What problem does it solve?
Unregulated, un-auditable experiments lead to false conclusions, wasted engineering time, and unactionable results that do not drive concrete product or architectural decisions. This skill enforces strict, falsifiability-first guardrails to ensure every experiment produces reliable, comparable, and decision-ready evidence.
Core Features & Use Cases
- Two-commit rule enforcement: Guarantees predictions are committed before any data is collected, eliminating post-hoc hypothesis editing and retrofitted verdicts.
- FOCUS intake gating: Filters out low-value probes at the idea stage to avoid wasting resources on questions that will not change how the team works.
- Standardized audit and scoring: Provides binary audit checklists, hypothesis scorecard templates, and retrospective workflows to make experiment results comparable across time and teams.
Use case: A team testing if a new database index reduces API p95 latency can use this skill to scaffold the experiment, commit the hypothesis first, run the frozen eval pack, score predictions against evidence, and capture the verdict in a long-lived retrospective that informs future architectural decisions.
Quick Start
Use the running-experiments skill to scaffold a new hypothesis-first experiment to test if adding a database index reduces API p95 latency below 200ms, run the eval pack, and write the final verdict with a hypothesis scorecard.
