Promote PF-seeded RTK consensus above Tokyo target

README.md

@@ -115,10 +115,14 @@ reference truth or runtime FGO, and requires declared false FIX at or below 1%.
 Latest `libgnss++` selected results are shown only as an external reference;
 every imported RTK idea must still pass the PF posterior and promotion gates.
 
-| Full-epoch matched 3D | PF-only locked current | latest libgnss++ reference | campaign target |
+| Full-epoch matched 3D | PF-only locked current | latest libgnss++ reference | v0.3 promotion floor |
 |---|---:|---:|---:|
-| Tokyo run1 | **31.40%** (3,744 / 11,924) | 80.02% | 81.00% |
-| Nagoya run1 | **69.55%** (5,274 / 7,583) | 85.84% | 86.00% |
+| Tokyo run1 | **46.51%** (5,546 / 11,924) | 80.02% | 45.00% |
+| Nagoya run1 | **75.37%** (5,715 / 7,583) | 85.84% | 69.55% non-degradation |
+
+The full WP172 dual-RTK candidate supply is measured at a conservative
+42.653 ms/epoch when executed sequentially (21.826 ms/epoch concurrently),
+under the 100 ms production budget. FIX declarations remain suppressed.
 
 <p align="center">
   <img src="docs/assets/figures/pf_only_rtk_stretch.svg" alt="PF-only Tokyo and Nagoya RTK stretch progress, external libgnss++ reference, targets, and truth-free promotion gates" width="900">

RELEASE_NOTES_v0.3.0.md

@@ -30,15 +30,25 @@ python tools/build_release_bundle.py --verify dist/reproducibility
   greedy comparator ends 10 m away.
 - GTX 1660 Ti maxima are 13.761 ms in normal mode and 75.907 ms in search mode,
   with no recorded deadline miss.
+- Full WP172 candidate supply replays at a conservative 42.653 ms/epoch when
+  its two RTK paths are summed sequentially (21.826 ms/epoch concurrently);
+  the guarded final trajectory is byte-identical.
 - The cross-domain campaign covers 3 cities, 9 sites/routes, 5 dates, and
   3 receivers. Epoch-weighted RMS changes from 17.107 m to 16.916 m, with
   Tokyo non-degradation and a Hong Kong gain.
 - The deterministic ROS replay contains 10 events and one controlled restart.
+- The PF-only WP172 production trajectory reaches 5,546/11,924 Tokyo epochs
+  below 50 cm (46.5112%), with 1,802 gained, zero lost, and zero false FIX.
+- The fail-closed production audit passes all 11 gates.
 
 ## Important limits
 
-This release does not claim the program's 45% Tokyo sub-50 cm target. The
-Phase 5 campaign validates RMS and non-degradation, not that target. The locked
-Phase 3 outage audit is synthetic, Hong Kong is reproduced from a tracked
-summary because raw data is absent, and the Windows memory figure is a
-conservative capacity estimate. Promotion remains fail closed.
+The 45% Tokyo sub-50 cm promotion floor is met, but WP172 deliberately
+suppresses FIX declarations; it does not claim RTK FIX coverage. The locked
+Tokyo rerun is an operational promotion audit rather than a virgin scientific
+holdout because earlier campaign diagnostics had inspected Tokyo run1. The
+numerical consensus/residual gates were frozen and checked on Nagoya. The
+locked Phase 3 outage audit is synthetic, Hong Kong is reproduced from a
+tracked summary because raw data is absent, and the Windows memory figure is a
+conservative capacity estimate. Promotion remains fail closed for future
+changes.

configs/evaluation/v030_production_promotion.json

@@ -7,13 +7,15 @@
     "gained_epochs_min": 1,
     "normal_latency_max_ms": 100.0,
     "search_latency_max_ms": 1000.0,
+    "wp172_sequential_average_max_ms": 100.0,
     "mandatory_negative_holdouts": 4,
     "minimum_cities": 3
   },
   "evidence": {
     "negative_holdouts": "internal_docs/phase1_holdout_detector_audit_2026_07_29.json",
-    "tokyo_production": "internal_docs/wp160_tokyo_screened_stability_promotion_2026_07_24.json",
+    "tokyo_production": "internal_docs/wp172_tokyo_final_holdout_2026_07_29.json",
     "runtime": "internal_docs/phase4_realtime_benchmark_2026_07_29.json",
+    "wp172_runtime": "internal_docs/wp172_runtime_benchmark_2026_07_29.json",
     "cross_domain": "internal_docs/phase5_cross_domain_result_2026_07_29.json",
     "ros2_replay_input": "internal_docs/phase6_ros2_replay_input_2026_07_29.json",
     "ros2_replay_result": "internal_docs/phase6_ros2_replay_result_2026_07_29.json",

configs/evaluation/wp172_pf_seeded_rtk_consensus.json

@@ -0,0 +1,25 @@
+{
+  "schema": "gnss_gpu_wp172_pf_seeded_rtk_consensus_contract_v1",
+  "frozen_from": "nagoya_run1_development",
+  "production_input_truth": false,
+  "runtime_fgo": false,
+  "pf_only": true,
+  "candidate_gate": {
+    "required_seeded_status": 4,
+    "max_independent_position_disagreement_m": 1.0,
+    "max_prefit_residual_rms_m": 4.0
+  },
+  "output_policy": {
+    "fallback": "retain_production_epoch",
+    "declare_fix": false
+  },
+  "holdout_policy": {
+    "tokyo_truth_usage": "post_freeze_full_denominator_audit_only",
+    "mandatory_negative_holdout_disposition": "source_class_mismatch",
+    "reason": "WP172 requires an independently agreeing PF-seeded RTK candidate; CPPR unsafe triples do not satisfy the input contract"
+  },
+  "m4_expected_sha256": {
+    "internal_docs/wp30_m4_production_config.json": "3FB163B16E59C6394A38206E826FE9E0411975597E8BC62E8A1AF67FBFFC8115",
+    "internal_docs/wp30_m4_tokyo_evidence_ledger.json": "555C4FC2992654F9073323985EF7C35DB46738FEBE8F3F54D442C8BC17B76787"
+  }
+}