docs(perception): record m4 temporal acceptance

docs/22_MILESTONE_4_OBJECT_CENTRIC_RECORDED_REALTIME_CV.md

@@ -2,7 +2,7 @@
 
 Date: 2026-08-05
 
-Status: in progress; M4.0–M4.4 accepted, M4.5 temporal/motion is next
+Status: in progress; M4.0–M4.5 accepted, M4.6 replay threat is next
 
 Audit base: `1b3e0b3` on `feat/simulation-polygon-s1`
 
@@ -844,8 +844,72 @@ observation bytes.
 All ten M4.4 acceptance requirements in the immutable result are true. This is
 algorithmic and contract acceptance on the recorded source, not a physical-live
 claim and not a Worker 006 cutover. No class name participates in association or
-range. M4.4 is closed; M4.5 temporal retention and class-independent motion is
+range. M4.4 is closed.
-the next implementation phase.
+
+### 2026-08-05 — M4.5 bounded temporal retention and motion acceptance
+
+The accepted E34 state machine and E51 diagnostic motion thresholds were
+extracted into product-owned `k1link.perception.temporal` and
+`k1link.perception.motion`. Neither runtime closure imports `k1link.compute`, an
+E/L experiment module or a device plugin. The versioned profile
+`config/perception/m4-temporal-motion-v1.json` binds the exact M4.4 frame ledger
+and freezes the following semantics:
+
+- only current qualified source points create occupied cells;
+- all components use ephemeral identities and a bounded eight-sample history;
+- association is deterministic spatial reassociation and uses neither detector
+  identity nor semantic class;
+- camera-only/conflicting evidence remains nonmetric uncertainty and cannot
+  create occupied temporal state;
+- current evidence is hit-backed, held evidence is `unknown`, and expired
+  evidence publishes no cells;
+- the occupied TTL is `0.75 s`, with exact deadline semantics independent from
+  the later replay frame that materializes the expiry event;
+- moving/static decisions use only map-frame centroid history, explicit span,
+  displacement, speed and evidence-sufficiency thresholds;
+- insufficient history, stale support, map-frame discontinuity, implausible
+  speed and the conservative threshold deadband all publish `unknown`.
+
+The canonical provider consumed the immutable M4.4 ledger and sealed a complete
+local Mission Core replay:
+
+- result ID:
+  `m4-temporal-replay-9ed5dcd249ed3bcb81661dd18e2b854a7ffedf3fd2b92b9c994c3c70c34533f2`;
+- frame ledger SHA-256:
+  `1bf1365bdb3f20214443d3f8b87a0fa88f9848af8ca0456b7ca364d37631c3fc`;
+- manifest SHA-256:
+  `c01692d18393877acd432bc6d7b829b8f9f70b36814e0ddfc3c1a31cab65daac`;
+- report SHA-256:
+  `5327782f3ee55bac61a7003fd594adb50f53cb10538d25cbc90a38e3a31ae7a4`;
+- `4,489 / 4,489` frames, `0` failed and all `37,457` M4.4 observations
+  accounted for;
+- `27,299` current occupied inputs and `10,158` nonmetric uncertainty inputs;
+- `4,305` ephemeral components created and `22,994` frame-local detector
+  identity changes survived spatial reassociation;
+- `27,299` current, `33,707` held and `4,288` expiry publications;
+- peak state was `34` active components, `324` cells per component and exactly
+  `8` history samples;
+- `11,365` moving, `1,451` stationary and `52,478` unknown publications;
+- zero map-frame jump candidates and zero implausible-speed publications;
+- maximum held age exactly `0.75 s`, zero occupied cells after the TTL deadline
+  and zero false-free claims;
+- expiry deadline delay `0 ms`; maximum discrete replay materialization delay
+  `283 ms`, identical to the separately reported accepted E34 measurement and
+  not an occupied ghost interval;
+- local temporal-plus-motion latency p50 `0.902 ms`, p95 `1.973 ms`, maximum
+  `5.057 ms`.
+
+Sixteen E46B source-scoped engineering frames are included as explicit aggregate
+clip checks. They remain non-independent and have no component correspondence,
+so they are evidence coverage, not a moving/static accuracy claim. Independent
+object-centric correctness remains the M4.8 gate. A separate repeat execution
+produced the same frame-ledger SHA-256, proving that runtime timing does not alter
+canonical temporal or motion bytes.
+
+All fifteen M4.5 acceptance requirements are true. Synthetic tests cover ID and
+semantic-hint changes, moving, stationary, held, expiry, camera-only uncertainty
+and map-frame discontinuity. M4.5 is closed; M4.6 replay-only threat assessment
+is the next implementation phase.
 
 ## Implementation order