[AZ-599] Batch 79: FT-P-02 Derkachi builder + _common.py extraction

- Add build_p02_fixtures.py: IMU CSV → tlog conversion (RAW_IMU +
  ATTITUDE pairs, centidegrees→radians yaw) and orchestrator that
  runs gps-denied replay against Derkachi MP4 + generated tlog,
  verifying ≥1 record_type="estimate" in the FDR archive.
- Extract run_gps_denied_replay + FDR-parent-dir helpers into
  sitl_replay_builder/_common.py; refactor build_p01_fixtures.py
  to import from _common (b78 tests preserved).
- Add 20 unit tests under e2e/_unit_tests/fixtures/test_sitl_
  replay_builder_p02.py covering AC-1..AC-5; total unit suite
  686/686 passing (regression gate AC-6).
- README updated to document FT-P-01 + FT-P-02 builders.
- Advance autodev state: last_completed_batch=79, current_batch=80;
  prune verbose detail blob.

Co-authored-by: Cursor <cursoragent@cursor.com>
This commit is contained in:
Oleksandr Bezdieniezhnykh
2026-05-17 13:40:07 +03:00
parent 2f1fb4d0d0
commit 4e0717e543
10 changed files with 1111 additions and 76 deletions
@@ -0,0 +1,88 @@
# Batch 79 Report — FT-P-02 Derkachi fixture builder
**Batch**: 79
**Date**: 2026-05-17
**Context**: Test implementation (greenfield Step 10 — Implement Tests)
**Tasks**: AZ-599 (3 cp) — 1 task (FT-P-02 Derkachi builder + b78 helper extraction)
**Cycle**: 1
**Verdict**: COMPLETE — PASS (self-reviewed; see `reviews/batch_79_review.md`)
## Summary
Replicates the b78 vertical-slice pattern for FT-P-02 (Derkachi
continuous-flight drift). Different from b78 in three ways:
1. **Video is already MP4** — no still-image encoding step.
2. **IMU is real recorded telemetry** — needs CSV → tlog conversion
with real motion data (vs. b78's synthetic stationary tlog).
3. **Output is the SUT's natural FDR archive directory** — FT-P-02
reads it via `fdr_reader.iter_records`, not via the per-call
`outbound_messages_*.json` schema b78 ships.
Also extracted `run_gps_denied_replay` + `write_observer_fixture`
into a shared `_common.py` so both b78 and b79 reference one
canonical implementation. Future per-scenario builders (FT-P-04
also uses Derkachi inputs, FT-P-05 needs its own, etc.) will reuse
the same helpers.
### AZ-599 — FT-P-02 Derkachi builder (3 cp)
* **`e2e/fixtures/sitl_replay_builder/_common.py`** (new): shared
`run_gps_denied_replay`, `write_observer_fixture`, `DEFAULT_CLI_BIN`.
* **`e2e/fixtures/sitl_replay_builder/build_p01_fixtures.py`**
(edited): re-imports the moved helpers from `_common.py`; b78
tests still pass unchanged.
* **`e2e/fixtures/sitl_replay_builder/build_p02_fixtures.py`** (new):
* `P02BuilderConfig` frozen dataclass.
* `convert_imu_csv_to_tlog(csv_path, output_tlog)` — parses
`data_imu.csv` rows, packs RAW_IMU + ATTITUDE pairs. Yaw from
`GLOBAL_POSITION_INT.hdg` (centidegrees → radians);
roll/pitch = 0 (fixed-wing cruise approximation).
* `verify_fdr_has_estimates(fdr_path)` — counts
`record_type="estimate"` records; raises if zero.
* `build_p02_fixtures(cfg, ...)` — orchestrator.
* `_main(argv=None)` — argparse CLI.
* **`e2e/fixtures/sitl_replay_builder/README.md`** (edited): two-
builder structure with scenario coverage table; FT-P-01 + FT-P-02
per-builder sections with strategy + usage + limitations.
* **`e2e/_unit_tests/fixtures/test_sitl_replay_builder_p02.py`**
(new): +20 tests.
* **`e2e/_unit_tests/test_directory_layout.py`** (edited): registers
the two new modules (`_common.py` + `build_p02_fixtures.py`).
## Out of scope (deferred)
* **Live capture EXECUTION** — same as b78, a manual operator step.
* **FT-P-04** (Derkachi F2F registration) — also uses Derkachi
inputs and could reuse `build_p02_fixtures`. Deferred to a
follow-up so this batch ships at 3 cp.
* **Other scenarios** (FT-P-03 / 05 / 07 / 08 / 10 / 11, FT-N-01..04).
* **iNav adapter** for FT-P-02.
* **True SCALED_IMU2 → RAW_IMU unit conversion** — pass-through;
re-visit if the SUT's tlog parser rejects.
* **Aggressive-manoeuvre ATTITUDE synthesis** (roll/pitch ≠ 0) —
re-visit if FT-P-04 or other scenarios show fusion drift.
## Test Results
* New unit tests: **20**.
* Full `e2e/_unit_tests` suite: **686 passed in 123 s** (previous:
664 → +22 net = +20 new tests + +2 directory-layout entries).
* No new linter errors.
* b78 builder tests (24) still pass after `_common.py` extraction.
## State
* Spec moved: `_docs/02_tasks/todo/AZ-599_ft_p_02_derkachi_builder.md`
`_docs/02_tasks/done/`.
* `_docs/_autodev_state.md` advanced to `last_completed_batch: 79`.
* No K=3 cumulative review this batch (next due after b81; the
b76-b78 cumulative just shipped one batch ago).
## Incidents
* **Disk-full during regression gate**: the first regression-gate
attempt failed with 212 conftest setup errors (all
`FileNotFoundError: /var/folders/.../pytest-of-...`). Surfaced
to the user with options; user freed space (37 Gi avail after);
re-ran the gate and it passed cleanly. Not a code issue.