Cost per usable record is the metric that keeps monitoring honest: it measures how much you spend to get one record that passes your quality gate (correct region signal, required fields present, and stable output). When this cost rises, it usually means your queue lost a constraint: pacing drifted, retries amplified, or region rules stopped being enforced.
Define the concept clearly
A “usable record” is not a page load. It is a sample that meets your monitoring definition. If your report depends on a price block and a location signal, a record without those fields is not usable even if the page returned HTTP 200.
Results it can change
Teams that optimize for success rate often scale the wrong thing. Cost per usable record pushes you to scale comparability: stable region signals, stable field completeness, and predictable failure cost.
What happens in the request path
When failures retry immediately, the queue pacing changes. That shifts routing and behavior and often reduces field completeness. Your costs rise because you spend more requests per usable record, even if headline success stays similar.
Workloads where it does not fit
For one-off investigative scraping where you only need a handful of pages, the metric matters less. For repeatable monitoring and comparisons, it is one of the fastest indicators of drift.
FAQ
Isn’t success rate enough?
No. Success rate ignores missing fields and unstable region signals. Monitoring decisions need usable records, not just responses.
What should I track alongside it?
Region sentinel consistency, field completeness, and retry budget usage. They explain why the cost changes.
How do I lower the cost?
Stabilize inputs: lock region rules per market queue, enforce pacing, and cap retries so failures cannot amplify.