Replace setup-python with setup-uv: the self-hosted Gitea runner lacks the
toolcache and lsb_release dependency, so the job failed before any project
step. uv installs its own Python 3.12; the checks are unchanged.
Align the README and docs with the committed artifacts: quick start serves
from baked artifacts, rebuild steps are separate, and the reference-model
comparison no longer overstates the result. Drop stale external-doc
references and comments that restated the code.
- make small trains from a CSV and prints metrics + confusion matrix
- make figures renders dependency-free SVGs from metrics.json
- scripts/e2e_train.py boots a real server and checks snapshot, robustness,
auth, one-job-at-a-time, and both training jobs with hot-swap
- CI/ruff/compileall now include scripts/
- train.py emits per-sensor MAE/WAPE/R2 vs the seasonal-naive baseline
- small.py emits the full confusion matrix and per-sensor classifier metrics
- metrics.json now carries the small-model summary for paper support
- regenerate baked artifacts; tests assert the tables are internally consistent
Adds ruff as a dev dependency with an explicit py312 E/F rule set, so the
gate stays stable across ruff releases. make check now runs uv sync --frozen,
ruff, compileall, and pytest; make lint runs ruff alone.
The Gitea job splits the same checks into named steps for readable failures
and has a 10-minute timeout. No secrets, image push, or cluster access.
GET /bootstrap returns the 8 sensors, hourly series for all of them,
context, meta, and versions in the CONTRACT.md shape. Snapshot artifacts
are projected to contract fields once at boot; incomplete artifacts return
503. Adds tests for shape, sensor count, hourly alignment, and versions.
Dockerfile now creates UID/GID 10001, uses COPY --chown=10001:10001, and
runs as USER 10001:10001. Adds make check and a minimal Gitea Actions
workflow that runs the same local gate: uv sync --frozen, compileall, pytest.
The baseline now reproduces mission/data.md exactly: MAE 5.40, WAPE 44.7%,
R2 0.462. The model's lag_week remains the previous observed same weekday/hour,
which handles sensor gaps and is documented in the model card.