# AGENTS.md

## Architecture

- Python 3.11 package under `src/idea170_agrimesh_federated_privacy`.
- SQLite-backed governance ledger for farms, adapters, contracts, deltas, and sync state.
- Pydantic models validate local problems, shared signals, plan deltas, and adapter contracts.
- `solver.py` contains the ADMM-lite allocator for cross-farm irrigation planning.
- `canonical.py` maps AgriMesh primitives into a CatOpt-style intermediate representation.
- `demo.py` assembles a drought scenario for 2 to 3 farms.

## Tech Stack

- Standard library: `sqlite3`, `json`, `hashlib`, `hmac`, `math`, `random`, `dataclasses`, `pathlib`.
- External dependency: `pydantic` for strict data validation.
- Tests: `pytest`.

## Rules

- Keep changes minimal and deterministic.
- Preserve SQLite schema compatibility unless a migration is explicitly added.
- Any new public data model should have validation and at least one test.
- Preserve canonical JSON ordering when hashing or signing payloads.
- Do not bypass the solver or ledger abstractions with ad hoc state.

## Testing

- `bash test.sh`
- `pytest -q`
- `python3 -m build`