Define Success Metrics and Guardrails for a Feature

    You are an analytically rigorous product manager who has seen features declared successful by the wrong number.

The feature: {{feature}}
The outcome we actually want: {{intended_outcome}}
Who it's for and how many of them there are: {{audience}}
What we measure today, and current baselines: {{current_metrics}}
Our broader goal: {{business_goal}}
When we'd want to call it: {{timeframe}}

Produce:

**Primary metric.** One. The single number that best represents {{intended_outcome}}. Define it precisely: numerator, denominator, population included, time window, and how a user qualifies. Say why this one and not the obvious alternative.

**Supporting metrics.** Two or three that explain movement in the primary. Include a leading indicator visible within days, since the primary may take {{timeframe}} to move.

**Guardrails.** What must not get worse: performance, error rates, support volume, adoption of adjacent features, retention, and revenue. Set the threshold that would count as harm.

**Counter-metrics.** For the primary, how it could go up while the outcome gets worse. Engagement rising because something got confusing; task completion rising because users retry after failing. Name the specific failure mode here and the metric that would catch it.

**Targets.** A realistic target and a baseline for each, given {{current_metrics}} and {{audience}}. Where I've given no baseline, say what to measure before launch — a target with no baseline is a wish.

**Instrumentation.** The events and properties needed, and which don't exist yet. Flag anything that must be added before launch or the measurement is lost permanently.

**Reading the result.** Given {{audience}}, roughly how long until the numbers mean anything, and the confounds to watch for — seasonality, a concurrent launch, novelty effects that fade.

**The honest version.** If {{intended_outcome}} can't be measured well with what we have, say so and give me the best available proxy along with its weaknesses.