Wire 4 stateless / cached helpers into airborne_bootstrap.build_pre_constructed: c282_ransac_filter, c5_imu_preintegrator (cached on calibration path), c5_se3_utils (helpers.se3_utils module as namespace handle), c5_wgs_converter. The original AZ-623 5th deliverable (c5_isam2_graph_handle) hit an unresolvable construction-order conflict between c4_pose (consumes the handle) and c5_state (creates it inside build_state_estimator's tuple return) under the umbrella's "MUST NOT touch any per-component factory signature" constraint. Per AZ-623 spec's escalation gate, scope was split: AZ-625 captures the handle ordering work; AZ-624 dependency edge updated to require both. Tests: tests/unit/runtime_root/test_az623_pre_constructed_phase_e.py adds 7 tests covering AC-623.1..3 (4 new keys + correct types, IMU preintegrator caching, operator-actionable error messages for empty / unreadable / malformed calibration paths). Autouse stubs added to test_az619/620/621/622 so prior phase tests remain isolated from new builders. Quality gates: ruff format clean, ruff lint clean, 24/24 phase tests pass, 247/247 runtime_root + c5_state regression suite passes. Code review verdict PASS_WITH_WARNINGS (3 Low findings; full report in _docs/03_implementation/reviews/batch_94_review.md). Co-authored-by: Cursor <cursoragent@cursor.com>
6.9 KiB
Batch Report
Batch: 94
Tasks: AZ-623 (Phase E narrowed: build_pre_constructed seeds c282_ransac_filter + c5_imu_preintegrator + c5_se3_utils + c5_wgs_converter; c5_isam2_graph_handle work split out to AZ-625)
Date: 2026-05-19
Cycle: 1
Task Results
| Task | Status | Files Modified | Tests | AC Coverage | Issues |
|---|---|---|---|---|---|
| AZ-623_pre_constructed_phase_e_ransac_c5_helpers (narrowed) | Done | 7 files | 24 passed | 4/4 ACs covered | 0 blocking |
Original AZ-623 had a 5th deliverable (c5_isam2_graph_handle) that uncovered an unresolvable construction-order conflict between c4_pose (consumes the handle) and c5_state (creates the handle inside build_state_estimator's tuple return) under the umbrella's "MUST NOT touch any per-component factory signature" constraint. Per the AZ-623 spec's own escalation gate ("if the resolution requires a Protocol seam change, escalate"), the question was surfaced; user chose to split scope:
- This batch lands the 4 stateless / cached helpers (narrowed AZ-623).
- New PBI AZ-625 ("AZ-618 Phase E.5: c5_isam2_graph_handle ordering", 3pt, parent AZ-618) holds the handle work.
- AZ-624 dependency edge updated to require both AZ-623 AND AZ-625.
Files Changed
Production
src/gps_denied_onboard/runtime_root/airborne_bootstrap.py— added imports (json,pathlib.Path,numpy,_types.calibration.CameraCalibration,helpers.imu_preintegrator.{ImuPreintegrator, make_imu_preintegrator},helpers.ransac_filter.RansacFilter,helpers.wgs_converter.WgsConverter); added module-level_IMU_PREINTEGRATOR_CACHEdict +clear_imu_preintegrator_cache()helper; added_load_camera_calibration(mirrors_replay_branch._load_camera_calibrationbut raisesAirborneBootstrapError); added 4 builders_build_c282_ransac_filter,_build_c5_imu_preintegrator(cached on calibration path),_build_c5_se3_utils(returns thehelpers.se3_utilsmodule as a namespace handle, matching existing C5 estimator's MagicMock fixture pattern),_build_c5_wgs_converter; extendedbuild_pre_constructedto populate the 4 new keys after the existing AZ-619..AZ-622 keys.
Tests
tests/unit/runtime_root/test_az623_pre_constructed_phase_e.py(NEW, 7 tests):test_ac_623_1_adds_c282_ransac_and_c5_helpers— AC-623.1 (4 keys + correct types).test_ac_623_1_keeps_existing_keys_intact— additivity invariant (AZ-619..AZ-622 keys still present).test_ac_623_2_imu_preintegrator_cached_across_calls— AC-623.2 (cache short-circuit via per-test counter).test_ac_623_2_imu_preintegrator_per_path_cache— AC-623.2 (per-path cache isolation).test_ac_623_3_empty_calibration_path_raises_named_error— AC-623.3 (operator-actionable error).test_ac_623_3_unreadable_calibration_path_raises_named_error— AC-623.3 (file not found path).test_ac_623_3_malformed_json_raises_named_error— AC-623.3 (JSON decode wrap).
tests/unit/runtime_root/test_az619_pre_constructed_phase_a.py— added autouse_stub_c5_builders.tests/unit/runtime_root/test_az620_pre_constructed_phase_b.py— added autouse_stub_c5_builders.tests/unit/runtime_root/test_az621_pre_constructed_phase_c.py— added autouse_stub_c5_builders.tests/unit/runtime_root/test_az622_pre_constructed_phase_d.py— added autouse_stub_c5_builders.
