basic_evaluator, bounds=BOUNDS)
+
+    assert _should_cap_straights_to_bounds(large_context)
+    assert not _should_cap_straights_to_bounds(small_context)
+
+
+def test_pair_local_context_clones_live_static_obstacles() -> None:
+    obstacle = Polygon([(20, -20), (40, -20), (40, 20), (20, 20)])
+    engine = RoutingWorld(clearance=2.0)
+    engine.add_static_obstacle(obstacle)
+    danger_map = DangerMap(bounds=BOUNDS)
+    danger_map.precompute([obstacle])
+    evaluator = CostEvaluator(engine, danger_map, bend_penalty=50.0, sbend_penalty=150.0)
+    finder = PathFinder(
+        AStarContext(
+            evaluator,
+            RoutingProblem(
+                bounds=BOUNDS,
+                nets=(
+                    NetSpec("pair_a", Port(0, 0, 0), Port(60, 0, 0), width=2.0),
+                    NetSpec("pair_b", Port(0, 10, 0), Port(60, 10, 0), width=2.0),
+                ),
+            ),
+            RoutingOptions(),
+        )
+    )
+    state = _RoutingState(
+        net_specs={
+            "pair_a": NetSpec("pair_a", Port(0, 0, 0), Port(60, 0, 0), width=2.0),
+            "pair_b": NetSpec("pair_b", Port(0, 10, 0), Port(60, 10, 0), width=2.0),
+        },
+        ordered_net_ids=["pair_a", "pair_b"],
+        results={},
+        needs_self_collision_check=set(),
+        start_time=0.0,
+        timeout_s=1.0,
+        initial_paths=None,
+        accumulated_expanded_nodes=[],
+        best_results={},
+        best_completed_nets=-1,
+        best_conflict_edges=10**9,
+        best_dynamic_collisions=10**9,
+        last_conflict_signature=(),
+        last_conflict_edge_count=0,
+        repeated_conflict_count=0,
+    )
+
+    local_context = finder._build_pair_local_context(state, {}, ("pair_a", "pair_b"))
+
+    assert finder.context.problem.static_obstacles == ()
+    assert len(local_context.problem.static_obstacles) == 1
+    assert len(local_context.cost_evaluator.collision_engine._static_obstacles.geometries) == 1
+    assert next(iter(local_context.problem.static_obstacles)).equals(obstacle)
+
+
 def test_route_astar_bend_collision_override_does_not_persist(basic_evaluator: CostEvaluator) -> None:
     context = _build_context(basic_evaluator, bounds=BOUNDS, bend_radii=(10.0,), bend_collision_type="arc")
 
@@ -301,6 +385,27 @@ def test_route_astar_supports_all_visibility_guidance_modes(
     assert validation["connectivity_ok"]
 
 
+def test_tangent_corner_mode_avoids_exact_visibility_graph_builds(basic_evaluator: CostEvaluator) -> None:
+    obstacle = Polygon([(30, 10), (50, 10), (50, 40), (30, 40)])
+    basic_evaluator.collision_engine.add_static_obstacle(obstacle)
+    basic_evaluator.danger_map.precompute([obstacle])
+    context = _build_context(
+        basic_evaluator,
+        bounds=BOUNDS,
+        bend_radii=(10.0,),
+        sbend_radii=(),
+        max_straight_length=150.0,
+        visibility_guidance="tangent_corner",
+    )
+
+    path = _route(context, Port(0, 0, 0), Port(80, 50, 0))
+
+    assert path is not None
+    assert context.metrics.total_visibility_builds == 0
+    assert context.metrics.total_visibility_corner_pairs_checked == 0
+    assert context.metrics.total_ray_cast_calls_visibility_build == 0
+
+
 def test_route_astar_repeated_searches_succeed_with_small_cache_limit(basic_evaluator: CostEvaluator) -> None:
     context = AStarContext(
         basic_evaluator,
@@ -318,3 +423,244 @@ def test_route_astar_repeated_searches_succeed_with_small_cache_limit(basic_eval
         path = _route(context, start, target)
         assert path is not None
         assert path[-1].end_port == target
+
+
+def test_self_collision_prunes_before_congestion_check(basic_evaluator: CostEvaluator) -> None:
+    context = _build_context(basic_evaluator, bounds=BOUNDS)
+    root = AStarNode(Port(0, 0, 0), 0.0, 0.0)
+    parent_result = Straight.generate(Port(0, 0, 0), 10.0, width=2.0, dilation=1.0)
+    parent = AStarNode(parent_result.end_port, g_cost=10.0, h_cost=0.0, parent=root, component_result=parent_result)
+    open_set: list[AStarNode] = []
+    closed_set: dict[tuple[int, int, int], float] = {}
+
+    add_node(
+        parent,
+        parent_result,
+        target=Port(20, 0, 0),
+        net_width=2.0,
+        net_id="netA",
+        open_set=open_set,
+        closed_set=closed_set,
+        context=context,
+        metrics=context.metrics,
+        congestion_cache={},
+        congestion_presence_cache={},
+        congestion_candidate_precheck_cache={},
+        congestion_net_envelope_cache={},
+        congestion_grid_net_cache={},
+        congestion_grid_span_cache={},
+        config=SearchRunConfig.from_options(context.options, self_collision_check=True),
+        move_type="straight",
+        cache_key=("overlap",),
+    )
+
+    assert not open_set
+    assert context.metrics.total_congestion_check_calls == 0
+    assert context.metrics.total_congestion_cache_misses == 0
+
+
+def test_closed_set_dominance_prunes_before_congestion_check(basic_evaluator: CostEvaluator) -> None:
+    context = _build_context(basic_evaluator, bounds=BOUNDS)
+    root = AStarNode(Port(0, 0, 0), 0.0, 0.0)
+    result = Straight.generate(Port(0, 0, 0), 10.0, width=2.0, dilation=1.0)
+    open_set: list[AStarNode] = []
+    closed_set = {result.end_port.as_tuple(): context.cost_evaluator.score_component(result, start_port=root.port)}
+
+    add_node(
+        root,
+        result,
+        target=Port(20, 0, 0),
+        net_width=2.0,
+        net_id="netA",
+        open_set=open_set,
+        closed_set=closed_set,
+        context=context,
+        metrics=context.metrics,
+        congestion_cache={},
+        congestion_presence_cache={},
+        congestion_candidate_precheck_cache={},
+        congestion_net_envelope_cache={},
+        congestion_grid_net_cache={},
+        congestion_grid_span_cache={},
+        config=SearchRunConfig.from_options(context.options),
+        move_type="straight",
+        cache_key=("dominated",),
+    )
+
+    assert not open_set
+    assert context.metrics.total_congestion_check_calls == 0
+    assert context.metrics.total_congestion_cache_misses == 0
+
+
+def test_no_dynamic_paths_skips_congestion_check(basic_evaluator: CostEvaluator) -> None:
+    context = _build_context(basic_evaluator, bounds=BOUNDS)
+    root = AStarNode(Port(0, 0, 0), 0.0, 0.0)
+    result = Straight.generate(Port(0, 0, 0), 10.0, width=2.0, dilation=1.0)
+    open_set: list[AStarNode] = []
+    closed_set: dict[tuple[int, int, int], float] = {}
+
+    add_node(
+        root,
+        result,
+        target=Port(20, 0, 0),
+        net_width=2.0,
+        net_id="netA",
+        open_set=open_set,
+        closed_set=closed_set,
+        context=context,
+        metrics=context.metrics,
+        congestion_cache={},
+        congestion_presence_cache={},
+        congestion_candidate_precheck_cache={},
+        congestion_net_envelope_cache={},
+        congestion_grid_net_cache={},
+        congestion_grid_span_cache={},
+        config=SearchRunConfig.from_options(context.options),
+        move_type="straight",
+        cache_key=("no-dynamic",),
+    )
+
+    assert open_set
+    assert context.metrics.total_congestion_check_calls == 0
+    assert context.metrics.total_congestion_cache_misses == 0
+
+
+def test_guidance_seed_matching_move_reduces_cost_and_advances_seed_index(
+    basic_evaluator: CostEvaluator,
+) -> None:
+    context = _build_context(basic_evaluator, bounds=BOUNDS)
+    root = AStarNode(Port(0, 0, 0), 0.0, 0.0, seed_index=0)
+    result = Straight.generate(Port(0, 0, 0), 10.0, width=2.0, dilation=1.0)
+    open_set: list[AStarNode] = []
+    unguided_open_set: list[AStarNode] = []
+    closed_set: dict[tuple[int, int, int], float] = {}
+
+    add_node(
+        root,
+        result,
+        target=Port(20, 0, 0),
+        net_width=2.0,
+        net_id="netA",
+        open_set=open_set,
+        closed_set=closed_set,
+        context=context,
+        metrics=context.metrics,
+        congestion_cache={},
+        congestion_presence_cache={},
+        congestion_candidate_precheck_cache={},
+        congestion_net_envelope_cache={},
+        congestion_grid_net_cache={},
+        congestion_grid_span_cache={},
+        config=SearchRunConfig.from_options(
+            context.options,
+            guidance_seed=(StraightSeed(length=10.0),),
+            guidance_bonus=5.0,
+        ),
+        move_type="straight",
+        cache_key=("guided",),
+    )
+    add_node(
+        AStarNode(Port(0, 0, 0), 0.0, 0.0),
+        result,
+        target=Port(20, 0, 0),
+        net_width=2.0,
+        net_id="netA",
+        open_set=unguided_open_set,
+        closed_set={},
+        context=context,
+        metrics=context.metrics,
+        congestion_cache={},
+        congestion_presence_cache={},
+        congestion_candidate_precheck_cache={},
+        congestion_net_envelope_cache={},
+        congestion_grid_net_cache={},
+        congestion_grid_span_cache={},
+        config=SearchRunConfig.from_options(context.options),
+        move_type="straight",
+        cache_key=("unguided",),
+    )
+
+    assert open_set
+    assert unguided_open_set
+    guided_node = open_set[0]
+    unguided_node = unguided_open_set[0]
+    assert guided_node.seed_index == 1
+    assert guided_node.g_cost < unguided_node.g_cost
+    assert context.metrics.total_guidance_match_moves == 1
+    assert context.metrics.total_guidance_match_moves_straight == 1
+    assert context.metrics.total_guidance_match_moves_bend90 == 0
+    assert context.metrics.total_guidance_match_moves_sbend == 0
+    assert context.metrics.total_guidance_bonus_applied == pytest.approx(5.0)
+    assert context.metrics.total_guidance_bonus_applied_straight == pytest.approx(5.0)
+    assert context.metrics.total_guidance_bonus_applied_bend90 == pytest.approx(0.0)
+    assert context.metrics.total_guidance_bonus_applied_sbend == pytest.approx(0.0)
+
+
+def test_guidance_seed_bend90_keeps_full_bonus(
+    basic_evaluator: CostEvaluator,
+) -> None:
+    context = _build_context(basic_evaluator, bounds=BOUNDS)
+    root = AStarNode(Port(0, 0, 0), 0.0, 0.0, seed_index=0)
+    result = Bend90.generate(Port(0, 0, 0), 10.0, width=2.0, direction="CCW", dilation=1.0)
+    open_set: list[AStarNode] = []
+    unguided_open_set: list[AStarNode] = []
+
+    add_node(
+        root,
+        result,
+        target=Port(10, 10, 90),
+        net_width=2.0,
+        net_id="netA",
+        open_set=open_set,
+        closed_set={},
+        context=context,
+        metrics=context.metrics,
+        congestion_cache={},
+        congestion_presence_cache={},
+        congestion_candidate_precheck_cache={},
+        congestion_net_envelope_cache={},
+        congestion_grid_net_cache={},
+        congestion_grid_span_cache={},
+        config=SearchRunConfig.from_options(
+            context.options,
+            guidance_seed=(result.move_spec,),
+            guidance_bonus=5.0,
+        ),
+        move_type="bend90",
+        cache_key=("guided-bend90",),
+    )
+    add_node(
+        AStarNode(Port(0, 0, 0), 0.0, 0.0),
+        result,
+        target=Port(10, 10, 90),
+        net_width=2.0,
+        net_id="netA",
+        open_set=unguided_open_set,
+        closed_set={},
+        context=context,
+        metrics=context.metrics,
+        congestion_cache={},
+        congestion_presence_cache={},
+        congestion_candidate_precheck_cache={},
+        congestion_net_envelope_cache={},
+        congestion_grid_net_cache={},
+        congestion_grid_span_cache={},
+        config=SearchRunConfig.from_options(context.options),
+        move_type="bend90",
+        cache_key=("unguided-bend90",),
+    )
+
+    assert open_set
+    assert unguided_open_set
+    guided_node = open_set[0]
+    unguided_node = unguided_open_set[0]
+    assert guided_node.seed_index == 1
+    assert unguided_node.g_cost - guided_node.g_cost == pytest.approx(5.0)
+    assert context.metrics.total_guidance_match_moves == 1
+    assert context.metrics.total_guidance_match_moves_straight == 0
+    assert context.metrics.total_guidance_match_moves_bend90 == 1
+    assert context.metrics.total_guidance_match_moves_sbend == 0
+    assert context.metrics.total_guidance_bonus_applied == pytest.approx(5.0)
+    assert context.metrics.total_guidance_bonus_applied_straight == pytest.approx(0.0)
+    assert context.metrics.total_guidance_bonus_applied_bend90 == pytest.approx(5.0)
+    assert context.metrics.total_guidance_bonus_applied_sbend == pytest.approx(0.0)

