Read experiments only when they are ready to be read. Show where people fall out, and whether the instrumentation can say why. Report engagement depth rather than frequency, because frequency rewards friction.
The Role Pack contains the method. The canonical profile contains the governed capability boundary. They reference each other without duplicating ServAI Core.
Read an experiment against its pre-registration and refuse before the registered sample is reached.
Report guardrail metrics whether or not they moved.
Compute required sample size only from parameters a person supplied.
Report retention as a curve against a versioned activity definition.
Report funnel conversion with time-to-stage, and flag any instrumentation change in the window.
Separate first use from repeat use when reporting feature adoption.
Report engagement depth alongside frequency and identify failed attempts.
Report churn indicators as association and never as cause.
Every method maps to a registered capability the profile can actually prepare or recommend. Repeating a precondition sentence is not evidence.
product.experiment.readoutCompare the pre-registered primary metric between arms, report lift with an interval, and check the pre-registered sample size was reached. Guardrail metrics are reported whether or not they moved.
product.experiment.sample_sizeRequired n from baseline rate, minimum detectable effect, power and significance level — all four stated by a person before the run.
product.retention.cohort_curveRetention by cohort aligned on periods since first use, with the activity definition pinned. Reports the curve, not a single rate.
product.funnel.stage_conversionConversion between consecutive stages, with time-to-next-stage, over a window with no instrumentation change.
product.adoption.feature_uptakeUptake by cohort against the release date, separating first use from repeat use — the second is the one that matters.
product.instrumentation.gap_reportCompares questions asked against events available and reports which questions the current instrumentation cannot answer.
product.engagement.depth_not_frequencyDepth measures — actions per session, features touched, task completion — alongside frequency, because frequency alone rewards friction.
product.churn.leading_indicatorsBehaviour differences between retained and churned cohorts in the period before churn, reported as association and never as cause.
The Role Pack cannot grant itself authority. The shared resolver binds only what the profile, organization, tools, data classification and human policy permit.
Choose from 1,000 named connector routes. Every connection remains tenant-specific, least-privilege, admitted and health-gated; write authority is never implied by catalogue presence.
ServAI proposes the exact role manifest. Your people review the profile, data scope, tools and approval path before anything becomes active.
This is a separately entitled cross-industry Role Pack. It can join an Industry Pack mission only when the organization owns both entitlements.