Pilot a fixed extraction contract
Freeze extraction fields and validate a small sample before scaling.
These examples and illustrative results are independently authored teaching materials, not measured model results.
Use case
Extract titles and prices from 1,000 product pages with standard, sale and out-of-stock layouts. Pilot a contract before batch execution exposes inconsistent fields.
Mechanism
Define title:string, price:number-or-null, currency and extraction status. Separate missing price from fetch failure rather than using 0. Sample each layout and check readiness, response nesting and per-page status. Manually compare fields with pages, record mismatches and repilot revisions. Scaling retains per-item validation, failures, retries and cost scope.
Bad example
Extract all 1,000 pages with shifting fields. Set missing price to 0 and treat outer API success as whole-batch success.
Good example
Pilot standard, sale and out-of-stock pages with fixed fields and null/error rules. Missing price is null, fetch failure is error, and sale-price selection must be explicit. Inspect inner status and source per page before scaling in batches with item checks.
Why the change matters
A pilot exposes layout/contract mismatches while changes are cheap. Explicit missing/failure states prevent complete-looking but semantically wrong data.
Observable expectation
Teaching pages yield normal price, the declared sale-price choice and null for absent price. DNS failure must not become 0. Check title/currency/status and reconcile sample counts with accepted/failing records. Without execution, provide a pilot design only.
Limits
Sample success does not establish population correctness; retain drift monitoring and item checks. Dynamic content, currency, login and blocking affect results. Frozen batching/delay guidance needs actual-contract verification, and paid calls require authorization.