Same visit, two versions — see exactly what changed.
Altered medical records detection surfaces the difference when near-duplicate pages differ — late amendments, addenda, and altered entries shown side by side, change by change, from the produced PDFs themselves. A delta is a signal, not a verdict: what changed and when each version was produced is documented; what it means stays a human read.
The near-misses are the point.
Deduplication removes pages that are identical. This feature keeps the ones that almost are: the same visit note appearing in two packets with different text is paired, not discarded — because in a version dispute, the difference between them is the evidence.
Change by change, not a vague "these differ."
The diff is at the level a cross-examination needs: which sentence was added, which words were edited, what disappeared. Each change is highlighted on both page images and cited to its exact location, ready to drop into an exhibit or a motion.
When it was produced matters as much as what changed.
Each version carries its provenance: which production it arrived in, when, and where it sits in that packet. A note that gained a sentence between the first production and the one after the demand letter — that sequence is documented on the card, from the PDFs alone, no EHR audit log required.
Documented, cited — never labeled tampering.
An amendment can be honest medicine; a late entry can be routine practice. So the output is deliberately modest and audit-grade: what changed, cited to both versions, and when each was produced. The conclusion belongs to the attorney or evaluator reading it — which is exactly what makes it usable in front of a judge.
See Verifiable AI CitationsFrom two packets to a cited delta card.
Three steps — the comparison a human reviewer would never have time to run on every page.
All packets, including supplemental drops — each hashed and logged at ingest.
Same-visit pages compared change by change; identical pages deduplicate, differing ones surface.
Side-by-side versions with production dates — export as a cited exhibit or keep working the file.
Who reads the deltas.
Anyone whose case can turn on a note that changed between productions.
Malpractice and PI teams get the version question answered from the produced record — and a cited exhibit for the deposition where it counts.
For law firmsAn IME opinion built on version B should say so. Evaluators see both versions before the exam, and the report cites the one relied on.
For evaluatorsAlteration detection, answered.
No — and it doesn't try. Amendments can be perfectly legitimate, and a late entry can be routine practice. The feature documents what changed between versions and when each version was produced; what the change means stays a human read. It never labels a delta as tampering.
No. It works from the produced PDFs themselves — the packets you actually received. When two productions contain near-duplicate pages for the same visit and they differ, the delta is surfaced from the documents alone, no EHR access required. If you later obtain the EHR audit log, the flagged deltas tell you exactly where to look.
Deduplication removes pages that are identical, so reviewers read less. Alteration detection does the opposite with the near-misses: pages that are almost identical but not quite are exactly the ones it keeps, pairs, and diffs — because in a version dispute, the difference is the evidence.
Change-level diffs: text added, text removed, and text edited between the two versions, highlighted side by side on the page images. Each change is cited to its exact location in both versions, so the finding drops straight into a deposition exhibit or a motion.
Yes. Each version carries its provenance: which production it arrived in, the production date, and where it sits in that packet. A note that reads differently in the packet produced after the demand letter than in the one produced before it — that timing is on the card.
Related capabilities
Identical pages removed and logged — the other half of this comparison.
ExploreLate-arriving packets folded in — and compared against what came before.
ExploreVersion deltas land beside the case's other flagged signals.
ExploreProof that the versions you're comparing arrived exactly as produced.
ExploreTwo productions in the file? See what changed between them.
Upload the packets and get the version pairs back, diffed and cited to both sources. Handled under our BAA; never used to train a model.