Enterprise scale and pilot measurement

Read the enterprise example correctly#

The enterprise-scale website view is an illustrative scenario. Its agent counts, outcomes, policy families and sample events explain how a portfolio view could work. They are not production telemetry, customer results or evidence that LSAS has operated at those volumes.

Modeled policy-assignment coverage describes an example assignment denominator. It does not establish observed assets, evaluated workloads, verified enforcement or regulatory compliance. Representative fictional sample events are not canonical execution receipts.

Define a representative qualification workload#

Specify applications, request shapes, content sizes, policies, adapters, concurrency, rate and evidence retention. Include both ordinary traffic and denied/held/error paths. Measure the actual deployed topology rather than extrapolating from a small contract fixture.

Latency and throughput need defined start/end points, periods and percentiles under documented conditions. Separate provider latency from governance overhead where instrumentation supports it. Event timestamps alone do not establish a reliable end-to-end duration.

Test failure behavior as part of capacity#

Qualification should verify that tenant isolation, current-authority checks and evidence publication remain correct under contention and interruption. Include cancellation, expiry, unavailable dependencies, recovery and retained evidence after restart. A fast response that bypasses a required control is not a successful benchmark.

Agree pilot acceptance before measurement#

Use the customer's baseline for evidence retrieval, manual review effort, traceability, workflow coverage and exception handling. Financial visibility is included only where a qualified integration supplies the required usage and rate provenance. Until a baseline exists, the result is measured during the pilot.

• Define the population and denominator.
• Name the collection method, frequency and owner.
• Record missing, unsuccessful and censored observations.
• Approve acceptance thresholds and limitations before comparing results.