# Post test readout including null results

> Use when a test has finished and you need an honest write-up, including when nothing moved.

## Fill in before running

- `{{TEST_NAME}}`
- `{{ORIGINAL_HYPOTHESIS}}`
- `{{VARIANT_DESCRIPTIONS}}`
- `{{DATES}}`
- `{{SAMPLE_SIZES}}`
- `{{PRIMARY_RESULTS}}`
- `{{SECONDARY_RESULTS}}`
- `{{ANOMALIES}}`

## Prompt

```
You are writing the readout for a completed experiment. Honesty about a null result is the point of this document.

TEST NAME: {{TEST_NAME}}
HYPOTHESIS AS WRITTEN BEFORE THE TEST: {{ORIGINAL_HYPOTHESIS}}
VARIANTS: {{VARIANT_DESCRIPTIONS}}
DATES AND DURATION: {{DATES}}
SAMPLE PER VARIANT: {{SAMPLE_SIZES}}
PRIMARY METRIC RESULTS: {{PRIMARY_RESULTS}}
GUARDRAIL AND SECONDARY METRICS: {{SECONDARY_RESULTS}}
ANYTHING UNUSUAL DURING THE TEST: {{ANOMALIES}}

Write the readout with these sections:
1. Result in one sentence - one of: variant won, variant lost, no detectable difference, or inconclusive due to insufficient power. Choose one and do not hedge across two.
2. The numbers, with confidence interval or significance as supplied. If I have not supplied significance, say it cannot be judged rather than calculating from incomplete inputs.
3. What this means for the hypothesis - was the underlying belief supported, contradicted, or left untested.
4. Guardrail check - anything that got worse.
5. Segments - only if I supplied segment data. If I did not, state that segment claims would be post hoc and are not included.
6. Decision - ship, revert, iterate, or investigate, with the reason.
7. What we learned even if nothing moved. A null result usually means the change was too small to matter or the problem is elsewhere. Say which you think it is and why.

Rules: never describe a non-significant difference as a "slight lift" or a "positive trend". Do not go hunting for a winning segment to rescue the result. Do not open with scene setting or close with a paragraph restating section one. No em dashes.
```

## Getting a better result

- Paste the hypothesis exactly as it was written beforehand, not a tidied version.
- If you did not pre-register segments, let it refuse to report them - that rule is doing real work.
- Section seven is what makes null results worth running; keep a running file of them.

---

From the QuQi prompt library - https://www.quqi.io/prompts/post-test-readout