Public FMV integration fixtures

The repository uses three publicly downloadable MPEG-2 Transport Streams, one independent negative-conformance bundle, and three independently authored VMTI vectors for opt-in end-to-end tests:

• Day Flight.mpg: 1280x720 H.264 daylight video and asynchronous KLVA.
• Night Flight IR.mpg: 1280x720 H.264 infrared video and asynchronous KLVA.
• Truck.ts: Esri's official FMV tutorial sample with H.264 video, AAC audio, and synchronous KLVA carried as fragmented Metadata Access Units.
• testfiles.zip: ImpleoTV STANAG4609 Inspector's public corpus of 21 deliberately damaged MPEG-2 Transport Streams.
• vmti_nested.klv, vmti_standalone.klv, and vmti_vtarget.klv: small Apache-2.0 vectors published by the independent libmisbklv implementation.

“Publicly downloadable” does not mean “openly licensed.” The FFmpeg directory and Esri item grant public download access but publish no media license. The project therefore records their source and hashes and downloads them only on demand; it does not redistribute them or describe them as open content.

For redistributable examples, the package generates two original synthetic FMVs from FFmpeg test sources and project-authored telemetry. They exercise H.264 video, Layer II audio, synchronous ST 0601, maps, recorded playback, and simulated-live playback without third-party footage:

stanag4609-player --demo day
stanag4609-player --demo thermal --simulate-live
stanag4609-demo-samples samples/demo --duration 30

The generated data is dedicated under CC0 1.0; the generator code remains MIT licensed. See DEMO_DATA_LICENSE.md in the source distribution.

These teaching streams are deliberately small and synthetic. They are not independent interoperability evidence and may contain verifier findings for profiles that require producer security policy, MIIS identity, or embedded video timestamps that a generic demo cannot truthfully invent.

The large media files are installed into the ignored samples/private/ directory rather than committed. The fetcher verifies Esri's source ZIP before extracting only the recorded video member and retains the ImpleoTV corpus as a ZIP so each test member can be authenticated before use. Source URLs, exact sizes, SHA-256 identities, stream observations, and expected results are recorded in references/fixtures.json.

python scripts/fetch_public_fixtures.py
pytest -m integration tests/integration/test_public_fmv.py

The FFmpeg-hosted flight files contain genuine, checksum-valid ST 0601 Local Sets. They also expose a useful deployed-system compatibility case: Target Width (Tag 22) is encoded in four bytes although the standard defines a two-byte value. The default strict decoder rejects this. FieldDecodingMode.PRESERVE retains the complete packet, decodes the conforming fields, and reports Tag 22 as a structured issue. This distinction prevents a permissive player from being mistaken for a conformance validator.

The daylight fixture also drives the complete FMVVerifier report. The integration test confirms that valid transport continuity, video/KLV discovery, and PCR cadence are reported as passes while every nonconforming four-byte Target Width is coalesced into one counted, tag-specific error.

The full daylight fixture is additionally passed through the no-op live transformer. The integration test fingerprints every reconstructed H.264 PES and complete KLV packet on both sides, proving elementary-stream identity. It also verifies all 13,331 source PCR observations are surfaced to the clock tap and retained in the transformed output through source-layout repacketization. The same real fixtures also pass through the constant-rate shaper without a clock anchor. Their complete output hashes remain byte-identical, proving that slot planning does not alter source packets unless PCR restamping is explicitly enabled.

Both fixtures satisfy the ST 1402.2 100 ms PCR interval under the exact program-aware clock validator, with no interval or regression diagnostics over 35,395 observed clock references. Both also build complete browser-player assets: FFmpeg produces playable H.264 MP4 media, the library produces exactly 6 daylight and 18 infrared timeline samples, and every sample exposes sensor, frame-center, and target geometry to the map and side panel.

The Esri fixture adds an independent audio-bearing, synchronous-KLVA case. It is also a valuable negative conformance sample: the current file has no metadata standard descriptor, contains Metadata AU sequence discontinuities, and exceeds the ST 1402 PCR interval in places. The integration test therefore asserts both successful H.264/AAC/KLV discovery and those concrete diagnostics; when the audio-pyav extra is installed, it also decodes all 6,946 timestamped AAC access units to 7,112,704 stereo PCM samples at 48 kHz through the public demux, audio timing, frame parser, and decoder APIs. Being playable in an FMV product is not treated as proof of conformance. The same fixture drives a complete reference-player asset build: all 711 displayable metadata samples are decoded, H.264/AAC is transcoded to a seekable MP4, the static UI is copied, and the first metadata sample is checked against the earliest audio/video PTS. This retains the source's 2,777-tick AAC lead-in instead of displaying every metadata update early relative to the MP4 clock.

Independent negative-conformance corpus

ImpleoTV publishes a small test-data-only release asset alongside its proprietary Inspector product. The project downloads only that public ZIP, not the application or installer. The integration suite verifies the ZIP identity, then separately verifies the size and SHA-256 of 20 representative members before asserting a fault-specific library diagnostic for each one. Covered families include corrupt KLV checksum/key/length, duplicate and missing ST 0601 items, missing KLV payload, continuity-counter errors, duplicate transport packets, excessive PTS/PCR/PAT/PMT intervals, PCR regression, invalid PES length, missing PAT/PMT, corrupt DVB SDT CRC, corrupt TS sync, and the transport-error indicator.

One archive member remains explicitly unasserted in the manifest: mpegts-pcr-pts-drift.ts. Its H.264 PTS leads PCR by approximately 0.600 to 0.899 seconds, which remains inside ST 1402 §7.4's published maximum ten-second lead. The corpus label alone is not treated as a normative failure, and the verifier does not invent an unsupported drift-rate threshold.

The corpus contains no ST 0903/VMTI packets. A separately examined OpenSensorHub UAS sample stream also contains 277 ST 0601 packets but no nested Item 74 or standalone ST 0903 Local Set. Although it is used by an upstream VmtiTest, that test permits zero targets, so it is not accepted as VMTI interoperability evidence here.

The three libmisbklv vectors are pinned to an exact upstream commit and downloaded rather than redistributed. All have valid outer packet checksums, but each contradicts a mandatory ST 0903.6 relationship: a nested packet reports two targets without Item 101, one standalone packet omits mandatory MIIS Item 13, and the other standalone packet omits mandatory FOV and MIIS items. The integration tests assert the exact strict-decoder rejection for each case. These independent negative vectors strengthen cross-implementation conformance evidence, but they are hand-authored KLV rather than a recorded VMTI FMV stream and therefore do not close the real-stream interoperability gap.

Additional public fixtures, especially ST 0903 VMTI and ST 1001-labelled audio streams, should be added to the manifest when found.

The separate ArcGIS multiplexer tutorial's Raw_Video.mpeg is intentionally not listed as an FMV fixture because it is an MPEG Program Stream with video and audio but no KLV. Its 866-row Raw_Metadata.csv companion is exercised as a producer input. The generated transport contains all 866 decodable ST 0601 packets, MPEG-2 video, and MPEG-1 Layer II audio; it passes the verifier's structural profile without warnings or errors and builds a player timeline with 866 geospatial samples. The FFmpeg remux command explicitly requests a 20 ms PCR period so the generated output remains within ST 1402's 100 ms limit.