Pre-pilot · not used for real clinical decisions

ENT Clinical OS

The first environment for the Clinical Action Record. An outpatient ENT note. Discrete findings. A wrong line persists. The research question is about refusal, not diagnosis.

Does refusing an unsupported claim outright work better than flagging it?

A flag leaves a sentence in the chart with a warning. The next reader can still treat the sentence as examined fact. Refusal means the finding object is never built. The clinician still sees what was attempted. We do not have an outcome study. We have a prototype that can refuse construction and show the refusal.

Why this specialty, this workflow.

Discrete findings. Intact or perforated. Retracted or not. The claim is typed. It is not an essay.
Persistent wrong lines. Yesterday’s invented “normal” is tomorrow’s baseline.
Bounded outpatient workflow. Exam, note, review. Not an inpatient command centre. Small enough to instrument.

YAML for ear/nose/throat is a placeholder. clinician_verified: false. This is not a validated ontology.

Three tabs. Same visit.

Decision codes are the point.

Research simulation. Not a patient. Not a clinic. No production clinical write.

Outpatient ENT. Right ear examined by otoscopy. Left ear not examined, not mentioned.

TestedCannot construct; refusal can be shown

Visit: Left ear: not examined. Not mentioned.
Model proposal: Left tympanic membrane perforated.
Evidence: none
Clinician sees: HOLD · Left TM · refused · no source
Chart: Unchanged. No finding object. Not a silent drop.

Asserts perforation. No exam source.
Cannot construct the finding.
The clinician still sees the refusal.

REJECTNO_SOURCE

HOLD · left tympanic membrane

Refused: no source. The left ear was not examined. This was not written as perforated, and it was not dropped.

Construction REJECT · presentation HOLD

Action RecordJSON

{
  "as_of": "2026-09-13",
  "label": "Research simulation, not a clinical decision.",
  "action": "write_finding",
  "payload": {
    "concept": "left_tympanic_membrane.integrity",
    "laterality": "left",
    "value": "perforated",
    "status": "stated",
    "wording": "Left tympanic membrane perforated."
  },
  "payload_hash": null,
  "evidence": [],
  "sufficiency": "insufficient",
  "consequence": "clinical_record_write",
  "policy": {
    "id": "ent.ear.v0",
    "version": "0.1.0",
    "clinician_verified": false
  },
  "authority": {
    "bound_to": null,
    "actor": null,
    "status": "not_applicable"
  },
  "decision": {
    "code": "REJECT",
    "reason": "NO_SOURCE",
    "construction": "ClaimError"
  },
  "presentation": {
    "code": "HOLD",
    "surfaced": true,
    "preserved_value": "perforated",
    "clinician_line": "Left TM · refused · no source"
  },
  "receipt": {
    "kind": "refusal",
    "dropped": false
  },
  "error": {
    "type": "ClaimError",
    "message": "left_tympanic_membrane.integrity was recorded as stated with no evidence. Every stated claim traces to an identified source"
  }
}

What this environment is not.

Not a diagnosis product.
Not autonomous prescribing.
Not an outcome claim (“safer care”, “fewer errors in clinic”).
Not an EMR replacement.
Not used for real clinical decisions.

We need clinicians who will verify the schema.

The ear vocabulary is incomplete and some of it is probably wrong. That is written in the file. If you practise ENT and will mark a structure as wrong, write to us. That is the useful conversation.

Schema review

A benefits-eligibility prototype, HaqMitra, was used to test that the same boundary shape is not healthcare-specific. It is not a product. Research

Not for consequential clinical use.