inire/tests/test_collision.py

@@ -1,6 +1,11 @@
+from shapely.geometry import box
+
 from inire.geometry.collision import RoutingWorld
+from inire.geometry.components import ComponentResult
 from inire.geometry.components import Straight
 from inire.geometry.primitives import Port
+from inire.router._astar_types import AStarMetrics
+from inire.seeds import StraightSeed
 
 
 def _install_static_straight(
@@ -82,6 +87,7 @@ def test_check_move_static_clearance() -> None:
 
 def test_verify_path_report_preserves_long_net_id() -> None:
     engine = RoutingWorld(clearance=2.0)
+    engine.metrics = AStarMetrics()
     net_id = "net_abcdefghijklmnopqrstuvwxyz_0123456789"
     path = [Straight.generate(Port(0, 0, 0), 20.0, width=2.0, dilation=1.0)]
     geoms = [poly for component in path for poly in component.collision_geometry]
@@ -91,10 +97,12 @@ def test_verify_path_report_preserves_long_net_id() -> None:
     report = engine.verify_path_report(net_id, path)
 
     assert report.dynamic_collision_count == 0
+    assert engine.metrics.total_verify_dynamic_candidate_nets == 0
 
 
 def test_verify_path_report_distinguishes_long_net_ids_with_shared_prefix() -> None:
     engine = RoutingWorld(clearance=2.0)
+    engine.metrics = AStarMetrics()
     shared_prefix = "net_shared_prefix_abcdefghijklmnopqrstuvwxyz_"
     net_a = f"{shared_prefix}A"
     net_b = f"{shared_prefix}B"
@@ -115,6 +123,111 @@ def test_verify_path_report_distinguishes_long_net_ids_with_shared_prefix() -> N
     report = engine.verify_path_report(net_a, path_a)
 
     assert report.dynamic_collision_count == 1
+    assert engine.metrics.total_verify_dynamic_candidate_nets == 1
+
+
+def test_verify_path_report_uses_net_envelopes_before_dynamic_object_scan() -> None:
+    engine = RoutingWorld(clearance=2.0)
+    engine.metrics = AStarMetrics()
+    path_a = [Straight.generate(Port(0, 0, 0), 20.0, width=2.0, dilation=1.0)]
+    path_b = [Straight.generate(Port(0, 0, 0), 20.0, width=2.0, dilation=1.0)]
+    path_far = [Straight.generate(Port(100, 100, 0), 20.0, width=2.0, dilation=1.0)]
+
+    engine.add_path(
+        "netB",
+        [poly for component in path_b for poly in component.collision_geometry],
+        dilated_geometry=[poly for component in path_b for poly in component.dilated_collision_geometry],
+    )
+    engine.add_path(
+        "netFar",
+        [poly for component in path_far for poly in component.collision_geometry],
+        dilated_geometry=[poly for component in path_far for poly in component.dilated_collision_geometry],
+    )
+
+    report = engine.verify_path_report("netA", path_a)
+
+    assert report.dynamic_collision_count == 1
+    assert engine.metrics.total_verify_dynamic_candidate_nets == 1
+    assert engine.metrics.total_verify_dynamic_exact_pair_checks == 1
+
+
+def test_verify_path_details_returns_conflicting_net_ids() -> None:
+    engine = RoutingWorld(clearance=2.0)
+    engine.metrics = AStarMetrics()
+    path_a = [Straight.generate(Port(0, 0, 0), 20.0, width=2.0, dilation=1.0)]
+    path_b = [Straight.generate(Port(0, 0, 0), 20.0, width=2.0, dilation=1.0)]
+
+    engine.add_path(
+        "netB",
+        [poly for component in path_b for poly in component.collision_geometry],
+        dilated_geometry=[poly for component in path_b for poly in component.dilated_collision_geometry],
+    )
+
+    detail = engine.verify_path_details("netA", path_a)
+
+    assert detail.report.dynamic_collision_count == 1
+    assert detail.conflicting_net_ids == ("netB",)
+
+
+def test_verify_path_details_reports_component_conflicts() -> None:
+    engine = RoutingWorld(clearance=2.0)
+    engine.metrics = AStarMetrics()
+    path_a = [Straight.generate(Port(0, 0, 0), 20.0, width=2.0, dilation=1.0)]
+    path_b = [
+        Straight.generate(Port(100, 0, 0), 10.0, width=2.0, dilation=1.0),
+        Straight.generate(Port(0, 0, 0), 20.0, width=2.0, dilation=1.0),
+    ]
+
+    engine.add_path(
+        "netB",
+        [poly for component in path_b for poly in component.collision_geometry],
+        dilated_geometry=[poly for component in path_b for poly in component.dilated_collision_geometry],
+        component_indexes=[0] * len(path_b[0].collision_geometry) + [1] * len(path_b[1].collision_geometry),
+    )
+
+    detail = engine.verify_path_details("netA", path_a, capture_component_conflicts=True)
+
+    assert detail.conflicting_net_ids == ("netB",)
+    assert detail.component_conflicts == ((0, "netB", 1),)
+
+
+def test_verify_path_details_deduplicates_component_conflicts() -> None:
+    engine = RoutingWorld(clearance=2.0)
+    engine.metrics = AStarMetrics()
+    query_component = ComponentResult(
+        start_port=Port(0, 0, 0),
+        collision_geometry=[box(0, 0, 10, 10), box(12, 0, 22, 10)],
+        end_port=Port(22, 0, 0),
+        length=22.0,
+        move_type="straight",
+        move_spec=StraightSeed(22.0),
+        physical_geometry=[box(0, 0, 10, 10), box(12, 0, 22, 10)],
+        dilated_collision_geometry=[box(0, 0, 10, 10), box(12, 0, 22, 10)],
+        dilated_physical_geometry=[box(0, 0, 10, 10), box(12, 0, 22, 10)],
+    )
+    blocker_component = ComponentResult(
+        start_port=Port(0, 0, 0),
+        collision_geometry=[box(5, 0, 17, 10)],
+        end_port=Port(17, 0, 0),
+        length=12.0,
+        move_type="straight",
+        move_spec=StraightSeed(12.0),
+        physical_geometry=[box(5, 0, 17, 10)],
+        dilated_collision_geometry=[box(5, 0, 17, 10)],
+        dilated_physical_geometry=[box(5, 0, 17, 10)],
+    )
+
+    engine.add_path(
+        "netB",
+        blocker_component.collision_geometry,
+        dilated_geometry=blocker_component.dilated_collision_geometry,
+        component_indexes=[0],
+    )
+
+    detail = engine.verify_path_details("netA", [query_component], capture_component_conflicts=True)
+
+    assert detail.conflicting_net_ids == ("netB",)
+    assert detail.component_conflicts == ((0, "netB", 0),)
 
