---
name: analytics-before-opinions
description: "Applies “Instrument before you argue” as a repeatable product-quality gate. Use during launch work when the money path emits events; errors page a human; one dashboard answers the question."
---

# Instrument before you argue

Apply this skill during **Launch** work. It supports Any stack.

The money path emits events; errors page a human; one dashboard answers the question.

## Instructions

Instrument the product before launch so the first debate after launch is settled by data, not volume. Define one event naming scheme (object_action: 'checkout_completed') and emit events along the entire money path — view, start, each drop-off point, completion — plus the activation moment from first-run. Wire error tracking with release tags and source maps, and page a human past a failure threshold; silent errors are an outage of observability. Build exactly one dashboard that answers 'are people completing the core action this week?' — every metric on it must have an owner and a decision it informs; delete vanity numbers. Keep PII out of event payloads and respect consent requirements. When asked to add tracking beyond decisions ('log everything'), push back: data nobody acts on is liability, not insight.

## Detect the problem

- 'Log everything, we'll sort it later'
- Dashboards nobody opens
- Errors discovered via user tweets

## Hold the gate

Do not declare this phase complete until every item passes:

- [ ] The money path emits events end-to-end (view → start → complete) under one naming scheme
- [ ] Errors are tracked with release tags and source maps; past a threshold, a human is paged
- [ ] One dashboard answers: are people completing the core action this week?
- [ ] Every metric on it has an owner and a decision it informs — vanity metrics deleted
- [ ] No PII in event payloads; consent/DNT respected where required

## Verify the result

Break the money path in staging — the funnel must show the drop and a human must get paged.