Data health.

How the pipeline protects the numbers

• Only fetched numbers. Adapters throw on anything unexpected; nothing is hand-typed.
• Sanity gate. A fetch whose plan count halves or whose median price moves more than 2.5× is rejected and the previous snapshot kept, so a half-broken scraper cannot publish garbage.
• Alerts. A provider stale for 3+ days opens a rolling GitHub issue that closes itself when the fetch recovers.
• Price history. Every change is recorded; offer pages show the latest movement.
• Catalogs refresh too. Model facts and hardware prices are re-fetched weekly with the same gate-and-alert pattern, so the model and rent-vs-buy pages do not drift.

workflow runs on GitHub ↗ · provider pages →