The bottleneck isn't the order. It's the protocol.

1 — What protocoling is

An approved order is not an instruction a scanner can follow

An approved imaging order is not an instruction a scanner can follow. Somebody has to turn "CT abdomen and pelvis" into this scanner, this contrast agent at this dose by this route, these phases, these delays, this coverage, this slice thickness, this breath-hold — and that somebody is usually a radiologist, working from a queue, interrupted, with half the clinical picture. That translation step is called protocoling, and it is where the money and the minutes actually go.

1. Fact 1

   Done by the most expensive person — for the wrong reason

   A radiologist's value is interpretation. Protocoling is clerical translation performed by an interpreter — non-interpretive work that consumes a material share of the workday without being the reason the hour is expensive.

2. Fact 2

   Where bad orders are finally discovered

   When the order omits a load-bearing fact — leukocytosis, prior pathology, implant status — the protocol is wrong, and the study is changed or cancelled after the patient is already scheduled. Every one of those is a phone call, a rebooking, and sometimes a wasted scanner slot.

3. Fact 3

   Variation is a quality defect nobody measures

   Twelve scanners in one system routinely produce more than twelve different acquisitions for the same clinical question. There is no published number, anywhere, for how much protocol variance a department has — which is why publishing one is worth so much.

4. Fact 4

   The scanner minute is nearly all fixed cost

   A minute recovered on a scanner is close to full contribution, because the capital, the room and the staffing are already paid for. Seconds in, minutes out — and the minutes come out of a bottleneck while the seconds come out of slack.

3 — What ARKA does

Six stages. Four autonomy tiers. Not autonomous.

Not autonomous. The radiologist stays the arbiter. Auto is a standing order a named physician signed — not a software decision — and every auto-protocoled study is reversible before acquisition.

1. 0

   Context

   Assemble what the scanner decision actually depends on: the reconstructed indication, plus kidney function, pregnancy status, allergies, implants and devices, weight, sedation history, breath-hold capability, IV access, NPO status, isolation precautions and acuity markers. Note what is not here: any image.

2. 1

   Candidate generation

   A deterministic, site-configurable table proposes an ordered set of candidate protocols. Where the table has no answer, retrieval over the site's own historical decisions for similar cases proposes one. A language model may sit behind this interface — boxed, replaceable, and never allowed to clear a gate or set the autonomy tier.

3. 2

   Seven safety gates

   Every candidate passes seven deterministic checks — renal contrast, pregnancy, mri conditionality, allergy, physiologic feasibility, sedation, coverage adequacy — each returning block, modify, warn or clear with a cited reason. A gate is never advisory and never probabilistic.

4. 3

   Ontology bridge

   Translate the surviving protocol through four layers: the clinical concept → the public LOINC/RSNA Radiology Playbook code → this site's RIS orderable and chargemaster entry → this specific scanner's own protocol identifier.

5. 4

   Tiered autonomy

   Auto, Assisted, Escalate, or Return-to-sender. Which tier applies is decided by a rule the site's radiologists have signed — attestation for a named indication class — not by a confidence score alone.

6. 5

   Return to sender

   When the engine finds that the order is wrong — not the protocol — it sends a structured message back to the ordering clinician with a one-click corrected order attached. The "are you sure you want to do this?" phone call, fired before the patient is on the table instead of after.

Autonomy tiers

• Auto

  Protocol applied inside a boundary a named licensed physician at that site signed — indication class, modality, region, gate-clear — at a recorded version and date. Reversible by a radiologist at any point before acquisition.

• Assisted

  Ranked candidates with the rule that fired, gate results, and losing alternatives on screen. The radiologist confirms or edits before write-back.

• Escalate

  The engine will not decide. Missing attestation, unresolved gate, non-deterministic candidate source, or confidence below the attested floor puts the row in the radiologist's action queue.

• Return to sender

  The order is the defect, not the protocol. A structured correction goes back to the ordering clinician with a one-click corrected order — before the patient reaches the table.

5 — What we measure

Twelve metrics. Baseline capture is mandatory and free.

Baseline capture is mandatory and free. Measured minutes and shares convert to dollars only with a buyer-supplied unit; the product is labeled modeled and never writes back.

6 — The TCO model

Recovered capacity first. Your rate second.

Radiologist capacity is modeled from protocol metrics 2, 3 and 9 — hours render immediately; currency waits for your teleradiology or locum contract rate. Scanner-minute dollars use a published modeled derivation. Avoided-utilisation dollars need your unit cost.