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

# How CiCi works

> Follow an evaluation from an intended behavior to an evidence-qualified decision.

CiCi separates three questions that are easy to blur together:

1. **Did the evaluation run?**
2. **Is the evidence sufficient?**
3. **What does the evidence support saying or doing?**

Finishing the runs does not automatically make the result validated. CiCi records execution, evidence sufficiency, and claim status separately.

## Evaluation lifecycle

<Steps>
  <Step title="Define the evaluation">
    Describe the intended behavior, choose the signals that represent it, and define the variants and scenarios to compare.
  </Step>

  <Step title="Run modeled evaluation">
    APMs evaluate the frozen scenarios and variants. These results are modeled evidence, not observations of real people.
  </Step>

  <Step title="Add human validation when required">
    Import participant responses, complete independent review and adjudication, and close collection when fieldwork is finished.
  </Step>

  <Step title="Evaluate sufficiency">
    CiCi checks coverage, review completion, precision, and other frozen evidence rules. A completed study can still conclude that evidence is insufficient.
  </Step>

  <Step title="Turn evidence into action">
    Review the results, generate a refinement brief and implementation packet, and create a versioned governance report.
  </Step>
</Steps>

## What changes as evidence develops

| Evidence state           | What you can do                                                                  | What you should not claim                                  |
| ------------------------ | -------------------------------------------------------------------------------- | ---------------------------------------------------------- |
| APM only                 | Inspect modeled patterns and prepare implementation-ready candidates for testing | That people actually behaved this way                      |
| Human validation pending | Monitor collection and review progress                                           | Preview a moving or partially adjudicated result           |
| Human validated          | Interpret observed evidence and the human-versus-APM comparison                  | More than the completed gates and study boundaries support |
| Insufficient evidence    | Use descriptive findings and specific remedies for the next study                | A magnitude or transfer claim that CiCi withheld           |

<Warning>
  A result marked **insufficient evidence** is a finding, not a failed run or a pending state. It means the study completed but did not support one or more intended claims.
</Warning>