feat(harness): init ESKF from adapter's first GT pose as synthetic GPS origin

Wires a real CoordinateTransformer into the processor and seeds the ESKF
with the dataset's first ground-truth lat/lon/alt before the frame loop.
Result on EuRoC MH_01 (100 frames):
  eskf_initialized: 0/100 → 100/100
  vo_success: 99/100 (unchanged)
  eskf_has_position: 100/100

Satellite measurements are now correctly rejected by the Mahalanobis gate
(Δ² ~10⁶) because ORB produces unit-scale translations (scale_ambiguous=True)
which drive the ESKF position to diverge rapidly. The gate is working as
intended — the remaining issue is VO metric scale, not ESKF initialisation.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

src/gps_denied/testing/harness.py

@@ -23,6 +23,7 @@ import cv2
 import numpy as np
 
 from gps_denied.core.chunk_manager import RouteChunkManager
+from gps_denied.core.coordinates import CoordinateTransformer
 from gps_denied.core.gpr import GlobalPlaceRecognition
 from gps_denied.core.graph import FactorGraphOptimizer
 from gps_denied.core.metric import MetricRefinement
@@ -30,6 +31,7 @@ from gps_denied.core.models import ModelManager
 from gps_denied.core.processor import FlightProcessor
 from gps_denied.core.recovery import FailureRecoveryCoordinator
 from gps_denied.core.vo import ORBVisualOdometry
+from gps_denied.schemas import GPSPoint
 from gps_denied.schemas.graph import FactorGraphConfig
 from gps_denied.testing.datasets.base import (
     DatasetAdapter,
@@ -88,6 +90,16 @@ class E2EHarness:
         for i, pose in enumerate(gt_poses):
             gt_by_idx[i] = pose
 
+        # Seed ESKF with first GT pose as synthetic GPS origin so the processor
+        # can propagate VO translations into a consistent ENU frame.
+        if gt_poses:
+            origin = gt_poses[0]
+            start_gps = GPSPoint(lat=origin.lat, lon=origin.lon)
+            processor._coord.set_enu_origin(self._flight_id, start_gps)  # noqa: SLF001
+            processor._init_eskf_for_flight(  # noqa: SLF001
+                self._flight_id, start_gps, altitude=origin.alt
+            )
+
         trace_fh = None
         if self._trace_path is not None:
             self._trace_path.parent.mkdir(parents=True, exist_ok=True)
@@ -182,9 +194,11 @@ class E2EHarness:
         graph = FactorGraphOptimizer(FactorGraphConfig())
         chunk_mgr = RouteChunkManager(graph)
         recovery = FailureRecoveryCoordinator(chunk_mgr, gpr, metric)
+        coord = CoordinateTransformer()
         proc.attach_components(
             vo=vo, gpr=gpr, metric=metric,
             graph=graph, recovery=recovery, chunk_mgr=chunk_mgr,
+            coord=coord,
         )
         return proc
 

tests/e2e/test_euroc.py

@@ -38,9 +38,11 @@ async def test_euroc_mh01_rmse_within_ceiling(euroc_mh01_root: Path):
     result = await harness.run()
     if result.estimated_positions_enu.shape[0] == 0:
         pytest.xfail(
-            "Pipeline emits GPS estimates via fallback satellite matching but ESKF never "
+            "Pipeline emits zero GPS estimates — ESKF+VO active (99/100 vo_success, "
-            "initialises (no start_gps call in harness). VO now engages at 99% with ORB. "
+            "100/100 eskf_initialized) but satellite outlier rejection blocks all fixes. "
-            "Next: wire ESKF init with a synthetic GPS origin in the harness."
+            "Root cause: ORB scale_ambiguous=True → unit-scale VO translation → ESKF "
+            "position diverges → Mahalanobis gate rejects satellite measurements (~10^6 >> 16.3). "
+            "Next: apply VO scale from ESKF velocity or switch to metric VO backend."
         )
     # Align lengths by truncating to shorter (estimates may lag GT at start)
     n = min(result.estimated_positions_enu.shape[0], result.ground_truth.shape[0])
@@ -51,7 +53,8 @@ async def test_euroc_mh01_rmse_within_ceiling(euroc_mh01_root: Path):
     if ate["rmse"] >= EUROC_MH01_RMSE_CEILING_M:
         pytest.xfail(
             f"ATE RMSE={ate['rmse']:.2f}m exceeds {EUROC_MH01_RMSE_CEILING_M}m ceiling. "
-            "VO engages (99%) but ESKF never initialises without a start_gps call in the "
+            "VO+ESKF both active but ORB translation is unit-scale (scale_ambiguous=True) → "
-            "harness — estimates come from satellite fallback only."
+            "ESKF position diverges rapidly → satellite Mahalanobis gate rejects all fixes. "
+            "Fix: recover metric scale from ESKF velocity or use cuVSLAM (metric VO)."
         )
     assert ate["rmse"] < EUROC_MH01_RMSE_CEILING_M, f"ATE RMSE={ate['rmse']:.2f}m"