Contributing

Review is the point of this project. Skeptical reviews of the physics, the statistics and the code are especially welcome.

Start with the project overview and plain-language methodology. Developer setup has installation and build commands. Public guides should define IMU on first use, explain the experiment without assuming specialist knowledge, describe the current evidence honestly, and link to readable pages rather than research data files. Keep session history and personal authorization references out of public explanations.

• Current priorities: NASA solver/calibration/adequacy work remains primary. The secondary stationary experiment needs physical slow-turn controls, temperature/alignment characterization and honest uncertainty before geographic pooling. Do not qualify consumer IMUs by agreement with the expected latitude curve.
• Physics and statistics (historical aircraft simulations): the technical methodology, equations and technical validation record preserve the assumptions and unresolved issues. If you think the analysis favours one model, show it with a synthetic session (lll synth --truth ...) that the pipeline gets confidently wrong. Synthetic truth must come from the geometric generator (analysis/tests/truthgen.py), never from lll.models.
• Evidence: historical tables come from analysis/tests/evidence.py; campaign results have separate frozen records. Keep the readable evidence summary current without replacing historical results. Regenerating simulation tables requires an appropriate compute budget; retain generated detail in docs/EVIDENCE_TECHNICAL.md, with its source/configuration limits explicit.
• Hardware: real captures from the IMUs (lll bench) are welcome as test fixtures under analysis/tests/fixtures/. They pin the decoder to real bytes.
• Data format: docs/FORMAT.md is the contract between app, server and analysis. Breaking changes need a schema_version bump. Backward-compatible optional metadata and event kinds must be documented; old archives remain immutable and missing metadata stays unknown. External provider tracks remain separate artifacts with their own provenance and terms.
• Shared vectors: docs/test_vectors.json holds the model math and hand-built WitMotion packets, and both analysis/tests and the Android unit tests check against it. If you change either, update the vectors and both implementations.
• Tests: pytest analysis/tests server/tests and cd android && ./gradlew testDebugUnitTest.
• Licensing: contributions are accepted under AGPL-3.0-or-later. New source files carry an SPDX header.