# CLAUDE.md

Guidance for AI coding agents working in this repository.

## What this is

Flood monitoring and forecasting for the Ping River, Chiang Mai. Public dashboard and
API at https://water.buildfor.life/ (never publish the server's private/mesh IP).
Production: one systemd unit, `ping-collector.service` (the Rust binary in `rust/`),
on a small VPS in `/opt/thailand-water-monitor` (a git checkout), user `water-monitor`;
plus the monthly `water-monitor-retrain.timer` (Python, `.venv`, uv-managed). Updated by
`git pull`, `cargo build --release`, `sudo bash scripts/install.sh`.

## Rules

- Python 3.11 only. `uv sync --python 3.11`; run everything as `uv run ...`.
- `make format` (black 88 / isort black profile, config in pyproject.toml; cargo fmt)
  before committing; CI fails on formatting. `make test` must stay green. Tests use
  synthetic data, never the network; database tests (both languages) run against a
  throwaway PostgreSQL schema from `TEST_DATABASE_URL` and skip without it.
- PostgreSQL only. No other database backends.
- Timestamps everywhere are Asia/Bangkok wall-clock with no offset. The dashboard
  parses them with `parseTs()` and renders with `timeZone: TZ`; keep it that way.
- Model changes go through the rolling-origin harness (`scripts/evaluate_variants.py`)
  and are judged on first-alert LEAD and false alarms, not MAE. Record results, positive
  or negative, in `docs/FLOOD_FORECASTING.md` section 5. Do not change what is deployed
  (`rise_rain` / hgb-v3) without a harness result that beats it on lead.
- `train_all()` must never silently produce a gauge-only (v2) model; the guard that
  raises `RainUnavailableError` stays.
- No `git add -A`: zero-byte shell-accident files (`#`, `$(wc`, ...) have been committed
  before. Stage files by name.
- Do not add Co-Authored-By trailers.
- The dashboard is a single file, `src/static/dashboard.html` (design and rules in
  docs/DASHBOARD_REDESIGN.md), EN + TH via the `t()` table: every user-visible string
  needs both languages, Thai is the default, no emoji. Every path it fetches must be a
  route in `rust/collector/src/web/mod.rs` (tests/test_dashboard.py checks). A replay or
  a simulation must never be shown as live data. Preview it against production with
  `uv run python scripts/dev_proxy.py`; the server reads it at start (restart).
- Rust (`rust/`, see rust/README.md): `cargo fmt --all`, `cargo clippy --all-targets
  -- -D warnings`, `cargo test`; minimum version 1.88
  (`CARGO_TARGET_DIR=target/msrv cargo +1.88 check --all-targets --locked`). Read
  `.env` only through `rust/collector/src/envfile.rs` (python-dotenv semantics, never
  dotenvy: the Python tools read the same file).
- The forecast is served by Rust (`rust/forecast`) from models trained in Python;
  `src/ml/predict.py` is the reference it is tested against. A change to
  `src/ml/features.py`, `predict.py`, the HII fill or the dam frame must be ported in
  the same change: regenerate `rust/forecast/testdata/gen_forecast_fixtures.py` and keep
  `cargo test -p ping-forecast` green. Changing thresholds, stages, leads or the
  notifier settings (`src/ml/export.py`): `uv run python -m src.ml.export --config`
  (a test fails otherwise) and rebuild the collector.
- Notifier texts are pinned by `rust/collector/testdata/notify_python_cases.json.gz`;
  a deliberate wording change updates that fixture in the same commit.
- API routes live in `rust/collector/src/web`. A route's parameters are declared once
  (`web/params.rs`); that list validates and documents (/docs). Keep response formats
  stable: external clients use them.
- The Python backfill stores (`src/hii_collector.py`, `src/rid_reservoir.py`,
  `src/ml/rain.py`) write the same tables as `rust/collector/src/db.rs`: a table or
  upsert change goes into both.

## Where things are

- `rust/collector` (`ping-collector`): collection (`rid.rs`, `hii.rs`, `dams.rs`,
  `rain.rs`), forecast run (`forecasting.rs`), ntfy/Matrix (`notify.rs`), website/API
  (`web/`), tables (`db.rs`). `rust/forecast`: inference.
- `src/ml/`: features/train/evaluate/predict/export, `rain.py` (Open-Meteo), `dam.py`,
  `hii_rain.py`. `src/hii_backfill.py`, `src/hii_collector.py`, `src/rid_reservoir.py`:
  archive backfills (`scripts/backfill_*.py`).
- `scripts/install.sh`: install/update on the server. `scripts/retrain.sh` +
  `water-monitor-retrain.timer`: monthly retrain with staged promote.
  `scripts/forecast_parity.py`: Rust vs Python forecast on a live database.
  `scripts/dev_proxy.py`: serve the working-copy dashboard against the live API.
- `docs/FLOOD_FORECASTING.md` is the authoritative model write-up; `docs/DATA_SOURCES.md`
  the source catalog; `docs/NOTIFICATIONS.md` the alerts; `docs/RUST_REWRITE.md` the
  history of the Python-to-Rust move.