Software Unit & Integration Test Report (SUITR)

Software Unit & Integration Test Report (SUITR)#

Project: Sentinel-2 MSI Synthetic L0 reconstruction pipeline (s2_msi_raw_generator) · DRD: ECSS-E-ST-40C Rev.1 (SUITR). Plan: SUITP. This report records the executed unit/integration campaign for the reverse-chain pipeline — a Sentinel-2B L1B run backwards through the exact inverse of the operational L0→L1B radiometric chain to reconstruct L1A → L0plus → Synthetic L0 — at the current baseline; validation-level results (S2 L1B E2E) are in the V&V report and the S2 L1B E2E validation.

1. Configuration baseline under test#

  • Repository ipf/s2-msi-raw-generator, branch main (v0.3.0 release baseline, 2026-07-02, plus documentation-only changes).

  • Latest main pipeline: #31052 — success (2026-07-02); blocking jobs unit-tests and pages both green.

2. Test environment executed#

  • CI: python:3.12-slim, pip install numpy pytest zarr, pytest tests/ -q --junitxml=report.xml (JUnit artifact retained on the job).

  • Local cross-check: Python 3.13.13, zarr v3 environment — same results (the zarr v2/v3 shim keeps both majors green; see DEC-09 in the DJF).

3. Results — grand totals#

Metric

Value

Test files

21

Test functions

131

Collected cases (with parametrisation)

206

Passed

201

Failed / errors

0

Skipped (environment-gated S2 L1B tests)

5

The 5 skips are the S2_GIPP_DIR / S2_L1A_INPUT-gated tests (SUITP §2); they pass when the data is supplied — verified on the SDE with the operational GIPP and the S2B L1B/L1A (evidence: the S2 L1B → Synthetic L0 run validated against the reference ESA L0 img, agreeing within ≤ ~4 DN on the 10/20 m bands; see the S2 L1B E2E validation and the V&V report).

4. Results by area#

All areas pass with 0 failures; the per-file inventory and item-under-test mapping is SUITP §3. Highlights of the exact (non-tolerance) assertions that passed:

  • Codec: compress_frame ∘ decompress_frame bit-exact on every fixture matrix and the env-gated S2 L1B window (test_ccsds122).

  • Packet grammar: reassemble_segments ∘ packetize_stream = identity; 14-bit counter continuity; offsets exactly tile the stream (test_isp_packetize).

  • Naming: parse_psfd_name round-trips every generated name, including flagged-default fields (test_naming).

  • Product contract: full 156-array L0 contract (12 det × 13 bands) with masks, STAC, sensor configuration and provenance (test_l0product, test_integration).

  • Radiometry: each inverse operator is unit-verified as the exact inverse of its operational forward operator — invert offset, relative-response/PRNU, dark, un-bin, SWIR re-stage, defective, crosstalk and on-board equalisation (rtol ~1e-9). MTF-deconvolution is OFF, so PSF and noise are not re-applied. Quantitative bounds vs typical values are tabulated in the V&V report §3.

5. Anomalies#

None open at this baseline. The historical doc/test bound-discrepancy disposition (closed) is recorded in the V&V report §5; the salted-hash() reseeding defect was fixed in the v0.3.0 cycle (CHANGELOG Fixed, REQ-QUAL-004 reproducibility — realized on the reverse and calibration paths).

6. Verdict#

PASS. All non-gated unit and integration tests pass at the current baseline; the environment-gated tests pass on the SDE with data. Exit criteria of the SUITP (§5) are met; open validation-level items are carried in the QR report.