feat(control-station): review native raw fisheye risk cases

docs/22_MILESTONE_4_OBJECT_CENTRIC_RECORDED_REALTIME_CV.md

@@ -1445,6 +1445,36 @@ configuration, dataset release, metrics and ONNX identity; only then is a
 candidate loaded through the existing Triton seam and measured by a new
 append-only M4.8R run plus an independent validation gate.
 
+### 2026-08-26 — M4.8Q native raw-fisheye risk review ready
+
+The terminal phase of the existing `m48t-risk-quality-temporal` work is now the
+content-addressed M4.8Q result
+`m48q-native-risk-quality-lab-e8961d4db8ffb751503b9646bc8eeebec6c80c948ce2f5b4fa5b89c471e673c1`.
+It replaces neither the historical M4.8T evidence nor either admitted M4 viewer.
+The result projects `24` deterministic diagnostic cases into the existing M4.8
+image-case review instrument, with source-pixel boxes rendered by the client and
+an explicit clean-image toggle. The JPEG review derivatives retain the exact
+`800×600` raster geometry; rectification, warp and geometric resampling are all
+false. Boxes are not baked into the images.
+
+Worker 006 selected the cases from the immutable full native reference-graph
+ledger without another inference pass. The selection covers person, animal,
+light-road-user, vehicle, low-confidence, fisheye-edge and both directions of
+native-versus-legacy count divergence, with at least 12 source frames between
+selected cases. The source graph delivered all `4,489/4,489` frames at
+`11.84338 FPS`; detector p95 was `21.19895 ms`, world-state completion p95 was
+`47.940779 ms`, GPU utilization p95 was `53%`, and maximum used GPU memory was
+`9,718 MiB`. The isolated case-mining container was removed after execution and
+the canonical Triton predecessor was verified healthy and unchanged.
+
+This closes evidence preparation, not semantic quality. The cases are
+diagnostic samples rather than independent route truth; native-versus-legacy
+count differences are not a verdict because the legacy `704×704` path stretches
+the raw 4:3 image. Accordingly `review_ready=true`, while ground truth,
+candidate acceptance, production acceptance, navigation, safety, commands and
+actuation authority remain false. The next action is operator adjudication in
+the existing M4.8 review instrument.
+
 ## Implementation order
 
 The implementation sequence is intentionally strict: