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

# Corpus Qualification

> The decision loop between enterprise data and an intended AI use.

Corpus qualification helps a team decide what content may support one declared
AI use, under what conditions, and on what evidence.

## The qualification loop

<Steps>
  <Step title="Declare the use">
    Define the intended AI system, audience, purpose, decision authority, and
    criteria before evaluating content.
  </Step>

  <Step title="Freeze the corpus">
    Establish the exact population under consideration so included, excluded,
    unsupported, failed, and unresolved items can all be accounted for.
  </Step>

  <Step title="Produce evidence">
    Generate attributable utility, risk, and sufficiency evidence without
    collapsing them into an intrinsic value or safety score.
  </Step>

  <Step title="Make and preserve the decision">
    The customer applies its policy and authority. The resulting record keeps
    the decision, evidence, unresolved states, and basis together.
  </Step>

  <Step title="Hand off the approved population">
    Project the exact approved items and conditions to the named downstream
    system, then requalify after a material change.
  </Step>
</Steps>

## Two different artifacts

* A **Corpus Qualification Record** explains the complete decision and its
  basis.
* A **Qualified Corpus Manifest** carries the exact customer-approved population
  and conditions to a named downstream consumer.

Neither artifact grants authority by itself. The customer remains responsible
for the decision and downstream use.

## What qualification does not claim

Qualification does not establish ownership, rights, factual origin, intrinsic
value, universal safety, or overall fitness for AI. It answers a narrower and
more useful question: what the available evidence supports for one declared use,
including where that evidence is incomplete.