docs(mission): record accepted M4.8R3 shadow
This commit is contained in:
@@ -1504,6 +1504,42 @@ anchors onto its immutable timeline; no third perception instrument exists.
|
||||
Navigation, commands, actuation, physical clearance and collision-safety
|
||||
authority remain false.
|
||||
|
||||
### 2026-08-26 — M4.8R3 bounded static-occupancy Worker shadow accepted
|
||||
|
||||
M4.8R3 closes the bounded Worker integration requested by M4.8R2. The sealed
|
||||
result is
|
||||
`m48r3-static-occupancy-shadow-d1577870b098bcd0b67cc51798234f224b348ad3954ead72f3ce6df414875a29`.
|
||||
It composes the accepted camera/geometry graph with an occupied-only low-step
|
||||
provider; RF-DETR, the native `800×600` fisheye raster and the semantic risk
|
||||
policy remain unchanged. Weak two-to-four-point geometry is published only
|
||||
when the exact `0.45 m` voxel is present in all six causal frames and its
|
||||
current range is at most `8 m`. Strong five-point components remain immediate.
|
||||
The provider never uses a semantic class, clears a cell, infers free space or
|
||||
runs another neural-network pass.
|
||||
|
||||
The isolated Worker 006 replay delivered all `4,489/4,489` frames at
|
||||
`11.79902 FPS`. World-state completion p95 was `56.74324 ms`; geometry p95/p99
|
||||
were `8.420913/13.732692 ms`; GPU utilization p95/maximum were `54%/58%` and
|
||||
maximum used GPU memory was `9,743 MiB`. The additive provider contributed a
|
||||
mean `0.587893 ms` per frame. Relative to the frozen native baseline, FPS
|
||||
regression was `0.3745552%` and world-state p95 increased by `8.802461 ms`,
|
||||
both inside the predeclared envelope.
|
||||
|
||||
All `9/9` critical-near assisted anchors are occupied in the delivered graph.
|
||||
At display frame `1856`, the two thin posts remain independent (`5` observed
|
||||
components in their review extent) and the two concrete hemispheres remain
|
||||
independent (`6` observed components); the system does not inherit the
|
||||
operator's old paired boxes. The ledger adds `613,146` occupied cells, loses
|
||||
zero baseline cells, makes zero false-free claims and records zero capacity
|
||||
drops. The two `8–12 m` approach anchors intentionally remain unknown because
|
||||
the sparse-persistence policy is bounded to the critical `0–8 m` band.
|
||||
|
||||
This accepts the reproducible bounded Worker shadow and its LAB evidence, not a
|
||||
production navigation cutover. The result remains replay-simulated;
|
||||
physical-live authority, commands, actuation, planner-authoritative free space
|
||||
and collision-safety acceptance are false. The next milestone boundary is
|
||||
M4.9 recorded-realtime release-candidate validation using this frozen graph.
|
||||
|
||||
## Implementation order
|
||||
|
||||
The implementation sequence is intentionally strict:
|
||||
|
||||
Reference in New Issue
Block a user