# scrape-giant

Small grocery-history pipeline for Giant receipts.

The project currently does four things:

1. scrape Giant in-store order history from an active Firefox session
2. enrich raw line items into a deterministic `items_enriched.csv`
3. aggregate retailer-facing observed products and build a manual review queue
4. create a first-pass canonical product layer plus conservative auto-links

The work so far is Giant-specific on the ingest side and intentionally simple on
the shared product-model side.

## Current flow

Run the commands from the repo root with the project venv active, or call them
directly through `./venv/bin/python`.

```bash
./venv/bin/python scraper.py
./venv/bin/python enrich_giant.py
./venv/bin/python build_observed_products.py
./venv/bin/python build_review_queue.py
./venv/bin/python build_canonical_layer.py
```

## Inputs

- Firefox cookies for `giantfood.com`
- `GIANT_USER_ID` and `GIANT_LOYALTY_NUMBER` in `.env`, shell env, or prompts
- Giant raw order payloads in `giant_output/raw/`

## Outputs

Current generated files live under `giant_output/`:

- `orders.csv`: flattened visit/order rows from the Giant history API
- `items.csv`: flattened raw line items from fetched order detail payloads
- `items_enriched.csv`: deterministic parsed/enriched line items
- `products_observed.csv`: retailer-facing observed product groups
- `review_queue.csv`: products needing manual review
- `products_canonical.csv`: shared canonical product rows
- `product_links.csv`: observed-to-canonical links

Raw json remains the source of truth:

- `giant_output/raw/history.json`
- `giant_output/raw/<order_id>.json`

## Scripts

- `scraper.py`: fetches Giant history/detail payloads and updates `orders.csv` and `items.csv`
- `enrich_giant.py`: reads raw Giant order json and writes `items_enriched.csv`
- `build_observed_products.py`: groups enriched rows into `products_observed.csv`
- `build_review_queue.py`: generates `review_queue.csv` and preserves review status on reruns
- `build_canonical_layer.py`: builds `products_canonical.csv` and `product_links.csv`

## Notes on the current model

- Observed products are retailer-specific: Giant, Costco.
- Canonical products are the first cross-retailer layer.
- Auto-linking is conservative:
  exact UPC first, then exact normalized name plus exact size/unit context, then
  exact normalized name when there is no size context to conflict.
- Fee rows are excluded from auto-linking.
- Unknown values are left blank instead of guessed.

## Verification

Run the test suite with:

```bash
./venv/bin/python -m unittest discover -s tests
```

Useful one-off rebuilds:

```bash
./venv/bin/python enrich_giant.py
./venv/bin/python build_observed_products.py
./venv/bin/python build_review_queue.py
./venv/bin/python build_canonical_layer.py
```

## Project docs

- `pm/tasks.org`: task log and evidence
- `pm/data-model.org`: file layout and schema decisions

## Status

Completed through `t1.7`:

- Giant receipt fetch CLI
- data model and file layout
- Giant parser/enricher
- observed products
- review queue
- canonical layer scaffold
- conservative auto-link rules

Next planned task is `t1.8`: add a Costco raw ingest path.