 
 def test_remove_path_clears_dynamic_path() -> None:
@@ -129,3 +242,247 @@ def test_remove_path_clears_dynamic_path() -> None:
     engine.remove_path("netA")
     assert list(engine._dynamic_paths.geometries.values()) == []
     assert len(engine._static_obstacles.geometries) == 0
+
+
+def test_dynamic_grid_updates_incrementally_on_add_and_remove() -> None:
+    engine = RoutingWorld(clearance=2.0)
+    engine.metrics = AStarMetrics()
+    path_a = [Straight.generate(Port(0, 0, 0), 20.0, width=2.0, dilation=1.0)]
+    path_b = [Straight.generate(Port(0, 4, 0), 20.0, width=2.0, dilation=1.0)]
+
+    engine.add_path(
+        "netA",
+        [poly for component in path_a for poly in component.collision_geometry],
+        dilated_geometry=[poly for component in path_a for poly in component.dilated_collision_geometry],
+    )
+    engine.add_path(
+        "netB",
+        [poly for component in path_b for poly in component.collision_geometry],
+        dilated_geometry=[poly for component in path_b for poly in component.dilated_collision_geometry],
+    )
+
+    dynamic_paths = engine._dynamic_paths
+    assert dynamic_paths.net_to_obj_ids["netA"]
+    assert dynamic_paths.net_to_obj_ids["netB"]
+    assert dynamic_paths.grid
+    assert engine.metrics.total_dynamic_grid_rebuilds == 0
+
+    engine.remove_path("netA")
+
+    assert "netA" not in dynamic_paths.net_to_obj_ids
+    assert "netB" in dynamic_paths.net_to_obj_ids
+    assert engine.metrics.total_dynamic_grid_rebuilds == 0
+    assert "netA" not in dynamic_paths.grid_net_obj_ids.get((0, -1), {})
+    assert "netB" in dynamic_paths.grid_net_obj_ids.get((0, 0), {})
+
+
+def test_dynamic_net_envelopes_update_incrementally_on_add_and_remove() -> None:
+    engine = RoutingWorld(clearance=2.0)
+    engine.metrics = AStarMetrics()
+    path_a = [Straight.generate(Port(0, 0, 0), 20.0, width=2.0, dilation=1.0)]
+    path_b = [Straight.generate(Port(0, 40, 0), 10.0, width=2.0, dilation=1.0)]
+
+    engine.add_path(
+        "netA",
+        [poly for component in path_a for poly in component.collision_geometry],
+        dilated_geometry=[poly for component in path_a for poly in component.dilated_collision_geometry],
+    )
+    engine.add_path(
+        "netB",
+        [poly for component in path_b for poly in component.collision_geometry],
+        dilated_geometry=[poly for component in path_b for poly in component.dilated_collision_geometry],
+    )
+
+    dynamic_paths = engine._dynamic_paths
+    assert set(dynamic_paths.net_envelopes) == {"netA", "netB"}
+    assert dynamic_paths.net_envelopes["netA"] == (-1.0, -2.0, 21.0, 2.0)
+    assert dynamic_paths.net_envelopes["netB"] == (-1.0, 38.0, 11.0, 42.0)
+    assert engine.metrics.total_dynamic_tree_rebuilds == 0
+
+    net_b_envelope_obj_id = dynamic_paths.net_envelope_obj_ids["netB"]
+    assert list(dynamic_paths.net_envelope_index.intersection((-5.0, 35.0, 15.0, 45.0))) == [net_b_envelope_obj_id]
+
+    engine.remove_path("netA")
+
+    assert "netA" not in dynamic_paths.net_envelopes
+    assert "netA" not in dynamic_paths.net_envelope_obj_ids
+    assert "netB" in dynamic_paths.net_envelopes
+    assert engine.metrics.total_dynamic_tree_rebuilds == 0
+
+
+
+def test_congestion_query_uses_per_polygon_bounds() -> None:
+    engine = RoutingWorld(clearance=2.0)
+    engine.metrics = AStarMetrics()
+
+    blocker = Straight.generate(Port(40, 4, 90), 2.0, width=2.0, dilation=1.0)
+    engine.add_path(
+        "netB",
+        [poly for poly in blocker.collision_geometry],
+        dilated_geometry=[poly for poly in blocker.dilated_collision_geometry],
+    )
+
+    move = ComponentResult(
+        start_port=Port(0, 0, 0),
+        collision_geometry=[box(0, 0, 10, 10), box(90, 0, 100, 10)],
+        end_port=Port(100, 0, 0),
+        length=100.0,
+        move_type="straight",
+        move_spec=StraightSeed(100.0),
+        physical_geometry=[box(0, 0, 10, 10), box(90, 0, 100, 10)],
+        dilated_collision_geometry=[box(0, 0, 10, 10), box(90, 0, 100, 10)],
+        dilated_physical_geometry=[box(0, 0, 10, 10), box(90, 0, 100, 10)],
+    )
+
+    assert engine.check_move_congestion(move, "netA") == 0
+    assert engine.metrics.total_congestion_candidate_nets == 0
+    assert engine.metrics.total_congestion_candidate_ids == 0
+    assert engine.metrics.total_congestion_exact_pair_checks == 0
+
+
+def test_congestion_touching_geometries_do_not_count_as_overlap() -> None:
+    engine = RoutingWorld(clearance=2.0)
+    engine.metrics = AStarMetrics()
+
+    existing = Straight.generate(Port(0, 0, 0), 10.0, width=2.0, dilation=1.0)
+    touching = Straight.generate(Port(12, 0, 0), 10.0, width=2.0, dilation=1.0)
+
+    engine.add_path(
+        "netB",
+        [poly for poly in existing.collision_geometry],
+        dilated_geometry=[poly for poly in existing.dilated_collision_geometry],
+    )
+
+    assert engine.check_move_congestion(touching, "netA") == 0
+
+
+def test_congestion_exact_checks_only_touch_relevant_move_polygons() -> None:
+    engine = RoutingWorld(clearance=2.0)
+    engine.metrics = AStarMetrics()
+
+    blocker = Straight.generate(Port(0, 0, 0), 10.0, width=2.0, dilation=1.0)
+    engine.add_path(
+        "netB",
+        [poly for poly in blocker.collision_geometry],
+        dilated_geometry=[poly for poly in blocker.dilated_collision_geometry],
+    )
+
+    move = ComponentResult(
+        start_port=Port(0, 0, 0),
+        collision_geometry=[box(0, -2, 10, 2), box(90, -2, 100, 2)],
+        end_port=Port(100, 0, 0),
+        length=100.0,
+        move_type="straight",
+        move_spec=StraightSeed(100.0),
+        physical_geometry=[box(0, -2, 10, 2), box(90, -2, 100, 2)],
+        dilated_collision_geometry=[box(0, -2, 10, 2), box(90, -2, 100, 2)],
+        dilated_physical_geometry=[box(0, -2, 10, 2), box(90, -2, 100, 2)],
+    )
+
+    assert engine.check_move_congestion(move, "netA") == 1
+    assert engine.metrics.total_congestion_candidate_nets == 1
+    assert engine.metrics.total_congestion_candidate_ids == 1
+    assert engine.metrics.total_congestion_exact_pair_checks == 1
+
+def test_congestion_grid_span_cache_reuses_broad_phase_candidates() -> None:
+    engine = RoutingWorld(clearance=2.0)
+    engine.metrics = AStarMetrics()
+    net_envelope_cache: dict[tuple[str, int, int, int, int], tuple[str, ...]] = {}
+    grid_net_cache: dict[tuple[str, int, int, int, int], tuple[str, ...]] = {}
+    cache: dict[tuple[str, int, int, int, int], dict[str, tuple[int, ...]]] = {}
+
+    blocker = Straight.generate(Port(15, 0, 0), 20.0, width=2.0, dilation=1.0)
+    engine.add_path(
+        "netB",
+        [poly for poly in blocker.collision_geometry],
+        dilated_geometry=[poly for poly in blocker.dilated_collision_geometry],
+    )
+
+    move_a = Straight.generate(Port(5, 0, 0), 20.0, width=2.0, dilation=1.0)
+    move_b = Straight.generate(Port(7, 0, 0), 20.0, width=2.0, dilation=1.0)
+
+    assert engine.check_move_congestion(
+        move_a,
+        "netA",
+        net_envelope_cache=net_envelope_cache,
+        grid_net_cache=grid_net_cache,
+        broad_phase_cache=cache,
+    ) == 1
+    assert engine.check_move_congestion(
+        move_b,
+        "netA",
+        net_envelope_cache=net_envelope_cache,
+        grid_net_cache=grid_net_cache,
+        broad_phase_cache=cache,
+    ) == 1
+    assert engine.metrics.total_congestion_candidate_nets == 2
+    assert engine.metrics.total_congestion_net_envelope_cache_misses == 1
+    assert engine.metrics.total_congestion_net_envelope_cache_hits == 1
+    assert engine.metrics.total_congestion_grid_net_cache_misses == 1
+    assert engine.metrics.total_congestion_grid_net_cache_hits == 1
+    assert engine.metrics.total_congestion_grid_span_cache_misses == 1
+    assert engine.metrics.total_congestion_grid_span_cache_hits == 1
+
+
+def test_has_possible_move_congestion_uses_presence_cache_and_skips_empty_spans() -> None:
+    engine = RoutingWorld(clearance=2.0)
+    engine.metrics = AStarMetrics()
+    presence_cache: dict[tuple[str, int, int, int, int], bool] = {}
+
+    blocker = Straight.generate(Port(10, 0, 0), 20.0, width=2.0, dilation=1.0)
+    engine.add_path(
+        "netB",
+        [poly for poly in blocker.collision_geometry],
+        dilated_geometry=[poly for poly in blocker.dilated_collision_geometry],
+    )
+    empty_move = Straight.generate(Port(200, 0, 0), 20.0, width=2.0, dilation=1.0)
+
+    assert not engine.has_possible_move_congestion(empty_move, "netA", presence_cache)
+    assert not engine.has_possible_move_congestion(empty_move, "netA", presence_cache)
+    assert engine.metrics.total_congestion_presence_cache_misses == 1
+    assert engine.metrics.total_congestion_presence_cache_hits == 1
+
+    occupied_move = Straight.generate(Port(0, 0, 0), 20.0, width=2.0, dilation=1.0)
+
+    assert engine.has_possible_move_congestion(occupied_move, "netA")
+
+
+def test_has_candidate_move_congestion_uses_candidate_precheck_cache() -> None:
+    engine = RoutingWorld(clearance=2.0)
+    engine.metrics = AStarMetrics()
+    candidate_precheck_cache: dict[tuple[str, int, int, int, int], bool] = {}
+    net_envelope_cache: dict[tuple[str, int, int, int, int], tuple[str, ...]] = {}
+    grid_net_cache: dict[tuple[str, int, int, int, int], tuple[str, ...]] = {}
+
+    blocker = Straight.generate(Port(10, 0, 0), 20.0, width=2.0, dilation=1.0)
+    engine.add_path(
+        "netB",
+        [poly for poly in blocker.collision_geometry],
+        dilated_geometry=[poly for poly in blocker.dilated_collision_geometry],
+    )
+    empty_move = Straight.generate(Port(200, 0, 0), 20.0, width=2.0, dilation=1.0)
+    occupied_move = Straight.generate(Port(0, 0, 0), 20.0, width=2.0, dilation=1.0)
+
+    assert not engine.has_candidate_move_congestion(
+        empty_move,
+        "netA",
+        candidate_precheck_cache,
+        net_envelope_cache,
+        grid_net_cache,
+    )
+    assert not engine.has_candidate_move_congestion(
+        empty_move,
+        "netA",
+        candidate_precheck_cache,
+        net_envelope_cache,
+        grid_net_cache,
+    )
+    assert engine.has_candidate_move_congestion(
+        occupied_move,
+        "netA",
+        candidate_precheck_cache,
+        net_envelope_cache,
+        grid_net_cache,
+    )
+    assert engine.metrics.total_congestion_candidate_precheck_misses >= 2
+    assert engine.metrics.total_congestion_candidate_precheck_hits >= 1

inire/tests/test_example_performance.py

@@ -6,7 +6,7 @@ from typing import TYPE_CHECKING
 
 import pytest
 
-from inire.tests.example_scenarios import SCENARIOS, ScenarioOutcome
+from inire.tests.example_scenarios import SCENARIOS, ScenarioOutcome, snapshot_example_07_no_warm_start
 
 if TYPE_CHECKING:
     from collections.abc import Callable
@@ -15,6 +15,7 @@ if TYPE_CHECKING:
 RUN_PERFORMANCE = os.environ.get("INIRE_RUN_PERFORMANCE") == "1"
 PERFORMANCE_REPEATS = 3
 REGRESSION_FACTOR = 1.5
+NO_WARM_START_REGRESSION_SECONDS = 15.0
 
