Files
gps-denied-onboard/_docs/03_implementation
Oleksandr Bezdieniezhnykh fbeeab60b3 [AZ-337] [AZ-338] [AZ-508] Cumulative review batches 46-48
Verdict: PASS_WITH_WARNINGS. Per-batch reviews already validated
each task's ACs; this cumulative review focuses on cross-batch
drift and surfaces 1 Medium + 2 Low maintainability findings:

- F1 (Medium): `_iso_ts_from_clock` Clock-injected helper duplicated
  across 4 files (c2_vpr/net_vlad + ultra_vpr + _faiss_bridge,
  c12_operator_orchestrator/operator_reloc_service). B46 + B47
  carry inline comments anticipating AZ-508 would consolidate this,
  but AZ-508 (Batch 48) scoped itself narrower (stdlib-only,
  Excluded the Clock-injected variant). Recommend a 2-point follow-up
  PBI adding `iso_ts_from_clock(clock)` to helpers/iso_timestamps.py
  before AZ-339 / AZ-340 / AZ-358 / AZ-389 add more copies.

- F2 (Low): `_assert_engine_output_dim` near-duplicated between
  NetVLAD and UltraVPR. Defer consolidation until 5 c2_vpr strategies
  are in flight (after AZ-339 / AZ-340).

- F3 (Low): Clock-driven helper outputs `+00:00`; canonical FDR `ts`
  is `Z`. Fold into F1 follow-up PBI.

No Critical or High findings; auto-chain to next batch allowed.

Co-authored-by: Cursor <cursoragent@cursor.com>
2026-05-13 23:26:58 +03:00
..