feat(perception): integrate calibrated operator pipeline

Add calibrated K1 projection, recorded and near-live perception qualification, unified Rerun operator layers, bounded replay admission, audited viewer controls, worker experiments, and lab evidence.

tests/test_e3_rectified_segmentation.py

@@ -0,0 +1,107 @@
+from __future__ import annotations
+
+import importlib.util
+import math
+import sys
+from pathlib import Path
+from types import ModuleType
+
+import numpy as np
+import pytest
+
+WORKER_ROOT = (
+    Path(__file__).parents[1] / "experiments" / "perception" / "worker"
+)
+PROFILE_PATH = WORKER_ROOT / "e3_k1_camera1_profile.json"
+
+
+def _worker_module() -> ModuleType:
+    path = WORKER_ROOT / "run_e3_rectified_segmentation.py"
+    sys.path.insert(0, str(WORKER_ROOT))
+    spec = importlib.util.spec_from_file_location("e3_rectified_segmentation", path)
+    assert spec is not None and spec.loader is not None
+    module = importlib.util.module_from_spec(spec)
+    spec.loader.exec_module(module)
+    return module
+
+
+def test_e3_profile_is_pinned_to_camera1_kb4_and_exact_model() -> None:
+    worker = _worker_module()
+
+    profile, digest = worker._profile(PROFILE_PATH)
+
+    assert len(digest) == 64
+    assert profile["source"]["calibration_slot"] == "camera_1"
+    assert profile["source"]["distortion_kb4"] == pytest.approx(
+        [
+            -0.023164451386679667,
+            -0.0014974198594105452,
+            -0.001039213149441563,
+            -0.000035237331915978814,
+        ]
+    )
+    assert profile["model"]["revision"] == "8d6b6d1a3f7b50d441afd7d247c2ed10db186e8f"
+    assert profile["model"]["files"]["model.safetensors"]["sha256"] == (
+        "c265da9a74f58f5c3f4826d23ca4ca78beac0b106cca5842beca61580de5b782"
+    )
+
+
+def test_kb4_inverse_round_trips_valid_field_angles() -> None:
+    worker = _worker_module()
+    profile, _digest = worker._profile(PROFILE_PATH)
+    coefficients = np.asarray(profile["source"]["distortion_kb4"])
+    theta = np.linspace(0.0, math.radians(96.4), 512)
+    squared = theta * theta
+    distorted = theta * (
+        1.0
+        + coefficients[0] * squared
+        + coefficients[1] * squared**2
+        + coefficients[2] * squared**3
+        + coefficients[3] * squared**4
+    )
+
+    recovered = worker._invert_kb4(distorted, coefficients)
+
+    assert recovered == pytest.approx(theta, abs=1e-10)
+
+
+def test_five_view_rectification_covers_the_k1_valid_circle() -> None:
+    worker = _worker_module()
+    profile, _digest = worker._profile(PROFILE_PATH)
+    width, height = profile["source"]["resolution"]
+    _fx, _fy, cx, cy = profile["source"]["intrinsic_fx_fy_cx_cy"]
+    y, x = np.mgrid[:height, :width]
+    valid_mask = np.hypot(x - cx, y - cy) <= 293.0
+
+    maps = worker._rectification_maps(profile, valid_mask)
+
+    assert maps["coverage"]["coverage_fraction"] == 1.0
+    assert maps["coverage"]["covered_pixel_count"] == int(valid_mask.sum())
+    assert maps["coverage"]["overlap_count"]["min"] >= 1
+    assert maps["coverage"]["overlap_count"]["max"] <= 3
+
+
+def test_tile_fusion_uses_the_selected_view_and_masks_outside_fov() -> None:
+    worker = _worker_module()
+    valid_mask = np.asarray([[True, True], [False, True]])
+    maps = {
+        "selected_tile": np.asarray([[0, 1], [-1, 1]], dtype=np.int8),
+        "tiles": [
+            {
+                "tile_map_x": np.zeros((2, 2), dtype=np.float32),
+                "tile_map_y": np.zeros((2, 2), dtype=np.float32),
+            },
+            {
+                "tile_map_x": np.ones((2, 2), dtype=np.float32),
+                "tile_map_y": np.ones((2, 2), dtype=np.float32),
+            },
+        ],
+    }
+    semantics = [
+        np.asarray([[4, 4], [4, 4]], dtype=np.uint8),
+        np.asarray([[7, 7], [7, 7]], dtype=np.uint8),
+    ]
+
+    result = worker._fuse_tiles(semantics, maps, valid_mask)
+
+    assert result.tolist() == [[4, 7], [0, 7]]