# Database Governance

The repository intentionally keeps SQLite for the current private-team scale.
This directory separates four different responsibilities:

| Layer | Database | Grain | Owner |
|---|---|---|---|
| L1 | `L1/L1.db` | One raw network capture per match | Import pipeline |
| L2 | `L2/L2.db` | Normalized match, player, round and event facts | L2 Builder |
| L3 | `L3/L3.db` | Roster features and profile marts | L3 Builder |
| Web | `Web/Web_App.sqlite` | Lineups, comments, jobs and editorial data | Flask app |

## Rules

1. Paths are defined only in `database/paths.py`.
2. Schemas live next to their owning database.
3. Builders may read the previous layer and write only their own layer.
4. User-generated Web data is never restored as part of an ETL rollback.
5. A full import must run through `database/pipeline.py`.
6. Pipeline runs are serialized by `database/.pipeline.lock`.
7. L1/L2/L3 are backed up before a full pipeline run.
8. Missing metrics are stored as `NULL`, not fabricated zero values.
9. `Admin -> Data Integrity` is the operational source of truth.
10. Web schema changes increment `Config.WEB_SCHEMA_VERSION`.

## Entry Points

```bash
make l1          # Import output_arena JSON into L1
make l2          # Rebuild normalized facts
make l3          # Rebuild active-roster features
make pipeline    # Run L1 -> L2 -> L3 with backup and validation
make check       # Compile and run tests
```

## Directory Policy

- `L1/`, `L2/`, `L3/`, `Web/`: active code, schema and database.
- `backups/`: generated rollback snapshots; ignored by Git.
- `schema_bkp/`: historical schema research only; not used at runtime.
- `L1B/`: reserved demo-parser integration; not used at runtime.
- `L3/Roadmap/`: historical design notes; not used at runtime.

Large-scale directory moves are deliberately deferred until the legacy
builders no longer depend on their current module layout.