What problem or claim is examined
Generated documentation can widen claims, miss edge cases, or become stale if it is treated as authoritative without source links and reviewer ownership.
System or demonstration context
A legacy codebase contains years of implementation knowledge but incomplete documentation, slowing modernization estimates and onboarding.
Source state and inputs
- Bounded repositories and file types
- Human review required
- Source state retained
Method
- Prompt contract
- Source-reference schema
- Reviewer worksheet
- Unsupported-claim log
- Versioned approval
Representative cases
- Define typed outputs
- Require source references
- Generate bounded drafts
- Review against code and tests
- Record accepted, rejected, and unresolved claims
- Version output with source state
Observed result or demonstration output
Shows how AI can accelerate draft documentation while human reviewers retain authority over approved technical claims.
Decision supported
Shows how AI can accelerate draft documentation while human reviewers retain authority over approved technical claims.
Artifacts retained
- Prompt contract
- Draft documentation
- Reviewer worksheet
- Unsupported-claim log
- Approved documentation release
Limitations and boundary
This method does not claim that AI fully understands a codebase or can replace accountable technical review.
Boundary statement: This method does not claim that AI fully understands a codebase or can replace accountable technical review.