 # Baselines are measured from clean 6a28dcf-style runs without plotting.
 BASELINE_SECONDS = {
@@ -67,3 +68,20 @@ def test_example_like_runtime_regression(scenario: tuple[str, Callable[[], Scena
         f"{REGRESSION_FACTOR:.1f}x baseline {BASELINE_SECONDS[name]:.4f}s "
         f"from timings {timings!r}"
     )
+
+
+@pytest.mark.performance
+@pytest.mark.skipif(not RUN_PERFORMANCE, reason="set INIRE_RUN_PERFORMANCE=1 to run runtime regression checks")
+def test_example_07_no_warm_start_runtime_regression() -> None:
+    snapshot = snapshot_example_07_no_warm_start()
+
+    assert snapshot.total_results == 10
+    assert snapshot.valid_results == 10
+    assert snapshot.reached_targets == 10
+    assert snapshot.metrics.warm_start_paths_built == 0
+    assert snapshot.metrics.warm_start_paths_used == 0
+    assert snapshot.duration_s <= NO_WARM_START_REGRESSION_SECONDS, (
+        "example_07_large_scale_routing_no_warm_start runtime "
+        f"{snapshot.duration_s:.4f}s exceeded guardrail "
+        f"{NO_WARM_START_REGRESSION_SECONDS:.1f}s"
+    )

inire/tests/test_example_regressions.py

@@ -14,7 +14,16 @@ from inire import (
 )
 from inire.router._stack import build_routing_stack
 from inire.seeds import Bend90Seed, PathSeed, StraightSeed
-from inire.tests.example_scenarios import SCENARIOS, _build_evaluator, _build_pathfinder, _net_specs, AStarMetrics
+from inire.tests.example_scenarios import (
+    SCENARIOS,
+    _build_evaluator,
+    _build_pathfinder,
+    _net_specs,
+    AStarMetrics,
+    snapshot_example_05,
+    snapshot_example_07_no_warm_start,
+    trace_example_07_no_warm_start,
+)
 
 
 EXPECTED_OUTCOMES = {
@@ -36,6 +45,43 @@ def test_examples_match_legacy_expected_outcomes(name: str, run) -> None:
     assert outcome[1:] == EXPECTED_OUTCOMES[name]
 
 
+def test_example_05_avoids_dynamic_tree_rebuilds() -> None:
+    snapshot = snapshot_example_05()
+
+    assert snapshot.valid_results == 3
+    assert snapshot.metrics.dynamic_tree_rebuilds == 0
+
+
+def test_example_07_no_warm_start_canary_improves_validity() -> None:
+    snapshot = snapshot_example_07_no_warm_start()
+
+    assert snapshot.total_results == 10
+    assert snapshot.reached_targets == 10
+    assert snapshot.valid_results == 10
+    assert snapshot.metrics.warm_start_paths_built == 0
+    assert snapshot.metrics.warm_start_paths_used == 0
+    assert snapshot.metrics.pair_local_search_pairs_considered >= 1
+    assert snapshot.metrics.pair_local_search_accepts >= 1
+    assert snapshot.metrics.pair_local_search_nodes_expanded <= 128
+    assert snapshot.metrics.nodes_expanded <= 2500
+    assert snapshot.metrics.congestion_check_calls <= 6000
+
+
+def test_example_07_no_warm_start_trace_finishes_without_conflict_edges() -> None:
+    run = trace_example_07_no_warm_start()
+
+    assert len(run.results_by_net) == 10
+    assert sum(result.is_valid for result in run.results_by_net.values()) == 10
+    assert sum(result.reached_target for result in run.results_by_net.values()) == 10
+    assert run.metrics.pair_local_search_pairs_considered >= 1
+    assert run.metrics.pair_local_search_accepts >= 1
+
+    final_entry = run.conflict_trace[-1]
+    assert final_entry.stage == "final"
+    assert len(final_entry.completed_net_ids) == 10
+    assert final_entry.conflict_edges == ()
+
+
 def test_example_06_clipped_bbox_margin_restores_legacy_seed() -> None:
     bounds = (-20, -20, 170, 170)
     obstacles = (

inire/tests/test_performance_reporting.py

@@ -16,10 +16,38 @@ def test_snapshot_example_01_exposes_metrics() -> None:
     assert snapshot.metrics.route_iterations >= 1
     assert snapshot.metrics.nets_routed >= 1
     assert snapshot.metrics.nodes_expanded > 0
+    assert snapshot.metrics.score_component_calls >= 0
+    assert snapshot.metrics.danger_map_lookup_calls >= 0
     assert snapshot.metrics.move_cache_abs_misses >= 0
+    assert snapshot.metrics.guidance_match_moves >= 0
+    assert snapshot.metrics.guidance_match_moves_straight >= 0
+    assert snapshot.metrics.guidance_match_moves_bend90 >= 0
+    assert snapshot.metrics.guidance_match_moves_sbend >= 0
+    assert snapshot.metrics.guidance_bonus_applied >= 0.0
+    assert snapshot.metrics.guidance_bonus_applied_straight >= 0.0
+    assert snapshot.metrics.guidance_bonus_applied_bend90 >= 0.0
+    assert snapshot.metrics.guidance_bonus_applied_sbend >= 0.0
     assert snapshot.metrics.ray_cast_calls >= 0
+    assert snapshot.metrics.ray_cast_calls_expand_forward >= 0
     assert snapshot.metrics.dynamic_tree_rebuilds >= 0
+    assert snapshot.metrics.visibility_corner_index_builds >= 0
     assert snapshot.metrics.visibility_builds >= 0
+    assert snapshot.metrics.congestion_grid_span_cache_hits >= 0
+    assert snapshot.metrics.congestion_grid_span_cache_misses >= 0
+    assert snapshot.metrics.congestion_candidate_nets >= 0
+    assert snapshot.metrics.congestion_net_envelope_cache_hits >= 0
+    assert snapshot.metrics.congestion_net_envelope_cache_misses >= 0
+    assert snapshot.metrics.congestion_grid_net_cache_hits >= 0
+    assert snapshot.metrics.congestion_grid_net_cache_misses >= 0
+    assert snapshot.metrics.congestion_lazy_resolutions >= 0
+    assert snapshot.metrics.congestion_lazy_requeues >= 0
+    assert snapshot.metrics.congestion_candidate_ids >= 0
+    assert snapshot.metrics.verify_dynamic_candidate_nets >= 0
+    assert snapshot.metrics.refinement_candidates_verified >= 0
+    assert snapshot.metrics.pair_local_search_pairs_considered >= 0
+    assert snapshot.metrics.pair_local_search_attempts >= 0
+    assert snapshot.metrics.pair_local_search_accepts >= 0
+    assert snapshot.metrics.pair_local_search_nodes_expanded >= 0
 
