Track A — Catalog Operations

This track is not coding. It's the operational work of building a trustworthy product catalog: configuring shops, running scrapes, and reviewing what comes back. It's also the long pole to launch — reviewing thousands of products takes calendar weeks no matter how fast the code work goes.

How the pipeline works today

Shops are configured in the database via the admin UI (/admin/scraper), not in code:

1. A ScraperShop holds the shop name plus its BrightData PLP and PDP collector ids.
2. Each ScraperSource under it is one category listing URL (e.g. /fpv/moteurs → MOTOR).
3. A PLP scrape creates ScrapedProduct rows in status pending.
4. You triage them in the review queue: approve → ready, reject → skip.
5. A PDP scrape enriches ready products (full description, images, specs) → pdp_done (or antibot_blocked on failure — retriable).
6. Import turns a pdp_done scraped product into a catalog Product, applying the per-category spec mappings (ScraperSpecMapping — scraped key → SpecType).
7. Imported products land in the product review queue (/admin/products) where name, brand, category and specs get validated before the product is live and searchable.

1. Foundation pass (do once, before mass review)

Everything you review inherits the category/spec structure that exists at review time. Fixing it later means re-touching reviewed products, so freeze it first (part of M0):

• Category tree final: frame, motor, FC, ESC, AIO, VTX, camera, antenna, receiver, radio, goggles, battery/charger, props, GPS, accessories. Decide now what's out of scope (e.g. spare screws, apparel).
• Spec templates per category (/admin catalog screens): motor → KV/stator/shaft; VTX → power/band; battery → cells/capacity/C-rating; frame → size/wheelbase… Only specs buyers filter on.
• Spec mappings seeded per shop × category: run a small test scrape first (below), look at the raw keys each shop emits, map them once. Unmapped keys are silently lost at import.
• Naming convention written down: Brand Model Variant (e.g. "iFlight XING2 2306 1855KV"), no shop suffixes ("- Drone-FPV-Racer"), no marketing text. Consistency here is what makes dedup and matching possible.

2. Per-shop runbook

Repeat for each shop, in rollout order (§4). Don't parallelize shops — finish one before starting the next, so mistakes stay contained.

3. Triage rules

Decide these once so every batch is judged the same way:

• Approve: single identifiable product, in a supported category, matchable to a brand.
• Skip: bundles/kits ("frame + motors combo"), shop gift cards, apparel, repair services, products with no usable image, categories you declared out of scope.
• Variants (KV options, prop colors): one catalog product per meaningfully different variant (a 1855KV and 2555KV motor are different products; prop colors are not). Follow the naming convention.
• Duplicates within a shop (same product in two source categories): approve once, skip the rest.
• Use bulk select per brand/keyword to clear obvious groups fast; hand-review only the ambiguous tail.

4. Shop rollout order

Update the Status column as you go — this table is the tracker.

5. Acceptance bar

Per shop: 0 pending scraped products; ≥95% of ready reached pdp_done; imported products all through product review; no shop-suffixed names in the first 100 search results.

Catalog launch bar (gates M2):

• Every core category (frame, motor, FC/ESC/AIO, VTX, camera, goggles, radio, battery) has enough products that a buyer creating a deal finds their gear — spot-check by building 5 realistic used-quad listings end-to-end.
• No duplicate products in the top-100 of any core category.
• Cross-shop: products carried by both shops show both prices.

Throughput planning: expect 1,000–3,000 products per shop. At a sustainable 100–150 reviews/day, a shop is roughly 2–3 weeks of daily batches — that's why Track A starts at M0 and never pauses.

6. Coding dependencies of this track

Small Track B/C tasks that this track needs — schedule them into M2:

Until cron lands: manually re-run price scrapes weekly on completed shops so availability/prices don't rot while you review the next shop.