Emergency medical dispatch software captures the conversation between the caller and the telecommunicator. It scripts the questions, recommends the response, and reads the pre-arrival instructions. That part of the job is well served.
The radio is a different stream. The dispatch that goes out, the unit that acknowledges, the crew reporting from scene. It is transmitted, heard, and gone, and no protocol system is built to hold it. When a call is questioned three months later, the account nobody can produce is the one that carried the operational truth.
CareCompile Dispatch holds that stream. Permanent, searchable, scored, with the original audio behind every line.
| Centre | Mean | Against 70 | Spread | Determinant | To disp. |
|---|---|---|---|---|---|
| Yellowstone County 911 9 reviewed |
87 | ±8.3 | 89% | 25s | |
| Missoula County 911 5 reviewed, 1 below 70 |
77 | ±25.8 | 60% | 30s | |
| Great Falls Cascade County 911 9 reviewed, 3 below 70 |
70 | ±13.7 | 33% | 50s | |
| Butte-Silver Bow 911 9 reviewed, 8 below 70 |
58 | ±19.2 | 33% | 64s |
Same measures, computed identically at every centre. A centre cannot see how it compares, and a programme cannot correct variation it never measures.
Trunk-recorder, rtl_airband and SDRTrunk do not expose an API. They write one file per transmission to a directory. We read that directory, which is the one integration that does not break when your recorder updates.
Speech to text runs on dedicated GPU hardware, not a public API. Operational radio is sensitive material and it should not leave the building to be understood.
A poorly captured transmission cannot produce a confidently asserted finding. Confidence is carried forward and caps the score, so a bad capture never masquerades as a clean one.
Recordings are held on the filesystem, unreachable by URL, and served only through an authenticated endpoint. Every listen writes an audit row against the person who opened it.
Scripted case entry. Key questions for each chief complaint. Call prioritisation. Sequential pre-arrival instruction panels. Protocol content maintained by a clinical authority and pushed as updates. Certification and recertification for every telecommunicator who uses it.
That is a regulated clinical product with real vendors behind it and decades of protocol governance. We do not build it, we do not resell it, and we will say so in the room before anyone has to ask.
The dispatch transmission. The unit acknowledging. The responding crew on scene. Mutual aid, air medical, the landing zone, the hand-off. None of it passes through the protocol system, so none of it is held by one.
It is the record of what the service actually did, and today it exists only as an audio file in a folder that nobody searches, or as nothing at all.
The practical consequence: we are not competing with a protocol vendor and we do not displace one. A centre keeps whatever it runs for call-taking. We hold the stream that system was never built to reach, and the two records read better together than either does alone.
An accredited programme has to pull recorded calls and score them for protocol compliance. That work is manual, slow, and the first thing to slip when a centre is short handed. So it gets sampled, and the sample is small.
Transcribe everything and the sample stops being a compromise. Score every call on the measures that are actually observable on air, and a supervisor stops guessing which calls to pull.
Observable on air, and therefore defensible. Not graded against licensed protocol text.
| Measure | What it answers |
|---|---|
| DETERMINANT | Was a determinant assigned and stated |
| PRE-ARRIVAL | Was instruction given before the unit arrived |
| ADDRESS | Was the location confirmed on the transmission |
| CALLBACK | Was a callback number confirmed |
| TO DISPATCH | Seconds from call to unit assignment |
| CAPTURE | Transcription confidence, which caps all of the above |
The same measures applied identically at every participating answering point. Comparable because they are computed the same way, not because everyone agreed to fill in the same form.
A state average hides the centre that is struggling. We report the variation between centres and the count of calls below threshold, because that is where a programme actually intervenes.
Onboarding a centre is a recorder output directory and a channel definition. Call-takers do not learn anything new, which is what makes a small rural centre as cheap to add as a large one.
Centre-level access to its own calls and scores, with the comparison available to the programme. Audio access is authenticated and logged at every level.
Every transmission transcribed and threaded to an incident, scored on measures observable on air, with the recording behind each finding and an audit trail on who listened. Running today on the console.
The same measures computed identically at every answering point, reported as a spread rather than an average, so a state can see which centre is drifting instead of learning it from a complaint.
Dispatch radio carries the patient before the ambulance does: complaint, age, acuity, unit, estimated arrival. Structured from what was said on air and handed to the emergency department as HL7 before the doors open, it turns a surprise into a preparation. Designed against our existing HL7 and FHIR stack. Not built, not deployed, and gated behind a rule that refuses to send when confidence or completeness falls short.
The chain described here runs today in a security and incident-monitoring deployment, transcribing operational radio and turning it into structured incidents with the audio retained behind each one. What this page describes is that engine, aimed at emergency medical dispatch.
The console is open to anyone we give access to. It runs on generated traffic so that a conversation can start without a data agreement, a procurement, or a single real recording changing hands.
We have no deployed United States public-safety customer today. We are not an emergency medical dispatch protocol vendor, we do not sell call-taking software, and we do not license or resell clinical protocol content.
The pre-arrival notification described above is designed and not built. Nothing on this page should be read as a claim that it is running anywhere.
Everything on the console is synthetic. No real dispatch traffic, no real addresses, no real patients.
Recording and review of voice traffic is governed by state law and consent rules differ between states. A deployment is designed around the law of the state it runs in, decided before the first recording is processed rather than after.
If you are procuring an emergency medical dispatch system and want a quality assurance layer behind it, or you are a prime looking for that component, that is the conversation where we are useful.