 
 def test_record_performance_baseline_script_writes_selected_scenario(tmp_path: Path) -> None:
@@ -43,3 +71,236 @@ def test_record_performance_baseline_script_writes_selected_scenario(tmp_path: P
     assert payload["generator"] == "scripts/record_performance_baseline.py"
     assert [entry["name"] for entry in payload["scenarios"]] == ["example_01_simple_route"]
     assert (tmp_path / "performance.md").exists()
+
+
+def test_diff_performance_baseline_script_writes_selected_scenario(tmp_path: Path) -> None:
+    repo_root = Path(__file__).resolve().parents[2]
+    record_script = repo_root / "scripts" / "record_performance_baseline.py"
+    diff_script = repo_root / "scripts" / "diff_performance_baseline.py"
+    baseline_dir = tmp_path / "baseline"
+    baseline_dir.mkdir()
+    output_path = tmp_path / "diff.md"
+
+    subprocess.run(
+        [
+            sys.executable,
+            str(record_script),
+            "--output-dir",
+            str(baseline_dir),
+            "--scenario",
+            "example_01_simple_route",
+        ],
+        check=True,
+    )
+    subprocess.run(
+        [
+            sys.executable,
+            str(diff_script),
+            "--baseline",
+            str(baseline_dir / "performance_baseline.json"),
+            "--include-performance-only",
+            "--scenario",
+            "example_01_simple_route",
+            "--output",
+            str(output_path),
+        ],
+        check=True,
+    )
+
+    report = output_path.read_text()
+    assert "Performance Baseline Diff" in report
+    assert "example_01_simple_route" in report
+
+
+def test_diff_performance_baseline_script_can_append_measurement_log(tmp_path: Path) -> None:
+    repo_root = Path(__file__).resolve().parents[2]
+    record_script = repo_root / "scripts" / "record_performance_baseline.py"
+    diff_script = repo_root / "scripts" / "diff_performance_baseline.py"
+    baseline_dir = tmp_path / "baseline"
+    baseline_dir.mkdir()
+    log_path = tmp_path / "optimization.md"
+
+    subprocess.run(
+        [
+            sys.executable,
+            str(record_script),
+            "--output-dir",
+            str(baseline_dir),
+            "--scenario",
+            "example_01_simple_route",
+        ],
+        check=True,
+    )
+    subprocess.run(
+        [
+            sys.executable,
+            str(diff_script),
+            "--baseline",
+            str(baseline_dir / "performance_baseline.json"),
+            "--include-performance-only",
+            "--scenario",
+            "example_01_simple_route",
+            "--metric",
+            "duration_s",
+            "--metric",
+            "valid_results",
+            "--metric",
+            "nodes_expanded",
+            "--metric",
+            "visibility_corner_index_builds",
+            "--label",
+            "Step 0 - Baseline",
+            "--notes",
+            "Tooling smoke test.",
+            "--log",
+            str(log_path),
+        ],
+        check=True,
+    )
+
+    report = log_path.read_text()
+    assert "Step 0 - Baseline" in report
+    assert "Tooling smoke test." in report
+    assert "| example_01_simple_route | duration_s |" in report
+    assert "| example_01_simple_route | valid_results |" in report
+    assert "| example_01_simple_route | visibility_corner_index_builds |" in report
+
+
+def test_diff_performance_baseline_script_renders_current_metrics_for_added_scenario(tmp_path: Path) -> None:
+    repo_root = Path(__file__).resolve().parents[2]
+    record_script = repo_root / "scripts" / "record_performance_baseline.py"
+    diff_script = repo_root / "scripts" / "diff_performance_baseline.py"
+    baseline_dir = tmp_path / "baseline"
+    baseline_dir.mkdir()
+    output_path = tmp_path / "diff_added.md"
+
+    subprocess.run(
+        [
+            sys.executable,
+            str(record_script),
+            "--output-dir",
+            str(baseline_dir),
+            "--scenario",
+            "example_01_simple_route",
+        ],
+        check=True,
+    )
+    subprocess.run(
+        [
+            sys.executable,
+            str(diff_script),
+            "--baseline",
+            str(baseline_dir / "performance_baseline.json"),
+            "--include-performance-only",
+            "--scenario",
+            "example_07_large_scale_routing_no_warm_start",
+            "--metric",
+            "duration_s",
+            "--metric",
+            "nodes_expanded",
+            "--output",
+            str(output_path),
+        ],
+        check=True,
+    )
+
+    report = output_path.read_text()
+    assert "| example_07_large_scale_routing_no_warm_start | duration_s | - |" in report
+    assert "| example_07_large_scale_routing_no_warm_start | nodes_expanded | - |" in report
+
+
+def test_record_conflict_trace_script_writes_selected_scenario(tmp_path: Path) -> None:
+    repo_root = Path(__file__).resolve().parents[2]
+    script_path = repo_root / "scripts" / "record_conflict_trace.py"
+
+    subprocess.run(
+        [
+            sys.executable,
+            str(script_path),
+            "--output-dir",
+            str(tmp_path),
+            "--scenario",
+            "example_05_orientation_stress",
+        ],
+        check=True,
+    )
+
+    payload = json.loads((tmp_path / "conflict_trace.json").read_text())
+    assert payload["generated_at"]
+    assert payload["generator"] == "scripts/record_conflict_trace.py"
+    assert [entry["name"] for entry in payload["scenarios"]] == ["example_05_orientation_stress"]
+    assert (tmp_path / "conflict_trace.md").exists()
+
+
+def test_record_conflict_trace_script_supports_performance_only_scenario(tmp_path: Path) -> None:
+    repo_root = Path(__file__).resolve().parents[2]
+    script_path = repo_root / "scripts" / "record_conflict_trace.py"
+
+    subprocess.run(
+        [
+            sys.executable,
+            str(script_path),
+            "--output-dir",
+            str(tmp_path),
+            "--include-performance-only",
+            "--scenario",
+            "example_07_large_scale_routing_no_warm_start",
+        ],
+        check=True,
+    )
+
+    payload = json.loads((tmp_path / "conflict_trace.json").read_text())
+    assert [entry["name"] for entry in payload["scenarios"]] == ["example_07_large_scale_routing_no_warm_start"]
+
+
+def test_record_frontier_trace_script_writes_selected_scenario(tmp_path: Path) -> None:
+    repo_root = Path(__file__).resolve().parents[2]
+    script_path = repo_root / "scripts" / "record_frontier_trace.py"
+
+    subprocess.run(
+        [
+            sys.executable,
+            str(script_path),
+            "--output-dir",
+            str(tmp_path),
+            "--scenario",
+            "example_05_orientation_stress",
+        ],
+        check=True,
+    )
+
+    payload = json.loads((tmp_path / "frontier_trace.json").read_text())
+    assert payload["generated_at"]
+    assert payload["generator"] == "scripts/record_frontier_trace.py"
+    assert [entry["name"] for entry in payload["scenarios"]] == ["example_05_orientation_stress"]
+    assert (tmp_path / "frontier_trace.md").exists()
+
+
+def test_characterize_pair_local_search_script_writes_outputs(tmp_path: Path) -> None:
+    repo_root = Path(__file__).resolve().parents[2]
+    script_path = repo_root / "scripts" / "characterize_pair_local_search.py"
+
+    subprocess.run(
+        [
+            sys.executable,
+            str(script_path),
+            "--output-dir",
+            str(tmp_path),
+            "--num-nets",
+            "6",
+            "--seeds",
+            "41",
+            "--repeats",
+            "1",
+        ],
+        check=True,
+    )
+
+    payload = json.loads((tmp_path / "pair_local_characterization.json").read_text())
+    assert payload["generated_at"]
+    assert payload["generator"] == "scripts/characterize_pair_local_search.py"
+    assert payload["grid"]["num_nets"] == [6]
+    assert payload["grid"]["seeds"] == [41]
+    assert payload["grid"]["repeats"] == 1
+    assert len(payload["cases"]) == 1
+    assert (tmp_path / "pair_local_characterization.md").exists()

inire/tests/test_visibility.py

@@ -2,6 +2,7 @@ from shapely.geometry import box
 
 from inire.geometry.collision import RoutingWorld
 from inire.geometry.primitives import Port
+from inire.router._astar_types import AStarMetrics
 from inire.router.visibility import VisibilityManager
 
 
@@ -18,3 +19,111 @@ def test_point_visibility_cache_respects_max_distance() -> None:
     assert len(near_corners) == 3
     assert len(far_corners) > len(near_corners)
     assert any(corner[0] >= 100.0 for corner in far_corners)
+
+
+def test_visibility_manager_is_lazy_until_queried() -> None:
+    engine = RoutingWorld(clearance=0.0)
+    engine.metrics = AStarMetrics()
+    engine.add_static_obstacle(box(10, 20, 20, 30))
+
+    visibility = VisibilityManager(engine)
+
+    assert visibility.corners == []
+    assert engine.metrics.total_visibility_corner_index_builds == 0
+    assert engine.metrics.total_visibility_builds == 0
+
+    visibility.ensure_corner_index_current()
+
+    assert visibility.corners
+    assert engine.metrics.total_visibility_corner_index_builds == 1
+    assert engine.metrics.total_visibility_builds == 0
+
+
+def test_exact_corner_visibility_builds_graph_once_per_static_version() -> None:
+    engine = RoutingWorld(clearance=0.0)
+    engine.metrics = AStarMetrics()
+    engine.add_static_obstacle(box(10, 20, 20, 30))
+    visibility = VisibilityManager(engine)
+    origin = Port(10, 20, 0)
+
+    first = visibility.get_corner_visibility(origin, max_dist=100.0)
+    second = visibility.get_corner_visibility(origin, max_dist=100.0)
+
+    assert second == first
+    assert engine.metrics.total_visibility_corner_index_builds == 1
+    assert engine.metrics.total_visibility_builds == 1
+
+
+def test_clear_cache_invalidates_without_rebuilding_and_static_change_rebuilds_lazily() -> None:
+    engine = RoutingWorld(clearance=0.0)
+    engine.metrics = AStarMetrics()
+    engine.add_static_obstacle(box(10, 20, 20, 30))
+    visibility = VisibilityManager(engine)
+
+    visibility.get_corner_visibility(Port(10, 20, 0), max_dist=100.0)
+    assert engine.metrics.total_visibility_corner_index_builds == 1
+    assert engine.metrics.total_visibility_builds == 1
+
+    visibility.clear_cache()
+
+    assert visibility.corners == []
+    assert engine.metrics.total_visibility_corner_index_builds == 1
+    assert engine.metrics.total_visibility_builds == 1
+
+    engine.add_static_obstacle(box(40, 20, 50, 30))
+
+    visible = visibility.get_corner_visibility(Port(10, 20, 0), max_dist=100.0)
+
+    assert visible == []
+    assert engine.metrics.total_visibility_corner_index_builds == 2
+    assert engine.metrics.total_visibility_builds == 2
+
+
+def test_tangent_corner_candidate_query_matches_bruteforce_filter() -> None:
+    engine = RoutingWorld(clearance=0.0)
+    engine.add_static_obstacle(box(10, 20, 20, 30))
+    engine.add_static_obstacle(box(-35, -15, -25, -5))
+    engine.add_static_obstacle(box(35, -40, 45, -30))
+    visibility = VisibilityManager(engine)
+    radii = (10.0, 20.0)
+    min_forward = 5.0
+    max_forward = 60.0
+    tolerance = 2.0
+
+    for origin in (
+        Port(0, 0, 0),
+        Port(0, 0, 90),
+        Port(0, 0, 180),
+        Port(0, 0, 270),
+    ):
+        candidate_ids = set(
+            visibility.get_tangent_corner_candidates(
+                origin,
+                min_forward=min_forward,
+                max_forward=max_forward,
+                radii=radii,
+                tolerance=tolerance,
+            )
+        )
+        expected_ids: set[int] = set()
+        if origin.r == 0:
+            cos_v, sin_v = 1.0, 0.0
+        elif origin.r == 90:
+            cos_v, sin_v = 0.0, 1.0
+        elif origin.r == 180:
+            cos_v, sin_v = -1.0, 0.0
+        else:
+            cos_v, sin_v = 0.0, -1.0
+
+        for idx, (cx, cy) in enumerate(visibility.corners):
+            dx = cx - origin.x
+            dy = cy - origin.y
+            local_x = dx * cos_v + dy * sin_v
+            local_y = -dx * sin_v + dy * cos_v
+            if local_x <= min_forward or local_x > max_forward + 0.01:
+                continue
+            nearest_radius = min(radii, key=lambda radius: abs(abs(local_y) - radius))
+            if abs(abs(local_y) - nearest_radius) <= tolerance:
+                expected_ids.add(idx)
+
+        assert candidate_ids == expected_ids

