docs(m4): record E47 semantic shadow evidence
This commit is contained in:
DCCONSTRUCTIONS
@@ -2,7 +2,7 @@
|
||||
|
||||
Date: 2026-08-05
|
||||
|
||||
Status: in progress; M4.0–M4.6 accepted, M4.7 Worker 006 cutover is next
|
||||
Status: in progress; M4.0–M4.6 accepted, E47 diagnostic semantic/SLAM shadow is parallel, M4.7 Worker 006 cutover remains next
|
||||
|
||||
Audit base: `1b3e0b3` on `feat/simulation-polygon-s1`
|
||||
|
||||
@@ -112,7 +112,7 @@ Those systems remain separate platform workstreams.
|
||||
| Worker graph | The persistent process still executes `run_e15_shadow_inference.py serve` with E15/E19/E8/E3/E23 profiles | Open architectural blocker: runtime remains LAB-generation-specific |
|
||||
| Model service | Canonical Triton currently exposes pinned `yolox_s` and `pointpillars`; PointPillars was rejected as a K1 product candidate | Reuse `yolox_s`; do not reopen PointPillars |
|
||||
| CV provider interfaces | No executable `DetectorProvider`, `GeometryProvider`, `TemporalProvider` or object-map contract exists in the product package | Open architectural blocker |
|
||||
| LAB execution | 31 evidence definitions are fully classified; E33, E35 and E46J are canonical canaries, the remaining 28 are bounded legacy | Closed for LAB governance; no bulk legacy migration |
|
||||
| LAB execution | 33 evidence definitions are fully classified; M4, E33, E35 and E46J are canonical canaries, E47 is one bounded experimental adapter, and 28 remain read-only legacy | Closed for LAB governance; no bulk legacy migration |
|
||||
| Production import graph | 93 compute modules (72,456 lines), 51 web modules and many direct `web → compute.e*/l*` imports remain | Do not clean wholesale; prohibit new dependencies and isolate the new graph |
|
||||
| Telemetry | Generic pipeline telemetry, multi-contour enrollment, bounded journal and current Worker 006 Telegraf service exist | Reuse; add graph-stage and object-state metrics |
|
||||
| Independent truth | E46/E47/L3.4 freeze paths exist, but the two real reviews/adjudication are incomplete | Existing class truth is not a Milestone 4 blocker; a smaller object-centric truth contract is required |
|
||||
@@ -148,6 +148,51 @@ The following lines of research remain frozen unless a later measured gate fails
|
||||
- a DeepStream rewrite without a demonstrated runtime need;
|
||||
- mass migration or deletion of historical LAB implementations.
|
||||
|
||||
## E47 parallel semantic/SLAM shadow (2026-08-06)
|
||||
|
||||
E47 is an isolated experiment over the accepted M4.6 replay, not a new M4 stage
|
||||
and not a reason to renumber or postpone M4.7. It answers one narrower question:
|
||||
can an immutable semantic mask be bound to the exact RAVNOVES00 camera frame,
|
||||
projected through the admitted factory KB4 calibration onto the complete
|
||||
frame-local SLAM/LiDAR point index space, and aggregated for observations which
|
||||
already exist in the M4 geometry ledger?
|
||||
|
||||
The seam is deliberately model-neutral:
|
||||
|
||||
- a provider supplies a source/frame/model/preprocess-bound `uint8` class-ID mask;
|
||||
- the admitted KB4 projection assigns a semantic status to each source point;
|
||||
- the four statuses remain distinct: `labeled`, `ambiguous`, `unprojected` and
|
||||
`absent`;
|
||||
- observation semantics are detached diagnostics over immutable observation and
|
||||
source-point IDs;
|
||||
- semantic absence never means free and semantic output cannot create an
|
||||
obstacle or change identity, metric geometry, occupancy, motion or threat;
|
||||
- navigation, safety, command, physical-live and actuation authority remain
|
||||
false.
|
||||
|
||||
The full-route E4 EoMT result is the fixed control provider because its 4,489
|
||||
masks, model revision, weights digest and input identity already exist as sealed
|
||||
evidence. This does not promote EoMT to the production model. A later NVIDIA
|
||||
CitySemSegFormer or other candidate must enter through the same provider seam and
|
||||
be compared on frozen independent truth. `supervision` may be used as rendering
|
||||
or evaluation tooling, but it is not itself a segmentation model.
|
||||
|
||||
The temporal claim is intentionally narrower than physical synchronization.
|
||||
All `4,489` E4 mask rows and the sealed E10 source-pack entries have identical
|
||||
camera ordinals and identical recorded `session_seconds` (zero difference after
|
||||
conversion to nanoseconds), so E47 now verifies that binding fail-closed for
|
||||
every frame. The LiDAR and pose entries inside E10 still inherit the admitted E6
|
||||
`nearest-host-arrival-best-effort` binding. Their recorded per-frame deltas are
|
||||
checked against the existing `100 ms` limits, but this does not prove a shared
|
||||
hardware clock or zero camera-to-LiDAR skew. E47 therefore remains warning-grade
|
||||
diagnostic evidence and cannot be described as physically synchronized.
|
||||
|
||||
The reusable UI reads the same recorded clock in `VIDEO`, `CAMERA`, `3D` and
|
||||
`PLAN`: the mask is drawn below existing boxes and the matching semantic class
|
||||
IDs color only the exact current point increment. Any unknown mask class,
|
||||
point-index mismatch or artifact-accounting drift is an explicit integrity
|
||||
failure rather than an empty or cosmetically clean layer.
|
||||
|
||||
## Target reference graph
|
||||
|
||||
```text
|
||||
|
||||
Reference in New Issue
Block a user