# Experiment handoff

Develop three page-specific experiments from observed public product and pricing pages. Effects on purchases and activation are unmeasured.

1. Recheck observations against the current page and available first-party data.
2. Select one hypothesis. Agree on assignment, exposure, success window, session attribution, deduplication, guardrails and a stopping decision before changing copy. Complete the decisions in tracking-plan.json.
3. Implement one change on the user's chosen stack; verify routes, keyboard behavior and tracking in the authorized environment.
4. Run a preview and inspect desktop/mobile behavior before release.
5. Activate only with the user's existing authorization. Avoid simultaneous changes that confound the selected metric.
6. Analyze the actual collected cohort with an appropriate method; this tool does not call a winner.
7. Roll back if the defined guardrail or validity checks fail.

- Every proposed change is an untested hypothesis, never a measured uplift
- Counts are supplied nested-cohort observations; the generator does not fetch analytics
- Rates are descriptive and do not establish statistical significance or causality
- Priority and effort are supplied judgments, not generated estimates of profit
- Event contracts are implementation drafts; no tracking, identity collection or experiment is activated