Resources

AI case chronology drafting — with human review

Search interest in “AI chronology drafting” usually means one of two things: a tool that invents a court-ready timeline from a dump of files, or a tool that proposes draft events a litigator can verify. exhibit.law is the second kind on purpose.

Drafting runs after the file has been read — transcript text when it exists, EXIF or embedded dates when they are real moments — so a draft is grounded in something readable rather than a filename. Humans still confirm before those drafts harden on the timeline, calendar, and ICS feed.

What “drafting” means here

In exhibit.law, an AI draft is a timeline event with review_status not_reviewed: a suggested title, description, and occurred-at, often linked to the media that produced it. It is assistive triage, not a filed pleading and not a declaration that the date is true.

The chain is deliberate: extract metadata → transcribe or OCR by kind → draft_event last. Writing the event before a transcript exists would summarize a filename. Skipping review would convert a model guess into calendar fact.

  • Drafts stay not-reviewed until someone confirms them
  • Real extracted metadata beats a model-suggested date
  • No transcript and no extractable date means no draft — not an upload-minute stamp
  • Bulk Approve refuses upload-time placeholders until a person sets a real moment

Provenance per field

A drafted card often mixes a real EXIF capture time with model-written prose. exhibit.law surfaces occurred-at provenance separately from title and description — EXIF, embedded metadata, suggested by AI, or upload time as a warning — so reviewers do not treat storage timestamps as when the thing happened.

Product demo: sample review queue refusing approve on upload-time dates — not real evidence
Sample data — not real evidence. Review blocks Approve when occurred-at is only the upload minute.

Where the drafting runs

Under normal load, transcription, OCR/vision, and drafting run primarily on closed, self-hosted models on private infrastructure. Peak demand may use third-party APIs under no-training terms. Either way, output stays behind review. Details live on the AI Disclosure page — not in a marketing claim that AI “builds your case.”

Try a small drafting pass

Create a free account, upload a short A/V clip or a PDF with a text layer, process the batch, and open Review. Confirm or fix dates before anything hits the calendar.

Related: how to build a case chronology, case timeline overview, discovery management, and AI Disclosure.

Create free account · Contact · All resources