Grade the evidence

---
name: evidence-grader
description: "Audits the strength, freshness, and limits of evidence behind a product decision. Use when research, analytics, customer requests, or inferred signals are being used to justify a recommendation."
---

# Grade the evidence

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

Separate what happened from what was said, inferred, or merely hoped.

## Instructions

Audit the evidence behind this decision. Extract every material claim and attach its source, date, sample size or context, and collection method. Classify each item as observed behavior, payment or commitment, direct observation, stated request, quantitative analysis, inference, or assumption. Grade its strength using one rubric: directness, recency, representativeness, and reproducibility. Surface contradictory signals, selection bias, stale sources, and missing counter-evidence. Conclude with three lists: supported now, unsupported now, and unanswered. Recommend only what the available evidence earns; do not average weak signals into confidence.

## Detect the problem

- Several customer quotes treated as proof of demand
- A dashboard number with no date, cohort, or collection method
- Contradictory evidence omitted from the recommendation

## Hold the gate

Do not declare this phase complete until every item passes:

- [ ] Every material claim is paired with a named source, date, and sample or context
- [ ] Evidence is classified as behavior, payment, observation, request, analysis, inference, or assumption
- [ ] Strength is graded with one consistent rubric rather than persuasive wording
- [ ] Contradictions, selection bias, missing evidence, and stale evidence are surfaced
- [ ] The decision states what the evidence supports, does not support, and cannot yet answer

## Verify the result

Choose the strongest conclusion and trace it backwards to dated source evidence; if any link is an unlabeled inference, lower the grade.