scripts/characterize_pair_local_search.py

@@ -0,0 +1,177 @@
+#!/usr/bin/env python3
+from __future__ import annotations
+
+import argparse
+import json
+from dataclasses import asdict
+from datetime import datetime
+from pathlib import Path
+from time import perf_counter
+
+from inire.tests.example_scenarios import _run_example_07_variant
+
+
+def _parse_csv_ints(raw: str) -> tuple[int, ...]:
+    return tuple(int(part) for part in raw.split(",") if part.strip())
+
+
+def _run_case(num_nets: int, seed: int) -> dict[str, object]:
+    t0 = perf_counter()
+    run = _run_example_07_variant(
+        num_nets=num_nets,
+        seed=seed,
+        warm_start_enabled=False,
+    )
+    duration_s = perf_counter() - t0
+    return {
+        "duration_s": duration_s,
+        "summary": {
+            "total_results": len(run.results_by_net),
+            "valid_results": sum(1 for result in run.results_by_net.values() if result.is_valid),
+            "reached_targets": sum(1 for result in run.results_by_net.values() if result.reached_target),
+        },
+        "metrics": asdict(run.metrics),
+    }
+
+
+def _is_smoke_candidate(entry: dict[str, object]) -> bool:
+    summary = entry["summary"]
+    metrics = entry["metrics"]
+    return (
+        summary["valid_results"] == summary["total_results"]
+        and metrics["pair_local_search_accepts"] >= 1
+        and entry["duration_s"] <= 1.0
+    )
+
+
+def _select_smoke_case(cases: list[dict[str, object]]) -> dict[str, object] | None:
+    grouped: dict[tuple[int, int], list[dict[str, object]]] = {}
+    for case in cases:
+        key = (case["num_nets"], case["seed"])
+        grouped.setdefault(key, []).append(case)
+
+    candidates = []
+    for (num_nets, seed), repeats in grouped.items():
+        if repeats and all(_is_smoke_candidate(repeat) for repeat in repeats):
+            candidates.append({"num_nets": num_nets, "seed": seed})
+    if not candidates:
+        return None
+    candidates.sort(key=lambda item: (item["num_nets"], item["seed"]))
+    return candidates[0]
+
+
+def _render_markdown(payload: dict[str, object]) -> str:
+    lines = [
+        "# Pair-Local Search Characterization",
+        "",
+        f"Generated at {payload['generated_at']} by `{payload['generator']}`.",
+        "",
+        f"Grid: `num_nets={payload['grid']['num_nets']}`, `seed={payload['grid']['seeds']}`, repeats={payload['grid']['repeats']}.",
+        "",
+        "| Nets | Seed | Repeat | Duration (s) | Valid | Reached | Pair Pairs | Pair Accepts | Pair Nodes | Nodes | Checks |",
+        "| :-- | :-- | :-- | --: | --: | --: | --: | --: | --: | --: | --: |",
+    ]
+    for case in payload["cases"]:
+        summary = case["summary"]
+        metrics = case["metrics"]
+        lines.append(
+            "| "
+            f"{case['num_nets']} | "
+            f"{case['seed']} | "
+            f"{case['repeat']} | "
+            f"{case['duration_s']:.4f} | "
+            f"{summary['valid_results']} | "
+            f"{summary['reached_targets']} | "
+            f"{metrics['pair_local_search_pairs_considered']} | "
+            f"{metrics['pair_local_search_accepts']} | "
+            f"{metrics['pair_local_search_nodes_expanded']} | "
+            f"{metrics['nodes_expanded']} | "
+            f"{metrics['congestion_check_calls']} |"
+        )
+
+    lines.extend(["", "## Recommendation", ""])
+    recommended = payload["recommended_smoke_scenario"]
+    if recommended is None:
+        lines.append(
+            "No smaller stable pair-local smoke scenario satisfied the rule "
+            "`valid_results == total_results`, `pair_local_search_accepts >= 1`, and `duration_s <= 1.0` across all repeats."
+        )
+    else:
+        lines.append(
+            f"Recommended smoke scenario: `num_nets={recommended['num_nets']}`, `seed={recommended['seed']}`."
+        )
+    return "\n".join(lines)
+
+
+def main() -> None:
+    parser = argparse.ArgumentParser(description="Characterize pair-local search across example_07-style no-warm runs.")
+    parser.add_argument(
+        "--num-nets",
+        default="6,8,10",
+        help="Comma-separated num_nets values to sweep. Default: 6,8,10.",
+    )
+    parser.add_argument(
+        "--seeds",
+        default="41,42,43",
+        help="Comma-separated seed values to sweep. Default: 41,42,43.",
+    )
+    parser.add_argument(
+        "--repeats",
+        type=int,
+        default=2,
+        help="Number of repeated runs per (num_nets, seed). Default: 2.",
+    )
+    parser.add_argument(
+        "--output-dir",
+        type=Path,
+        default=None,
+        help="Directory to write pair_local_characterization.json and .md into. Defaults to <repo>/docs.",
+    )
+    args = parser.parse_args()
+
+    repo_root = Path(__file__).resolve().parents[1]
+    output_dir = repo_root / "docs" if args.output_dir is None else args.output_dir.resolve()
+    output_dir.mkdir(exist_ok=True)
+
+    num_nets_values = _parse_csv_ints(args.num_nets)
+    seed_values = _parse_csv_ints(args.seeds)
+
+    cases: list[dict[str, object]] = []
+    for num_nets in num_nets_values:
+        for seed in seed_values:
+            for repeat in range(args.repeats):
+                case = _run_case(num_nets, seed)
+                case["num_nets"] = num_nets
+                case["seed"] = seed
+                case["repeat"] = repeat
+                cases.append(case)
+
+    payload = {
+        "generated_at": datetime.now().astimezone().isoformat(timespec="seconds"),
+        "generator": "scripts/characterize_pair_local_search.py",
+        "grid": {
+            "num_nets": list(num_nets_values),
+            "seeds": list(seed_values),
+            "repeats": args.repeats,
+        },
+        "cases": cases,
+        "recommended_smoke_scenario": _select_smoke_case(cases),
+    }
+
+    json_path = output_dir / "pair_local_characterization.json"
+    markdown_path = output_dir / "pair_local_characterization.md"
+    json_path.write_text(json.dumps(payload, indent=2, sort_keys=True) + "\n")
+    markdown_path.write_text(_render_markdown(payload) + "\n")
+
+    if json_path.is_relative_to(repo_root):
+        print(f"Wrote {json_path.relative_to(repo_root)}")
+    else:
+        print(f"Wrote {json_path}")
+    if markdown_path.is_relative_to(repo_root):
+        print(f"Wrote {markdown_path.relative_to(repo_root)}")
+    else:
+        print(f"Wrote {markdown_path}")
+
+
+if __name__ == "__main__":
+    main()

scripts/diff_performance_baseline.py

@@ -0,0 +1,237 @@
+#!/usr/bin/env python3
+from __future__ import annotations
+
+import argparse
+import json
+from dataclasses import asdict
+from datetime import datetime
+from pathlib import Path
+
+from inire.tests.example_scenarios import PERFORMANCE_SCENARIO_SNAPSHOTS, SCENARIO_SNAPSHOTS
+from inire.results import RouteMetrics
+
+
+SUMMARY_KEYS = (
+    "duration_s",
+    "valid_results",
+    "reached_targets",
+    "route_iterations",
+    "nets_routed",
+    "nodes_expanded",
+    "ray_cast_calls",
+    "moves_generated",
+    "moves_added",
+    "congestion_check_calls",
+    "verify_path_report_calls",
+)
+
+
+def _snapshot_registry(include_performance_only: bool) -> tuple[tuple[str, object], ...]:
+    if not include_performance_only:
+        return SCENARIO_SNAPSHOTS
+    return SCENARIO_SNAPSHOTS + PERFORMANCE_SCENARIO_SNAPSHOTS
+
+
+def _available_metric_names() -> tuple[str, ...]:
+    return (
+        "duration_s",
+        "total_results",
+        "valid_results",
+        "reached_targets",
+        *RouteMetrics.__dataclass_fields__.keys(),
+    )
+
+
+def _current_snapshots(
+    selected_scenarios: tuple[str, ...] | None,
+    *,
+    include_performance_only: bool,
+) -> dict[str, dict[str, object]]:
+    allowed = None if selected_scenarios is None else set(selected_scenarios)
+    snapshots: dict[str, dict[str, object]] = {}
+    for name, run in _snapshot_registry(include_performance_only):
+        if allowed is not None and name not in allowed:
+            continue
+        snapshots[name] = asdict(run())
+    return snapshots
+
+
+def _load_baseline(path: Path, selected_scenarios: tuple[str, ...] | None) -> dict[str, dict[str, object]]:
+    payload = json.loads(path.read_text())
+    allowed = None if selected_scenarios is None else set(selected_scenarios)
+    return {
+        entry["name"]: entry
+        for entry in payload["scenarios"]
+        if allowed is None or entry["name"] in allowed
+    }
+
+
+def _metric_value(snapshot: dict[str, object], key: str) -> float | None:
+    if key in {"duration_s", "total_results", "valid_results", "reached_targets"}:
+        return float(snapshot[key])
+    if key not in snapshot["metrics"]:
+        return None
+    return float(snapshot["metrics"][key])
+
+
+def _validate_metrics(metric_names: tuple[str, ...]) -> None:
+    valid_names = set(_available_metric_names())
+    unknown = [name for name in metric_names if name not in valid_names]
+    if unknown:
+        raise SystemExit(
+            f"Unknown metric name(s): {', '.join(sorted(unknown))}. "
+            f"Valid names are: {', '.join(_available_metric_names())}"
+        )
+
+
+def _render_report(
+    baseline: dict[str, dict[str, object]],
+    current: dict[str, dict[str, object]],
+    metric_names: tuple[str, ...],
+) -> str:
+    scenario_names = sorted(set(baseline) | set(current))
+    lines = [
+        "# Performance Baseline Diff",
+        "",
+        "| Scenario | Metric | Baseline | Current | Delta |",
+        "| :-- | :-- | --: | --: | --: |",
+    ]
+    for scenario in scenario_names:
+        base_snapshot = baseline.get(scenario)
+        curr_snapshot = current.get(scenario)
+        if base_snapshot is None:
+            if curr_snapshot is None:
+                lines.append(f"| {scenario} | added | - | - | - |")
+                continue
+            for key in metric_names:
+                curr_value = _metric_value(curr_snapshot, key)
+                if curr_value is None:
+                    lines.append(f"| {scenario} | {key} | - | - | - |")
+                    continue
+                lines.append(f"| {scenario} | {key} | - | {curr_value:.4f} | - |")
+            continue
+        if curr_snapshot is None:
+            lines.append(f"| {scenario} | missing | - | - | - |")
+            continue
+        for key in metric_names:
+            base_value = _metric_value(base_snapshot, key)
+            curr_value = _metric_value(curr_snapshot, key)
+            if base_value is None:
+                lines.append(
+                    f"| {scenario} | {key} | - | {curr_value:.4f} | - |"
+                )
+                continue
+            if curr_value is None:
+                lines.append(
+                    f"| {scenario} | {key} | {base_value:.4f} | - | - |"
+                )
+                continue
+            lines.append(
+                f"| {scenario} | {key} | {base_value:.4f} | {curr_value:.4f} | {curr_value - base_value:+.4f} |"
+            )
+    return "\n".join(lines) + "\n"
+
+
+def _render_log_entry(
+    *,
+    baseline_path: Path,
+    label: str,
+    notes: tuple[str, ...],
+    report: str,
+) -> str:
+    lines = [
+        f"## {label}",
+        "",
+        f"Measured on {datetime.now().astimezone().isoformat(timespec='seconds')}.",
+        f"Baseline: `{baseline_path}`.",
+        "",
+    ]
+    if notes:
+        lines.extend(["Findings:", ""])
+        lines.extend(f"- {note}" for note in notes)
+        lines.append("")
+    lines.append(report.rstrip())
+    lines.append("")
+    return "\n".join(lines)
+
+
+def main() -> None:
+    parser = argparse.ArgumentParser(description="Diff the committed performance baseline against a fresh run.")
+    parser.add_argument(
+        "--baseline",
+        type=Path,
+        default=Path("docs/performance_baseline.json"),
+        help="Baseline JSON to compare against.",
+    )
+    parser.add_argument(
+        "--output",
+        type=Path,
+        default=None,
+        help="Optional file to write the report to. Defaults to stdout.",
+    )
+    parser.add_argument(
+        "--scenario",
+        action="append",
+        dest="scenarios",
+        default=[],
+        help="Optional scenario name to include. May be passed more than once.",
+    )
+    parser.add_argument(
+        "--metric",
+        action="append",
+        dest="metrics",
+        default=[],
+        help="Optional metric to include. May be passed more than once. Defaults to the summary metric set.",
+    )
+    parser.add_argument(
+        "--label",
+        default="Measurement",
+        help="Section label to use when appending to a log file.",
+    )
+    parser.add_argument(
+        "--notes",
+        action="append",
+        dest="notes",
+        default=[],
+        help="Optional short finding to append under the measurement section. May be passed more than once.",
+    )
+    parser.add_argument(
+        "--log",
+        type=Path,
+        default=None,
+        help="Optional Markdown log file to append the rendered report to.",
+    )
+    parser.add_argument(
+        "--include-performance-only",
+        action="store_true",
+        help="Include performance-only snapshot scenarios that are excluded from the default baseline corpus.",
+    )
+    args = parser.parse_args()
+
+    selected = tuple(args.scenarios) if args.scenarios else None
+    metrics = tuple(args.metrics) if args.metrics else SUMMARY_KEYS
+    _validate_metrics(metrics)
+    baseline = _load_baseline(args.baseline, selected)
+    current = _current_snapshots(selected, include_performance_only=args.include_performance_only)
+    report = _render_report(baseline, current, metrics)
+
+    if args.output is not None:
+        args.output.write_text(report)
+        print(f"Wrote {args.output}")
+    elif args.log is None:
+        print(report, end="")
+
+    if args.log is not None:
+        entry = _render_log_entry(
+            baseline_path=args.baseline,
+            label=args.label,
+            notes=tuple(args.notes),
+            report=report,
+        )
+        with args.log.open("a", encoding="utf-8") as handle:
+            handle.write(entry)
+        print(f"Appended {args.log}")
+
+
+if __name__ == "__main__":
+    main()

