A subpoena arrives. You have a week.

The demand names a patient, a date range and a party you have never corresponded with. Somewhere in the practice is every record that answers it, along with a corrected report that supersedes an earlier one, two copies of the same imaging study filed by different senders, and one document that should not go out at all. The work is not finding the files. It is producing them in a form that will still hold up when someone asks, eleven months later, what exactly you sent.

ChironAI OM assembles that production from the chart you already hold — numbered, indexed, attested — and keeps the record of it afterwards. The parts of the chart that are untidy are declared on the face of the packet rather than smoothed over, because those are the parts that get asked about.

Four things a records request actually asks of a practice.

1. 01

   Every record against the case, in one place

   The library

   Records arrive from everywhere — the employer, the claims administrator, prior treaters, imaging centres, the applicant’s attorney. Before you can answer a demand you have to know what you hold.

   • —The chart’s records organised at patient level and at case level, grouped by document type with a count per group.
   • —A preview offered only for formats a browser can actually render, rather than a viewer that fails in front of you.
   • —A records list that hit its page ceiling says so, rather than presenting a partial chart as though it were the whole one.
2. 02

   A production assembled against a specific request

   The packet

   A disclosure is not a folder of files. It is a response to a particular demand, from a particular party, on a particular date — and the packet has to carry that on its face.

   • —The requesting party, the purpose and the scope applied are recorded on the production itself and printed on its cover.
   • —Bates numbering with a per-production prefix seeded from the patient surname, every produced page numbered continuously, and the range recorded against each document.
   • —Records requested from a claims administrator or attorney, and the tokenized upload link an outside party returns them through.
3. 03

   It states what is wrong with the record

   The manifest

   Real charts are untidy. The same report is filed twice. A corrected version sits beside the original. A scan will not open. A packet that quietly tidies that away is the one that falls apart under examination, because the discrepancy is exactly what opposing counsel asks about.

   • —A superseded record is flagged on the manifest and preceded in the body by a banner, so no reader can mistake it for the current version.
   • —A byte-identical repeat is marked as a duplicate of the row that first carried it — noted, not silently dropped. Whether to omit it is the producing party’s decision at review, not a judgement the assembler makes on its own.
   • —What was deliberately not disclosed is listed as withheld, with its reason, in the same numbered sequence as everything else — so nothing is omitted silently, and what was withheld stays distinguishable from what was never considered at all.
   • —A document that cannot be rendered gets a placeholder page carrying the reason, rather than a gap in the numbering.
4. 04

   What went out, to whom, and when

   The record of the record

   The question that arrives months later is rarely "what is in the chart". It is "what did you send, and when did you send it".

   • —Every document in the chart shows whether it has already gone out, under which request, and in which Bates range.
   • —A production issued in error can be voided — with a reason and an actor, shown on every surface that lists it — while the record that the disclosure happened survives. Withdrawing a disclosure and erasing the evidence of it are different acts.
   • —A document produced only under a since-voided production says so, and stops saying so once it goes out again under a live one.
   • —A packet delivered to counsel travels by the same kind of expiring link, streamed by the server rather than redirected to storage, so no patient identifier is left in the recipient’s browser history or in any proxy between them and the practice.

A bound document, not a folder of files.

The other direction: a banker’s box arrives, and it is yours to read.

Disclosure runs both ways. The production that leaves the practice is one problem; the one that arrives is the harder one. A med-legal case can turn up as a thousand pages or three thousand, assembled by someone with no interest in making it easy to read, with the decisive four lines somewhere in the middle of it.

The record digest reads that file and returns a chronology. Eight of its behaviours are worth naming, and all eight are constraints rather than features — they describe what the system is structurally unable to do.

Those anchors are function names in the product, not a description of it. Why a constraint written in code is a different kind of promise from a constraint written in a prompt is the subject of how we prevent fabrication, which is the page to read next if the previous paragraph sounded like a claim rather than a mechanism.

One capability group out of thirteen.

Disclosure and production is a single group in the ChironAI OM inventory. The rest of it — the chart, the case lifecycle, causation, utilization review, the DWC forms and their clocks, and the revenue cycle — is enumerated on the capabilities page, and what shipped on which date is in the release notes.