A hospital, end to end: one ontology from the front desk to the pharmacy

01. The hospital twin

Hospital teams, when agreeing the model before importing data.

The middle slab is the hospital ontology. Patients, encounters, beds, wards, clinicians, orders, medicines, batches, fridges, suppliers, theatre sessions and roster shifts share reviewed links. Agents, what-if runs and reports read that model. Approved sources and memory sit below it.

An encounter is admitted_to a ward and assigned_to a bed. An order is prescribed a medication and dispensed_from a batch. A batch is stored_in a fridge and supplied_by a supplier. These are reviewed links, not facts an agent invents.

Hospital modeller: "What can we ask from today's import?"

Eunomia, coverage: "Check the approved terms, metrics and source coverage first. Missing inputs remain visible. Live HL7 or FHIR feeds and a dedicated hospital twin UI are coming."

02. Patient and family

Nora and Elise, from arrival to consent, discharge and follow-up.

The ontology holds Nora's patient reference, the ED encounter, a recorded preference to involve Elise, and later the discharge and follow-up references. Nora does not need to learn the word ontology. Staff use the model to answer her questions, then check the source. This is not a shipped patient portal.

Elise: "How long will Nora wait?"

Registration officer, asking Eunomia: "What was today's ED median wait?"

Eunomia, metric: "38 minutes in the selected ED summary. Source: ED-SUM-09, exported at 09:00. That describes earlier attendances, not Nora's place in the queue."

The triage nurse explains that urgency determines who is seen next. The treating team discusses surgery with Nora and records consent in the clinical system.

Nora: "Can you explain it with Elise here?"

Nurse, asking Eunomia: "Recall the recorded communication preference for this encounter."

Eunomia, recall: "Nora asked for Elise to join explanations. Source: intake note IN-104, 08:35. The consent form is a separate record, CONS-104."

A remembered preference is not consent to a procedure, or permission to share every record. At discharge, a nurse and pharmacist explain the approved instructions and check Nora's understanding.

Nora: "Which instructions do I take home, and when is follow-up?"

Discharge nurse, asking Eunomia: "Recall Nora's signed discharge plan and confirmed booking."

Eunomia, recall: "Plan DS-104 version 2 and medicine list MED-104 version 3 are the recorded documents. Booking BK-104 is confirmed for the fracture clinic on 23 November at 10:00. Source: booking export, 12:10."

Staff hand over the checked documents, including the care team's advice on when and how to seek help. They book the appointment in the booking system. Eunomia recalls its confirmation; it does not make the booking here.

03. Registration and admin

Intake, bed and theatre coordinators, while planning the admission.

Intake links Nora to encounter ENC-104, rather than creating a new person for every department. Admin sees identity and booking references, the bed request, the theatre session and the relevant roster. Clinical detail stays with the clinical team.

Bed manager: "Which staffed orthopaedic beds are free in the latest snapshot?"

Eunomia, recall: "Beds O-12 and O-14 are marked available. Source: BED-0900. O-12 has a cleaning note; check its release before assigning it."

The manager confirms a bed in the bed system, then imports the updated snapshot. The theatre coordinator links ENC-104 to tomorrow's planned session, TS-10A, and checks the nurse roster with the ward manager.

Theatre coordinator: "What does moving TS-10A affect?"

Eunomia, impact: "The linked case list, recovery-bed plan and roster shifts. Each path is shown. The booking team must confirm changes in the scheduling system."

Billing staff attach the episode and procedure coding references after clinical documentation is complete. A qualified coder assigns and checks the codes. Eunomia recalls the coding status and missing documents, not a guessed bill or an automatically chosen clinical code.

04. Triage and ward nurses

Triage and ward nurses, at assessment, rounds, handover and escalation.

The triage nurse records category 3 in this fictional encounter, based on their assessment. Eunomia recalls the recorded category; it does not triage Nora. The clinical vault holds references to observations, orders, falls-risk assessments and handovers, with who recorded them and when.

Ward nurse: "What needs checking at handover for Nora?"

Eunomia, recall: "Handover HN-104 records assisted mobilisation and a falls-risk plan. Observation set OBS-104 and the active orders are linked. Read the current chart before care."

At medication rounds, the nurse checks the live medication chart and the patient at the bedside. A copied order is context, not authority to administer a medicine.

Later, the hospital's existing observation system raises a deterioration alert. The nurse follows the hospital's escalation process immediately. Eunomia is not the monitor or alarm.

Nurse, after escalating: "Recall the latest handover and responsible team for this encounter."

Eunomia, recall: "HN-104, recorded at 14:00, links the ward consultant and current team roster. The imported observation snapshot is older than the alert. Use the live chart."

The time on a source matters as much as the words in it.

05. Doctors in ED, theatre and the ward

The ED registrar, surgeon and ward consultant, as Nora moves through care.

The ED registrar records Nora's history and examination, requests imaging and enters orders in the clinical system. The ontology links those orders and result references to ENC-104.

ED registrar: "Recall the history and outstanding results recorded for ENC-104."

