# CLAUDE.md

This file provides guidance to Claude Code (claude.ai/code) when working with code in this repository.

Default to using Bun instead of Node.js.

- Use `bun <file>` instead of `node <file>` or `ts-node <file>`
- Use `bun test` instead of `jest` or `vitest`
- Use `bun install` instead of `npm install` or `yarn install`
- Use `bun run <script>` instead of `npm run <script>`
- Bun automatically loads .env, so don't use dotenv.

## APIs

- `Bun.redis` for Redis. Don't use `ioredis`.
- Use Drizzle ORM with `postgres` driver for Postgres. Connection is in `src/db/index.ts`.
- Prefer `Bun.file` over `node:fs`'s readFile/writeFile.
- `Bun.$\`cmd\`` instead of execa.

## Commands

```sh
# Run all tests
bun test

# Run a single test file
bun test src/supplier/supplier-scoring.test.ts

# Type-check (no emit)
./node_modules/.bin/tsc --noEmit

# ASIN lead-list pipeline (LLM-based)
bun start leads.xlsx --out results.xlsx

# Supplier UPC pipeline (deterministic)
bun run upc-file --input input/supplier.xlsx --out output/supplier_ranked.xlsx

# Category discovery pipelines
bun run bestsellers
bun run monthly-sold
bun run mid-range

# Stalker pipeline
bun run stalker --input input/asins.xlsx

# Web API server
bun run start:web      # http://localhost:3000

# SP-API connectivity tests
bun run src/sp-test.ts
bun run src/sp-test.ts B07SN9BHVV
bun run src/sp-test.ts --sellability B07SN9BHVV

# Database migrations (Drizzle)
bun run db:generate
bun run db:migrate
```

## Architecture

Two distinct analysis pipelines share infrastructure (Keepa, SP-API, Redis, Postgres) but diverge in how they produce verdicts.

### ASIN Lead-list Pipeline (`src/index.ts` → `src/analysis-pipeline.ts`)

For spreadsheets containing known ASINs. Verdict is LLM-based (FBA/FBM/SKIP via LM Studio).

Flow: `reader.ts` parse → Redis cache check → `integrations/sp-api.ts` sellability gate (5 concurrent workers) → `integrations/keepa.ts` batch enrichment → `integrations/sp-api.ts` pricing + FBA fees (5 concurrent workers) → `integrations/llm.ts` batched analysis (5 products/batch) → `writer.ts` XLSX + Postgres.

### Supplier UPC Pipeline (`src/supplier/upc-file-analysis.ts`)

For supplier price lists containing UPC/EAN values. Verdict is deterministic (BUY/WATCH/SKIP); never calls LM Studio.

Flow: `supplier/upc-file-reader.ts` streaming parse (`.xlsx`) or row-window parse (`.xls`) → SP-API catalog UPC lookup first, Keepa UPC lookup as fallback → `integrations/keepa.ts` demand enrichment → `integrations/sp-api.ts` sellability + FBA fees → `supplier/supplier-scoring.ts` deterministic score → `supplier/supplier-export.ts` Excel workbook (`Ranked Leads`, `Skipped`, `Summary` sheets) + Postgres.

UPC resolution priority: SP-API catalog lookup → Keepa fallback (for no-match or request failure only).

### Category Pipelines

`src/categories/` — Keepa category browsing → SP-API sellability gate → LLM verdict. Each saves results to Postgres. Mid-range applies configurable filters (monthly sold, price, seller count, Amazon buy box share).

### Stalker Pipeline (`src/stalker/stalker.ts`)

Tracks competitor sellers across ASINs. Fetches storefronts, checks sellability of inventory items, and persists matched seller data to Postgres.

### Shared Infrastructure

| Module | Role |
|--------|------|
| `src/types.ts` | All shared interfaces (`ProductRecord`, `KeepaData`, `SpApiData`, `SupplierScore`, etc.) |
| `src/config.ts` | Env var loading via `Bun.env` |
| `src/db/index.ts` | Drizzle Postgres connection (shared pool) |
| `src/db/schema.ts` | Drizzle schema for all tables |
| `src/db/persistence.ts` | Product, observation, unified run-item, UPC resolution, and revision persistence |
| `src/integrations/keepa.ts` | Keepa API: batch ASIN fetch, UPC lookup, auto rate-limiting |
| `src/integrations/sp-api.ts` | SP-API: sellability, pricing+fees, UPC catalog lookup |
| `src/integrations/cache.ts` | Redis caching (24h TTL for lead-list; 12h for mid-range) |
| `src/integrations/llm.ts` | LLM integration (LM Studio / Claude) |
| `src/server.ts` | Bun HTTP server exposing REST endpoints for both pipelines |

### File Layout

- `src/integrations/` — external API clients (Keepa, SP-API, Redis cache, LLM, SearXNG)
- `src/categories/` — category discovery pipelines
- `src/stalker/` — competitor seller tracking pipeline
- `src/supplier/` — supplier UPC analysis pipeline
- `src/db/` — Drizzle schema and connection
- `input/` — source spreadsheets (git-ignored)
- `output/` — generated workbooks (git-ignored)

## Project Rules

- Keep the ASIN lead-list and category flows compatible with their current LLM-based FBA/FBM/SKIP analysis.
- The supplier UPC pipeline must not call LM Studio.
- Supplier UPC files resolve UPC/EAN through SP-API catalog lookup first; Keepa UPC lookup is fallback only (no-match or request-failure cases).
- Supplier workbook output must keep `Ranked Leads`, `Skipped`, and `Summary` sheets.
- Treat `products.asin` as the canonical normalized product identity; UPC values belong only in identifier and resolution records.
- Store time-varying data in observations or revisions and retain run history rather than overwriting prior analysis.
- When changing UPC supplier behavior, cover SP-API UPC parsing, deterministic scoring, and workbook export with `bun test`.