feat(perception): freeze blind detector gates

docs/01_IMPLEMENTATION_PLAN.md

@@ -14,6 +14,9 @@ Each gate produces evidence and an explicit GO, PAUSE or BLOCKED result.
 | Native pipeline telemetry | GO for producer/normalizer contract — lifecycle events and JSONL/MQTT sink boundaries exist and an E41 smoke run records real stage accounting. Durable Worker 006 MQTT wiring/deployment remains pending. |
 | Evidence storage | MEASURED — E44 finds 525,471,092 logical bytes, 312,753,179 unique-content bytes and 1.680146× amplification across 14 E30–E40 roots. Exact-content references/deduplication precede any format migration. |
 | Future transfer | PREREGISTERED — E43 freezes same-K1/mount/calibration/firmware, required streams, connected-component split and independent label reveal. Capture and labels do not yet exist. |
+| E31 binding sensitivity | MEASURED — E45 closes accounting for 87/87 accepted correspondences and finds no material monotonic residual association with represented image radius, rig speed or pose age. It does not supply calibration-target truth or outer-fisheye coverage. |
+| Detector Truth Island | PREPARED — E46 freezes 32 references with no prelabels, predictions, scores or candidate identity in the review package. Two independent reviews and adjudication are still required. |
+| Detector candidate comparison | FROZEN BEFORE TRUTH — E47 freezes raw-KB4 and fixed-valid-FOV-fill predictions from the same exact Mask R-CNN checkpoint. No accuracy result or winner exists before the E46 truth seal. |
 | Product interface | DEFERRED — no new windows, page anatomy or design changes are part of this stabilization increment. |
 
 The governing decision is

docs/20_RAVNOVES00_REFERENCE_SOURCE_PRODUCT_PLAN.md

@@ -142,6 +142,34 @@ across 14 admitted E30–E40 roots: `1.680146×` amplification and
 content-addressed referencing and exact deduplication, not an unmeasured
 format/database migration.
 
+E45 immutable diagnostic
+`e45-binding-sensitivity-b10fa117f110d5ac7942f73c62f1300c7d9ce833a187542b992e58fb24d1be72`
+accounts for all 87 accepted E31 correspondences and stratifies the existing
+support-centroid residual by KB4 image radius, rig motion and recorded source
+age. Residual rank association is small for image radius (`ρ=0.0421`),
+translation speed (`ρ=0.0076`) and pose age (`ρ=-0.0181`), so the accepted E31
+zero-offset binding remains unchanged. This is not calibration-target truth:
+the outer fisheye stratum is empty, measured static-landmark reprojection and
+physical mount dimensions remain unavailable, and calibrated-perception P0
+stays open.
+
+E46 immutable preparation
+`e46-detector-truth-island-d8ab2745679636dce374b720b562fce05d6d0a26be3eac88650224d7aa92267d`
+freezes 32 references: 16 anchors across eight route-time bins and all 16
+frames from four temporal groups. The reviewer package contains no model
+prelabels, predictions, scores or candidate identity. It references
+14,919,621 source bytes without copying images. The package is prepared, not
+truth: two different human reviewers and adjudication remain mandatory.
+
+E47 immutable candidate freeze
+`e47-detector-candidate-freeze-514bcca7a8a26313cab5ffcacca053d7a0ec6fe7cbef25f15faf3a11e48ee92f`
+stores raw-KB4 and fixed-valid-FOV-fill predictions for those exact 32 frames.
+Both candidates use the same Mask R-CNN checkpoint; only preprocessing differs.
+The freeze contains 417 and 394 task-ontology predictions respectively, but
+these counts are descriptive rather than accuracy evidence. No truth was
+joined, no winner was selected and no retraining is authorized before the E46
+truth seal.
+
 E43 immutable protocol
 `e43-future-capture-protocol-28f091b9648daffce988d44c183e21f56d77988061630de934f8003fb13701d8`
 preregisters the later same-K1/new-route transfer. It requires the same mount,
@@ -177,6 +205,11 @@ The current non-UI priority order is:
 4. design exact-content references/deduplication from E44 before any storage
    migration;
 5. preserve E43 unchanged until the owner supplies the new capture.
+6. complete two independent E46 reviews and adjudication without exposing
+   model prelabels or E47 predictions;
+7. reveal the sealed truth only after the E47 prediction freeze, then calculate
+   the preregistered detector metrics and select or reject a preprocessing
+   candidate.
 
 The following remain outside the present stabilization increment: