[jmacd867]

Closes #244

Summary

inference/ becomes one installed package, grouped by the concerns in GLOSSARY.md. Event classification, which the client never asked for, stays in its own folder and is marked out of scope.

 inference/
-├── angles.py, features.py, clustering.py, ...   # 12 scripts and modules, flat
-├── Data/  Models/  Results/  dataset/           # mixed case, code inside Data/
-├── requirements.txt
+├── data/  models/                               # lowercase, no code
+│   └── prepared/                                # was dataset/
+├── pilot_inference/
+│   ├── angles.py, paths.py                      # shared
+│   ├── capture/                                 # data_logger
+│   ├── performance/                             # features, clustering, synthetic pilots
+│   └── event_classification/                    # OUT OF SCOPE; test_model.py is now evaluate_model.py
 └── tests/                                       # mirrors pilot_inference/

Run scripts as uv run python -m pilot_inference.<group>.<module>, from inference/ or from the repo root with --project inference. pyproject.toml gains a build-system, so uv sync installs the package editable.

Also: GLOSSARY.md gains Segment, Instruction and Leg, and the event label terms move to an out-of-scope section. The root README.md, inference/README.md, AGENTS.md, docs/telemetry_schema.md and ADR 0002 are updated for the new paths and commands. The .pkl models stop being tracked (they were already in .gitignore), and train_model writes only bestModel.pkl, not an identical finalModel.pkl. requirements.txt and two empty files are removed.

Evidence

• Before: uv run pytest gave 73 passed. The flat scripts ran as python inference/cluster_pilots.py, and test_data_logger.py needed sys.path.insert(... / "Data") to import the logger.
  After: uv run pytest gives 73 passed (same tests, moved and re-imported). The logger test imports pilot_inference.capture.data_logger.

generate_synthetic_pilots -> compute_features -> cluster_pilots --k 3 --truth ...
before: adjusted Rand index 0.79 (documented in the old README)
after:  adjusted Rand index 0.792, run via python -m from inference/

Every event_classification script also ran end to end via python -m. The data logger needs a live relay, so it was not run. Anything needing X-Plane or iMotions is unverified.

Merge Danger

Door: two-way. It is moves and edits, and a revert restores the old layout.

Two things are not a clean undo for other people. The .pkl models are no longer tracked, so a fresh clone must run train_model before evaluate_model. raw_flight_data_updated.csv was meant to be pruned but is kept: it is the source of the committed labeled_flight_data.csv, and running label_generator as documented overwrites that file with the 597-row raw_flight_data.csv version (documented in inference/README.md).

Blast Radius: inference/

Anyone with a local inference/.venv must run uv sync again. Open branches that touch inference/ will conflict with the moves (PR #176 changes 6 files there). Open issues #212 and #229 cite the old Data/ and Results/ paths. Data/ to data/ is a case-only rename, so on a case-insensitive checkout a stale Data/ folder can linger until it is deleted by hand.

[Copilot]

Copilot was unable to review this pull request because the user who requested the review has reached their quota limit.

[Sanjana-Siddani]

Big cleanup, nice work! I followed the README from scratch: uv sync, all 73 tests pass, and every command works from inference/ and from the root. Clustering gives the same result as before (0.792), and train → evaluate works on a fresh clone without the .pkl files. One small idea for later: running the event-classification steps still overwrites the committed files in data/prepared/, so maybe stop tracking those at some point. Approving