Different measurements answer different questions
Address classification, delivery outcomes, and response time need separate evaluations. A synthetic score is not a promise about your recipients, and an engine timing is not an API response-time guarantee.
Read the result in context
Use each measurement only for the question it was designed to answer.
Synthetic classification
Synthetic fixtures exercise known address patterns and expected decisions. They do not establish real-world mailbox existence, user identity, or bounce rates.
Campaign outcomes
An observed delivery campaign describes its own recipients and sending conditions. Filtering can remove deliverable contacts as well as bounces; both outcomes matter.
Engine and API timing
A warmed local engine test excludes parts of a real request. DNS lookups, caches, network travel, concurrency, and service configuration affect end-to-end latency.
Current evidence boundary
Historical internal tests and campaign reports are not a guarantee for the deployed service. This page does not publish aggregate accuracy, bounce-reduction, or latency numbers without a fully scoped, reproducible public report.
A publishable report must name the dataset and collection date, engine version, sample selection, expected labels, decision thresholds, and measurement conditions. Campaign reporting must also count deliverable contacts suppressed and state when outcomes were collected.
Evaluate your own workflow
Test a representative sample and inspect both accepted and rejected records. Measure ownership or delivery separately when those are your goals.
For performance, record end-to-end requests from your integration with warm and cold caches, domain lookups, errors, and concurrency represented.
Compare workflow fit, not unrelated headline numbers
Review the checks, outputs, failure handling, and data processing each workflow requires.
Inspect an email result
Try the playground and review the returned reasons before deciding how your application should respond.