How We Measure Interview Copilot Response Time

Evidence status: This page is excluded from search results until Aissence publishes a reproducible benchmark. A response-time claim needs a defined start and end event, test environment, sample size, model configuration, failures, and percentile results.

Define the latency path before measuring it

Use Capture → Transcribe → Route → First token → Render. Record a timestamp at every boundary. “Response time” is ambiguous unless the benchmark states whether it begins with speech, final transcript text, request dispatch, or another event.

Publish the test environment

• Test date and application version
• Device, operating system, and processor
• Network location and connection type
• Speech-to-text provider and settings
• Model, provider, region, and generation settings
• Number and type of questions
• Warm and cold runs
• Timeouts, errors, and excluded samples

Report a distribution, not one attractive number

Publish the median, p90, p95, minimum, maximum, and failure rate. Separate first-visible-token latency from the time required to complete the answer. Link a downloadable CSV so readers can reproduce the summary.

Use a claim that names its conditions

In our [date] test of [n] questions on [device and network], median first-visible-token latency was [x] ms and p90 latency was [y] ms. [z] requests failed or timed out.

Publication requirement

Remove the search exclusion only after the raw measurements, calculation method, and current product version are available on this page. Until then, Aissence should not advertise a fixed response-time benchmark through editorial markup.