serendipitylxd/NavLock-BreachSynth

NavLock-BreachSynth

NavLock-BreachSynth is a fixed, test-only multimodal counterfactual benchmark for navigation-lock warning reliability assessment. It pairs each generated hazard scene with its factual source observation in NavLock-HY through the stable sample_idx field. Project code is available at <https://github.com/serendipitylxd/NavLock-BreachSynth>.

Release Scope

• —123 fixed test scenes and 1186 synchronized frames;
• —eight camera images per generated frame;
• —merged LiDAR and inserted-vessel point files per frame;
• —frame labels, generation metadata, final public expert labels, operational context, warning protocols, and aggregate statistics;
• —frozen RTMDet, Hydro3DNet, and warning-chain predictions without model checkpoints.

The hazard-stage distribution is 29 entry-path-not-clear, 81 unauthorized-entry, and 13 insufficient-capacity scenes. Generated-vessel 3D evidence is complete in 90 scenes and missing or incomplete in 33 scenes; background/original LiDAR remains available in the latter group.

Files

The sensor payload is compressed into 19 deterministic tar.gz archives totaling approximately 6.86 GiB. A scene is never split across archives. Extract every shard into the same destination:

mkdir -p NavLock-BreachSynth
for archive in data/test-scenes-*.tar.gz; do
  tar -xzf "$archive" -C NavLock-BreachSynth
done

index/test_scenes.jsonl is the scene-level index and index/test_frames.jsonl is the frame-level index used by the Hub viewer. metadata/release_manifest.json records shard membership and checksums. checksums.sha256 covers every published file except itself.

Evaluation Boundary

All samples are TEST_ONLY. They must not be moved into training or validation splits, used for detector fitting, threshold selection, or early stopping. The primary task is paired WARN versus NO_WARN evaluation. Hazard-stage recognition is secondary and uses ENTRY_PATH_NOT_CLEAR, UNAUTHORIZED_ENTRY, and INSUFFICIENT_CAPACITY only for generated hazards.

The 123 scenes are designed safety tests, not observations of real breach frequency. Delayed warning is recorded as delay; only absence of warning over the complete observed clip is a detection failure. Results do not establish deployed-system reliability or real-world event prevalence.

Source Data and License

This release is derived from NavLock-HY and follows its current controlled academic-preview policy. The dataset card therefore uses license: other. Apache-2.0 applies to the GitHub code repository, not to these sensor files. See DATA_LICENSE.md and the NavLock-HY dataset card before reuse.

Citation

@article{lu2026navlockbreachsynth,
  title   = {NavLock-BreachSynth: Rule-Constrained Multimodal Counterfactual
             Synthesis for Navigation-Lock Warning Reliability Assessment},
  author  = {Lu, Xiaodong and Li, Boju and Wu, Shishuang and Liu, Jun and Xu, Sudong},
  journal = {Reliability Engineering \& System Safety},
  year    = {2026},
  note    = {Manuscript in preparation}
}