> ## Documentation Index
> Fetch the complete documentation index at: https://docs.snowdoughnut.com/llms.txt
> Use this file to discover all available pages before exploring further.

# Run an experiment

> How a test gets briefed, given an ID, launched, monitored, reviewed, and recorded as a learning, so the result is findable after the people who ran it have moved on.

Use this whenever you're changing something to find out whether it works, rather than because you already know it does.

## Before you start

| You need                          | Where it lives                                                              |
| :-------------------------------- | :-------------------------------------------------------------------------- |
| An ID pattern                     | [Experiment ID](/experiment-id)                                             |
| A metric definition you trust     | [Marketing reporting data dictionary](/marketing-reporting-data-dictionary) |
| Enough traffic to detect a change | [Statistical significance](/statistical-significance)                       |

## The path

<Steps>
  <Step title="1. Read the concepts once">
    [Experiment IDs](/experiment-ids) explains why every test needs a traceable identifier. [Statistical significance](/statistical-significance) explains what a result can and can't tell you, including why checking early changes what the number means.
  </Step>

  <Step title="2. Fill in the experiment brief">
    Copy the [experiment brief](/experiment-brief) and fill in the hypothesis, the primary metric, the guardrail metrics, the sample needed, and the decision rule you'll follow either way.

    Write the decision rule before launch. Deciding what counts as a win after seeing the data is how a flat test becomes a success story.
  </Step>

  <Step title="3. Register the ID">
    Issue an ID from your [experiment ID](/experiment-id) pattern and record it before launch.
  </Step>

  <Step title="4. Apply the ID everywhere">
    The ID goes on every surface the test touches: campaign and ad names ([conventions](/ad-set-group-naming-conventions)), [UTM values](/utm-standards), and the analytics events. This is what makes results traceable later.
  </Step>

  <Step title="5. Launch and leave it alone">
    Run to the sample size you calculated. Monitor the guardrail metrics for damage, not the primary metric for good news. See [creating an experiment](/creating-an-experiment) for the full process.
  </Step>

  <Step title="6. Review against the decision rule">
    Run the [experiment review process](/experiment-review-process). Compare the result to the rule you wrote in step 2, not to what you hoped for.
  </Step>

  <Step title="7. Record the learning">
    File the outcome in the [experiment learning library](/experiment-learning-home), including the tests that showed nothing. A null result stops someone re-running the same test next quarter.
  </Step>

  <Step title="8. Roll it out or roll it back">
    A win becomes the new default and gets written into the relevant standard. A loss gets reverted and recorded. Either way the brief in the library carries the outcome.
  </Step>
</Steps>

## The mistakes worth naming

| Mistake                                                         | What it does                                    |
| :-------------------------------------------------------------- | :---------------------------------------------- |
| Deciding the win condition after seeing data                    | Any flat test can be read as a win              |
| Checking significance repeatedly ([peeking](/glossary#peeking)) | Inflates false positives                        |
| No guardrail metric                                             | A lift in one number hides damage in another    |
| No ID on the surfaces                                           | The result can't be traced back to what changed |
| Not recording null results                                      | The same test gets run again                    |

## Related resources

* [**Experiment brief**](/experiment-brief) The workbook.
* [**Creating an experiment**](/creating-an-experiment) The detailed process.
* [**Experiment review process**](/experiment-review-process) How results get judged.
* [**Statistical significance**](/statistical-significance) What the numbers mean.
* [**Experiment learning library**](/experiment-learning-home) Where outcomes are recorded.
* [**Experimentation best practices**](/experimentation-best-practices) How to run tests whose results survive scrutiny.