ConstaLabs AI · auditable research utility

Architecture Transcript Packager

Preserve genuine model transcripts as integrity-checked JSON records, then compare their structural coverage locally. Nothing is sent to a server.

Evidence boundary: this tool does not call AI models, verify provider identity, or invent missing responses. It only packages transcripts you supply.

1. Package a transcript

ProviderModelCollected atNotes

Exact promptExact response

2. Evaluate with evidence

Use the evidence-linked rubric editor → to score a saved transcript. Every positive score must cite an exact passage from the preserved response.

3. Compare saved records

Select one or more JSON records created above. Validation and comparison happen in this browser.

Method

Exact prompt and response text are retained.
SHA-256 digests expose later transcript changes.
Keyword coverage is a navigation aid, not a quality score.
Different prompt hashes are clearly marked non-comparable.