Analysis
Evidence-grounded AI coding review
Review cited AI coding suggestions with rationale and explicit human decisions.
Implemented task
Public task
Open a review item or select uncoded evidence.
Expected outcome: A human-authored or human-reviewed evidence decision is persisted with exact revision history.
Declared product contract
Inputs and task boundary
Inputs
- A project coding suggestion with an exact evidence quote and rationale
Input constraints
- A suggestion remains unconfirmed until an authorized human or declared review policy decides it.
User actions
- Open a pending coding suggestion.
- Inspect the cited evidence, proposed code, confidence, and rationale.
- Confirm, modify, reject, or route the suggestion to dispute.
Outputs
- A durable human decision or unresolved dispute bound to exact evidence.
Decision boundary
Where it fits
- Researchers reviewing evidence-grounded coding suggestions.
Outside the boundary
Where it does not fit
- Treating generated codes as reviewed findings without a visible decision.
Verification
Product proof
journey
proof.m2-analysis-kernel
- Verified
- Review due
- Expires
- All five locator kinds, overlap, multiple codes, and exact evidence return are covered.
- Codebook, memo, history, research connections, and withdrawal child journeys pass.
- Every child artifact is hashed and rerun from a fresh database.
Failure boundary: The proof covers the completed M2 kernel, not M3 cases, matrices, or later collaboration rounds.
OpenVerbatim is an open-source (Apache-2.0) qualitative data analysis platform for coding and analyzing interview transcripts. AI-suggested codes stay marked as suggestions until a human reviewer confirms or rejects them, and every decision is kept in an audit trail. The full feature set is available when self-hosted; there is no paid feature wall.