Construct a matched product experiment and inspect its learning
A hands-on PSM-I lab. You produce the real artefact and 10 automated checks verify it behaves the way the exam expects.
Try this labAll PSM-I practice
- Certification
- PSM-I
- Format
- Structured configuration
- Difficulty
- medium
- Estimated time
- 35 min
- Automated checks
- 10
The brief
Repair product-experiment.json. protocol:{hypothesis,sourceRef,treatment,cohorts:[IDs],primary,guardrail,minimumPerArm,minimumLift,maximumErrorIncrease} must tie the treatment to F0/F1, include novice, retain the visible measures and minimum policy bounds. Stricter thresholds are allowed. assignments:[{unit,arm}] uses control or the chosen treatment. Include each selected cohort's full matched pairs exactly once with one control/one treatment per pair, sufficient arm samples and actual case/cost limits. plan:{recruitAt,exposeAt,observeAt,inspectAt} must respect recruitment, day3 availability, seven exposure days and inspection by day18. Unsupported windows contribute no observations. baselines:[{cohort,ref,units,completed,rate}] preserves each selected cohort's historical facts. metrics:[{cohort,controlUnits,treatmentUnits,controlCompleted,treatmentCompleted,controlErrors,treatmentErrors,controlRate,treatmentRate,completionLift,controlErrorRate,treatmentErrorRate,errorIncrease}]. Derive observed counts from assigned replay cases only with a supported window. Rates divide actual arm counts; lift/errorIncrease subtract control from treatment. Zero denominators yield null. decisions:[{cohort,recommendation,nextAction,nextAt,owner,hours}] must respond separately to actual sample, lift and guardrail facts. A failed hypothesis or guardrail requires useful learning, not manufactured expansion. Schedule exact action roles/hours after inspection within combined daily capacity and day22. Passing evidence may still support cautious repeat testing. report contains the same pooled count/rate fields without cohort, plus treatmentCost,followUpFinish,learningReady. Pool numerators and denominators; report latest nextAt only if all next steps fit, otherwise null. learningReady requires a supported protocol, matched allocation, budget, window and responsive feasible next steps; it does not mean the hypothesis succeeded. Preserve partial progress and repair without Reset.
What the checks verify
Your work is graded on 10 independent properties, not on matching one reference answer.
- Protocol, assignment, baseline, metric and next-step records are finite bounded and unique.
- The sourced hypothesis, audience and measures retain the stated decision policy.
- Actual selected cases create complete balanced within-cohort comparisons.
- Actual case count and treatment demand fit the supplied experiment constraints.
- Recruitment, availability, full exposure and timely inspection support observed model data.
- Historical cohort populations and rates retain immutable source evidence.
- Per-cohort counts and rate differences reflect actual assignments and supported observation.
- Each cohort's recommendation responds to its own observed hypothesis and guardrail result.
- Conditional next steps preserve real roles, effort and post-inspection daily capacity.
- Pooled evidence, treatment cost and conditional next-step finish are honestly reconciled.
Where this sits in the PSM-I blueprint
- Domain
- Managing Products with Agility
- Objective
- Product Value
- Skill
- Outcomes and Value
Part of PSM-I preparation
Labs are written by ExamNova to teach the decisions the exam tests. They are not reproductions of vendor lab content.