# A3 — Dataset building

> Build recurring, schema-validated web datasets. Stable ids, typed extractors, and self-healing when pages change.

**URL:** https://athree.dev/use-cases/datasets/  
**Canonical markdown:** https://athree.dev/use-cases/datasets.md

## Problem

Teams need clean, refreshable datasets from the public web — listings, prices, catalogs — without brittle one-off scrapers or paying for an LLM on every refresh.

## Approach

Define schemas and missions. A3 learns extraction stages per site, dedupes by stable listing ids / canonical URLs, and replays cheaply on a schedule. When markup drifts, learning / self-heal updates stages instead of rewriting the whole pipeline.

## Outcomes

- Schema-validated rows suitable for analytics or model training
- Predictable cost after learning
- Merge-friendly refreshes via stable identifiers

## Related

- Concepts: https://athree.dev/concepts.md
- E-commerce: https://athree.dev/use-cases/e-commerce.md