Specs
_docs/02_tasks/todo/AZ-623_*.md→ narrowed (handle work removed); ARCHIVED to_docs/02_tasks/done/._docs/02_tasks/todo/AZ-624_*.md— Dependencies updated (AZ-625 added)._docs/02_tasks/todo/AZ-625_c5_isam2_graph_handle_ordering.md(NEW) — 3pt, parent AZ-618, holds the handle ordering work.
AC Test Coverage: 4 of 4 covered
| AC | Test | Status |
|---|---|---|
| AC-623.1 | test_ac_623_1_adds_c282_ransac_and_c5_helpers + test_ac_623_1_keeps_existing_keys_intact |
Covered |
| AC-623.2 | test_ac_623_2_imu_preintegrator_cached_across_calls + test_ac_623_2_imu_preintegrator_per_path_cache |
Covered |
| AC-623.3 | test_ac_623_3_empty_calibration_path_raises_named_error (+ 2 variants for unreadable/malformed) |
Covered |
| AC-623.4 | File tests/unit/runtime_root/test_az623_pre_constructed_phase_e.py exists |
Covered |
Code Review Verdict: PASS_WITH_WARNINGS
Full report: _docs/03_implementation/reviews/batch_94_review.md. Three Low findings:
- F1 (Low / Maintainability) —
_load_camera_calibrationduplicates_replay_branch._load_camera_calibration(different exception class). Defer to a hygiene PBI (~2pt) bundled with batch 93's F2 leftover. - F2 (Low / Maintainability) — empty-path check duplicated across
_build_c5_imu_preintegratorand_load_camera_calibration(defense-in-depth). No change recommended. - F3 (Low / Style) —
c5_se3_utilsreturns a Python module as a namespace handle. Documented; matches existing C5 test fixture pattern. YAGNI — defer Protocol introduction.
No Critical / High / Medium findings. Auto-fix not invoked.
Auto-Fix Attempts: 0
Stuck Agents: None
Test Run Summary
- Targeted test set (AZ-619..AZ-623): 24 passed in 1.07s.
- Regression check (
tests/unit/runtime_root/+tests/unit/c5_state/): 247 passed in 1.78s.
Scope-Split Decision Trail
The split was driven by an irreconcilable construction-order issue between two umbrella constraints:
- Constraint A (umbrella AZ-618): "MUST NOT touch any per-component factory signature" + "All changes confined to runtime_root/airborne_bootstrap.py, runtime_root/__init__.py, and the new test file."
- Constraint B (existing seam):
build_state_estimatorreturns(StateEstimator, ISam2GraphHandle)as a tuple — the handle is a structural side-product of the estimator, not separately constructable. - Constraint C (existing seam): C4's
build_pose_estimatorconsumesc5_isam2_graph_handlefrompre_constructed, socompose_roottopologically requires it to exist before C4 runs, which is before C5 runs.
Possible resolutions all touched at least one of A/B/C; the user-approved Option B preserves all three constraints by:
- Keeping the C4 ↔ C5 seam intact (no Protocol changes).
- Promoting the (estimator, handle) build into
airborne_bootstrap, where AZ-618 explicitly permits orchestration. - Documenting the strategy in AZ-625's spec § "Decision" with full rationale.
Tier-2 / Deferred Work
- AZ-625 (3pt) — handle ordering work, blocks AZ-624.
- Hygiene PBI to consolidate
_load_camera_calibrationbetween airborne_bootstrap and_replay_branch(F1) and_is_build_flag_ontriple-duplication (batch 93 F2). Bundle as ~2pt cleanup PBI post-AZ-618.
Next Batch
- Batch 95: AZ-625 (Phase E.5: c5_isam2_graph_handle ordering) — 3pt.
- Then Batch 96: AZ-624 (Phase F: wire main() + AC-1..AC-5 verification incl. Jetson tier-2) — 2pt.
- Cumulative review window: next due at batch 96 (K=3 from last cumulative at 88-92).