scripts/record_conflict_trace.py

@@ -0,0 +1,228 @@
+#!/usr/bin/env python3
+from __future__ import annotations
+
+import argparse
+import json
+from collections import Counter
+from dataclasses import asdict
+from datetime import datetime
+from pathlib import Path
+
+from inire.results import RoutingRunResult
+from inire.tests.example_scenarios import TRACE_PERFORMANCE_SCENARIO_RUNS, TRACE_SCENARIO_RUNS
+
+
+def _trace_registry(include_performance_only: bool) -> tuple[tuple[str, object], ...]:
+    if include_performance_only:
+        return TRACE_SCENARIO_RUNS + TRACE_PERFORMANCE_SCENARIO_RUNS
+    return TRACE_SCENARIO_RUNS
+
+
+def _selected_runs(
+    selected_scenarios: tuple[str, ...] | None,
+    *,
+    include_performance_only: bool,
+) -> tuple[tuple[str, object], ...]:
+    if selected_scenarios is None:
+        return (("example_07_large_scale_routing_no_warm_start", dict(TRACE_PERFORMANCE_SCENARIO_RUNS)["example_07_large_scale_routing_no_warm_start"]),)
+
+    registry = dict(TRACE_SCENARIO_RUNS + TRACE_PERFORMANCE_SCENARIO_RUNS)
+    allowed_standard = dict(_trace_registry(include_performance_only))
+    runs = []
+    for name in selected_scenarios:
+        if name in allowed_standard:
+            runs.append((name, allowed_standard[name]))
+            continue
+        if name in registry:
+            runs.append((name, registry[name]))
+            continue
+        valid = ", ".join(sorted(registry))
+        raise SystemExit(f"Unknown trace scenario: {name}. Valid scenarios: {valid}")
+    return tuple(runs)
+
+
+def _result_summary(run: RoutingRunResult) -> dict[str, object]:
+    return {
+        "total_results": len(run.results_by_net),
+        "valid_results": sum(1 for result in run.results_by_net.values() if result.is_valid),
+        "reached_targets": sum(1 for result in run.results_by_net.values() if result.reached_target),
+        "results_by_net": {
+            net_id: {
+                "outcome": result.outcome,
+                "reached_target": result.reached_target,
+                "report": asdict(result.report),
+            }
+            for net_id, result in run.results_by_net.items()
+        },
+    }
+
+
+def _build_payload(
+    selected_scenarios: tuple[str, ...] | None,
+    *,
+    include_performance_only: bool,
+) -> dict[str, object]:
+    scenarios = []
+    for name, run in _selected_runs(selected_scenarios, include_performance_only=include_performance_only):
+        result = run()
+        scenarios.append(
+            {
+                "name": name,
+                "summary": _result_summary(result),
+                "metrics": asdict(result.metrics),
+                "conflict_trace": [asdict(entry) for entry in result.conflict_trace],
+            }
+        )
+    return {
+        "generated_at": datetime.now().astimezone().isoformat(timespec="seconds"),
+        "generator": "scripts/record_conflict_trace.py",
+        "scenarios": scenarios,
+    }
+
+
+def _count_stage_nets(entry: dict[str, object]) -> int:
+    return sum(
+        1
+        for net in entry["nets"]
+        if net["report"]["dynamic_collision_count"] > 0
+    )
+
+
+def _canonical_component_pair(
+    net_id: str,
+    self_component_index: int,
+    other_net_id: str,
+    other_component_index: int,
+) -> tuple[tuple[str, int], tuple[str, int]]:
+    left = (net_id, self_component_index)
+    right = (other_net_id, other_component_index)
+    if left <= right:
+        return (left, right)
+    return (right, left)
+
+
+def _render_markdown(payload: dict[str, object]) -> str:
+    lines = [
+        "# Conflict Trace",
+        "",
+        f"Generated at {payload['generated_at']} by `{payload['generator']}`.",
+        "",
+    ]
+
+    for scenario in payload["scenarios"]:
+        lines.extend(
+            [
+                f"## {scenario['name']}",
+                "",
+                f"Results: {scenario['summary']['valid_results']} valid / "
+                f"{scenario['summary']['reached_targets']} reached / "
+                f"{scenario['summary']['total_results']} total.",
+                "",
+                "| Stage | Iteration | Conflicting Nets | Conflict Edges | Completed Nets |",
+                "| :-- | --: | --: | --: | --: |",
+            ]
+        )
+
+        net_stage_counts: Counter[str] = Counter()
+        edge_counts: Counter[tuple[str, str]] = Counter()
+        component_pair_counts: Counter[tuple[tuple[str, int], tuple[str, int]]] = Counter()
+        trace_entries = scenario["conflict_trace"]
+        for entry in trace_entries:
+            lines.append(
+                "| "
+                f"{entry['stage']} | "
+                f"{'' if entry['iteration'] is None else entry['iteration']} | "
+                f"{_count_stage_nets(entry)} | "
+                f"{len(entry['conflict_edges'])} | "
+                f"{len(entry['completed_net_ids'])} |"
+            )
+            seen_component_pairs: set[tuple[tuple[str, int], tuple[str, int]]] = set()
+            for edge in entry["conflict_edges"]:
+                edge_counts[tuple(edge)] += 1
+            for net in entry["nets"]:
+                if net["report"]["dynamic_collision_count"] > 0:
+                    net_stage_counts[net["net_id"]] += 1
+                for component_conflict in net["component_conflicts"]:
+                    pair = _canonical_component_pair(
+                        net["net_id"],
+                        component_conflict["self_component_index"],
+                        component_conflict["other_net_id"],
+                        component_conflict["other_component_index"],
+                    )
+                    seen_component_pairs.add(pair)
+            for pair in seen_component_pairs:
+                component_pair_counts[pair] += 1
+
+        lines.extend(["", "Top nets by traced dynamic-collision stages:", ""])
+        if net_stage_counts:
+            for net_id, count in net_stage_counts.most_common(10):
+                lines.append(f"- `{net_id}`: {count}")
+        else:
+            lines.append("- None")
+
+        lines.extend(["", "Top net pairs by frequency:", ""])
+        if edge_counts:
+            for (left, right), count in edge_counts.most_common(10):
+                lines.append(f"- `{left}` <-> `{right}`: {count}")
+        else:
+            lines.append("- None")
+
+        lines.extend(["", "Top component pairs by frequency:", ""])
+        if component_pair_counts:
+            for pair, count in component_pair_counts.most_common(10):
+                (left_net, left_index), (right_net, right_index) = pair
+                lines.append(f"- `{left_net}[{left_index}]` <-> `{right_net}[{right_index}]`: {count}")
+        else:
+            lines.append("- None")
+
+        lines.append("")
+
+    return "\n".join(lines)
+
+
+def main() -> None:
+    parser = argparse.ArgumentParser(description="Record conflict-trace artifacts for selected trace scenarios.")
+    parser.add_argument(
+        "--scenario",
+        action="append",
+        dest="scenarios",
+        default=[],
+        help="Optional trace scenario name to include. May be passed more than once.",
+    )
+    parser.add_argument(
+        "--include-performance-only",
+        action="store_true",
+        help="Include performance-only trace scenarios when selecting from the standard registry.",
+    )
+    parser.add_argument(
+        "--output-dir",
+        type=Path,
+        default=None,
+        help="Directory to write conflict_trace.json and conflict_trace.md into. Defaults to <repo>/docs.",
+    )
+    args = parser.parse_args()
+
+    repo_root = Path(__file__).resolve().parents[1]
+    output_dir = repo_root / "docs" if args.output_dir is None else args.output_dir.resolve()
+    output_dir.mkdir(exist_ok=True)
+
+    selected = tuple(args.scenarios) if args.scenarios else None
+    payload = _build_payload(selected, include_performance_only=args.include_performance_only)
+    json_path = output_dir / "conflict_trace.json"
+    markdown_path = output_dir / "conflict_trace.md"
+
+    json_path.write_text(json.dumps(payload, indent=2, sort_keys=True) + "\n")
+    markdown_path.write_text(_render_markdown(payload) + "\n")
+
+    if json_path.is_relative_to(repo_root):
+        print(f"Wrote {json_path.relative_to(repo_root)}")
+    else:
+        print(f"Wrote {json_path}")
+    if markdown_path.is_relative_to(repo_root):
+        print(f"Wrote {markdown_path.relative_to(repo_root)}")
+    else:
+        print(f"Wrote {markdown_path}")
+
+
+if __name__ == "__main__":
+    main()

scripts/record_frontier_trace.py

@@ -0,0 +1,205 @@
+#!/usr/bin/env python3
+from __future__ import annotations
+
+import argparse
+import json
+from collections import Counter
+from dataclasses import asdict
+from datetime import datetime
+from pathlib import Path
+
+from inire.tests.example_scenarios import TRACE_PERFORMANCE_SCENARIO_RUNS, TRACE_SCENARIO_RUNS
+
+
+def _trace_registry(include_performance_only: bool) -> tuple[tuple[str, object], ...]:
+    if include_performance_only:
+        return TRACE_SCENARIO_RUNS + TRACE_PERFORMANCE_SCENARIO_RUNS
+    return TRACE_SCENARIO_RUNS
+
+
+def _selected_runs(
+    selected_scenarios: tuple[str, ...] | None,
+    *,
+    include_performance_only: bool,
+) -> tuple[tuple[str, object], ...]:
+    if selected_scenarios is None:
+        default_registry = dict(TRACE_PERFORMANCE_SCENARIO_RUNS)
+        return (("example_07_large_scale_routing_no_warm_start", default_registry["example_07_large_scale_routing_no_warm_start"]),)
+
+    registry = dict(TRACE_SCENARIO_RUNS + TRACE_PERFORMANCE_SCENARIO_RUNS)
+    allowed_standard = dict(_trace_registry(include_performance_only))
+    runs = []
+    for name in selected_scenarios:
+        if name in allowed_standard:
+            runs.append((name, allowed_standard[name]))
+            continue
+        if name in registry:
+            runs.append((name, registry[name]))
+            continue
+        valid = ", ".join(sorted(registry))
+        raise SystemExit(f"Unknown trace scenario: {name}. Valid scenarios: {valid}")
+    return tuple(runs)
+
+
+def _build_payload(
+    selected_scenarios: tuple[str, ...] | None,
+    *,
+    include_performance_only: bool,
+) -> dict[str, object]:
+    scenarios = []
+    for name, run in _selected_runs(selected_scenarios, include_performance_only=include_performance_only):
+        result = run()
+        scenarios.append(
+            {
+                "name": name,
+                "summary": {
+                    "total_results": len(result.results_by_net),
+                    "valid_results": sum(1 for entry in result.results_by_net.values() if entry.is_valid),
+                    "reached_targets": sum(1 for entry in result.results_by_net.values() if entry.reached_target),
+                },
+                "metrics": asdict(result.metrics),
+                "frontier_trace": [asdict(entry) for entry in result.frontier_trace],
+            }
+        )
+    return {
+        "generated_at": datetime.now().astimezone().isoformat(timespec="seconds"),
+        "generator": "scripts/record_frontier_trace.py",
+        "scenarios": scenarios,
+    }
+
+
+def _render_markdown(payload: dict[str, object]) -> str:
+    lines = [
+        "# Frontier Trace",
+        "",
+        f"Generated at {payload['generated_at']} by `{payload['generator']}`.",
+        "",
+    ]
+
+    for scenario in payload["scenarios"]:
+        lines.extend(
+            [
+                f"## {scenario['name']}",
+                "",
+                f"Results: {scenario['summary']['valid_results']} valid / "
+                f"{scenario['summary']['reached_targets']} reached / "
+                f"{scenario['summary']['total_results']} total.",
+                "",
+                "| Net | Hotspots | Closed-Set | Hard Collision | Self Collision | Cost | Samples |",
+                "| :-- | --: | --: | --: | --: | --: | --: |",
+            ]
+        )
+
+        reason_counts: Counter[str] = Counter()
+        hotspot_counts: Counter[tuple[str, int]] = Counter()
+        for net_trace in scenario["frontier_trace"]:
+            sample_count = len(net_trace["samples"])
+            lines.append(
+                "| "
+                f"{net_trace['net_id']} | "
+                f"{len(net_trace['hotspot_bounds'])} | "
+                f"{net_trace['pruned_closed_set']} | "
+                f"{net_trace['pruned_hard_collision']} | "
+                f"{net_trace['pruned_self_collision']} | "
+                f"{net_trace['pruned_cost']} | "
+                f"{sample_count} |"
+            )
+            reason_counts["closed_set"] += net_trace["pruned_closed_set"]
+            reason_counts["hard_collision"] += net_trace["pruned_hard_collision"]
+            reason_counts["self_collision"] += net_trace["pruned_self_collision"]
+            reason_counts["cost"] += net_trace["pruned_cost"]
+            for sample in net_trace["samples"]:
+                hotspot_counts[(net_trace["net_id"], sample["hotspot_index"])] += 1
+
+        lines.extend(["", "Prune totals by reason:", ""])
+        if reason_counts:
+            for reason, count in reason_counts.most_common():
+                lines.append(f"- `{reason}`: {count}")
+        else:
+            lines.append("- None")
+
+        lines.extend(["", "Top traced hotspots by sample count:", ""])
+        if hotspot_counts:
+            for (net_id, hotspot_index), count in hotspot_counts.most_common(10):
+                lines.append(f"- `{net_id}` hotspot `{hotspot_index}`: {count}")
+        else:
+            lines.append("- None")
+
+        lines.extend(["", "Per-net sampled reason/move breakdown:", ""])
+        if scenario["frontier_trace"]:
+            for net_trace in scenario["frontier_trace"]:
+                reason_move_counts: Counter[tuple[str, str]] = Counter()
+                hotspot_sample_counts: Counter[int] = Counter()
+                for sample in net_trace["samples"]:
+                    reason_move_counts[(sample["reason"], sample["move_type"])] += 1
+                    hotspot_sample_counts[sample["hotspot_index"]] += 1
+                lines.append(f"- `{net_trace['net_id']}`")
+                if reason_move_counts:
+                    top_pairs = ", ".join(
+                        f"`{reason}` x `{move}` = {count}"
+                        for (reason, move), count in reason_move_counts.most_common(3)
+                    )
+                    lines.append(f"  sampled reasons: {top_pairs}")
+                else:
+                    lines.append("  sampled reasons: none")
+                if hotspot_sample_counts:
+                    top_hotspots = ", ".join(
+                        f"`{hotspot}` = {count}" for hotspot, count in hotspot_sample_counts.most_common(3)
+                    )
+                    lines.append(f"  hotspot samples: {top_hotspots}")
+                else:
+                    lines.append("  hotspot samples: none")
+        else:
+            lines.append("- None")
+
+        lines.append("")
+
+    return "\n".join(lines)
+
+
+def main() -> None:
+    parser = argparse.ArgumentParser(description="Record frontier-trace artifacts for selected trace scenarios.")
+    parser.add_argument(
+        "--scenario",
+        action="append",
+        dest="scenarios",
+        default=[],
+        help="Optional trace scenario name to include. May be passed more than once.",
+    )
+    parser.add_argument(
+        "--include-performance-only",
+        action="store_true",
+        help="Include performance-only trace scenarios when selecting from the standard registry.",
+    )
+    parser.add_argument(
+        "--output-dir",
+        type=Path,
+        default=None,
+        help="Directory to write frontier_trace.json and frontier_trace.md into. Defaults to <repo>/docs.",
+    )
+    args = parser.parse_args()
+
+    repo_root = Path(__file__).resolve().parents[1]
+    output_dir = repo_root / "docs" if args.output_dir is None else args.output_dir.resolve()
+    output_dir.mkdir(exist_ok=True)
+
+    selected = tuple(args.scenarios) if args.scenarios else None
+    payload = _build_payload(selected, include_performance_only=args.include_performance_only)
+    json_path = output_dir / "frontier_trace.json"
+    markdown_path = output_dir / "frontier_trace.md"
+
+    json_path.write_text(json.dumps(payload, indent=2, sort_keys=True) + "\n")
+    markdown_path.write_text(_render_markdown(payload) + "\n")
+
+    if json_path.is_relative_to(repo_root):
+        print(f"Wrote {json_path.relative_to(repo_root)}")
+    else:
+        print(f"Wrote {json_path}")
+    if markdown_path.is_relative_to(repo_root):
+        print(f"Wrote {markdown_path.relative_to(repo_root)}")
+    else:
+        print(f"Wrote {markdown_path}")
+
+
+if __name__ == "__main__":
+    main()

scripts/record_performance_baseline.py

@@ -7,7 +7,7 @@ from dataclasses import asdict
 from datetime import date
 from pathlib import Path
 
-from inire.tests.example_scenarios import SCENARIO_SNAPSHOTS
+from inire.tests.example_scenarios import PERFORMANCE_SCENARIO_SNAPSHOTS, SCENARIO_SNAPSHOTS
 
 
 SUMMARY_METRICS = (
@@ -24,10 +24,20 @@ SUMMARY_METRICS = (
 )
 
 
-def _build_payload(selected_scenarios: tuple[str, ...] | None = None) -> dict[str, object]:
+def _snapshot_registry(include_performance_only: bool) -> tuple[tuple[str, object], ...]:
+    if not include_performance_only:
+        return SCENARIO_SNAPSHOTS
+    return SCENARIO_SNAPSHOTS + PERFORMANCE_SCENARIO_SNAPSHOTS
+
+
+def _build_payload(
+    selected_scenarios: tuple[str, ...] | None = None,
+    *,
+    include_performance_only: bool = False,
+) -> dict[str, object]:
     allowed = None if selected_scenarios is None else set(selected_scenarios)
     snapshots = []
-    for name, run in SCENARIO_SNAPSHOTS:
+    for name, run in _snapshot_registry(include_performance_only):
         if allowed is not None and name not in allowed:
             continue
         snapshots.append(run())
@@ -46,6 +56,7 @@ def _render_markdown(payload: dict[str, object]) -> str:
         f"Generated on {payload['generated_on']} by `{payload['generator']}`.",
         "",
         "The full machine-readable snapshot lives in `docs/performance_baseline.json`.",
+        "Use `scripts/diff_performance_baseline.py` to compare a fresh run against that snapshot.",
         "",
         "| Scenario | Duration (s) | Total | Valid | Reached | Iter | Nets Routed | Nodes | Ray Casts | Moves Gen | Moves Added | Dyn Tree | Visibility Builds | Congestion Checks | Verify Calls |",
         "| :-- | --: | --: | --: | --: | --: | --: | --: | --: | --: | --: | --: | --: | --: | --: |",
@@ -77,6 +88,7 @@ def _render_markdown(payload: dict[str, object]) -> str:
             "## Full Counter Set",
             "",
             "Each scenario entry in `docs/performance_baseline.json` records the full `RouteMetrics` snapshot, including cache, index, congestion, and verification counters.",
+            "These counters are currently observational only and are not enforced as CI regression gates.",
             "",
             "Tracked metric keys:",
             "",
@@ -101,6 +113,11 @@ def main() -> None:
         default=[],
         help="Optional scenario name to include. May be passed more than once.",
     )
+    parser.add_argument(
+        "--include-performance-only",
+        action="store_true",
+        help="Include performance-only snapshot scenarios that are excluded from the default baseline corpus.",
+    )
     args = parser.parse_args()
 
     repo_root = Path(__file__).resolve().parents[1]
@@ -108,7 +125,7 @@ def main() -> None:
     docs_dir.mkdir(exist_ok=True)
 
     selected = tuple(args.scenarios) if args.scenarios else None
-    payload = _build_payload(selected)
+    payload = _build_payload(selected, include_performance_only=args.include_performance_only)
     json_path = docs_dir / "performance_baseline.json"
     markdown_path = docs_dir / "performance.md"