Eunomia, recall: "ED note ED-104 and imaging request IMG-104 are linked. The export marks the report as pending. No final result is present in this snapshot."

The surgeon reviews the actual imaging, examines Nora and agrees the plan with her. Before theatre, they check the list, consent and the hospital's safety process.

Surgeon: "Which encounter and consent reference are linked to Nora on tomorrow's list?"

Eunomia, recall: "ENC-104 is scheduled_in TS-10A. CONS-104 is the linked consent reference. Source: theatre export TH-1600. Confirm against the current list and consent record."

After surgery, the ward consultant reviews recovery with nurses, pharmacy and allied health. For discharge, the consultant signs the summary in the clinical system. Eunomia can recall the signed version and linked results. It does not decide Nora is fit for discharge or create an unsigned plan that looks final.

06. Pharmacy and ward stock

The pharmacy lead, at stock review, cold-chain checks and a supplier shortage.

Pharmacy lead Mira Cole sees medicines by ward and imprest, the agreed stock kept on each ward. Every stock position needs a medicine, location, usable quantity and time. Batches carry expiry dates and costs. Cold-chain batches link to fridges. Supplier links hold delivery and shortage notes.

Mira: "How many days of IV fluid cover does Ward O have?"

Eunomia, metric: "6 days: 120 usable bags divided by the agreed 20 bags a day. Receipt: stock position IV-O-09, 09:00. Quarantined stock is excluded."

Mira: "What stock expires in the next 30 days?"

Eunomia, metric: "A$480 in the selected batch snapshot. Receipt: 40 units in batch EXP-01 at A$12 each. Review the expiry filter and stock count before action."

A fictional supplier, Wattle Medical Supply, reports a delayed IV fluid delivery. Mira asks for an impact view of the linked stock positions and ward demand records. She checks alternatives with procurement and clinicians. Eunomia does not choose a substitute medicine.

For controlled drugs, the model would hold a restricted reference to the hospital's S8 register and its reconciliation status. The legally required register, authorised checks and local procedures remain in the hospital's approved system. This example does not claim Eunomia is an S8 register.

07. Executive and operations

Operations and finance leads, at the daily capacity review.

Operations gets dated, aggregate snapshots in its admin vault. The executive can ask about bed occupancy, ED waits, theatre utilisation, staffing gaps and the cost per bed day. Finance and clinical teams agree the definitions first.

Operations director: "Where are we tight today?"

Eunomia, metrics: "Bed occupancy is 90%: 315 of 350 staffed beds. Theatre utilisation is 75%: 360 of 480 available minutes. Two roster shifts are unfilled. Allocated cost per occupied bed day is A$1,200: A$378,000 divided by 315. Receipts: BED-0900, TS-DAY-09, ROSTER-09 and FIN-09."

The numbers use the same time window and the selected summary records. If ten beds close for staffing, the denominator must change. A building with 350 physical beds is not always a hospital with 350 usable beds. The executive sees the reporting gap when a summary is missing, rather than a reassuring zero.

08. Scenario: an overnight fridge failure

Pharmacy and ward teams, after the 07:00 cold-chain incident import.

At 07:00, an imported incident says fridge F-02 failed overnight. The pharmacy team follows its cold-chain procedure and quarantines affected stock pending assessment. The failure report comes from the hospital's monitoring process, not a Eunomia sensor feed.

Mira: "What does F-02 affect, including today's due doses?"

Eunomia, impact: "F-02 links to insulin batch INS-27, then ward stock in O and M. Linked order records identify four patient references with doses due today. Sources: incident CF-02 at 07:00, batch snapshot at 06:00 and due-order export at 07:10."

The agent follows the recorded paths, checks due dates and shows the reason each record is included. For a path longer than one neighbourhood query, it walks the next links explicitly. These are potentially affected orders, not evidence that a patient received compromised stock. An authorised pharmacist and ward team check the current orders and arrange safe supply.

09. Scenario: a two-week flu surge

Operations, workforce and pharmacy leads, before changing a roster or order.

The operations director asks the question before changing the roster or placing an order. The team prepares one approved planning record for each resource, covering the same 14-day window. These hold baseline demand, usable capacity and the team's assumed surge requirement.

Operations director: "Test 20% more arrivals for 14 days. Use our agreed resource assumptions. Save nothing."

Eunomia, what-if: "Under those assumptions, arrivals move from 2,800 to 3,360. Peak occupied-bed demand moves from 315 to 378, which is 28 above capacity. Required nurse shifts move from 280 to 336, which is 36 above the 300 available. Oseltamivir demand moves from 100 to 120 courses against 110 available. IV fluid demand moves from 1,000 to 1,200 bags against 1,100 available. Nothing was saved."

The team explicitly assumes each listed requirement rises by 20%. It scales those properties in scenario_run; approved metrics read demand and capacity to show the gaps. Peak beds are a peak count, while shifts and stock are totals for the fortnight. They are not added together. These flu planning totals are separate from Ward O's daily stock example.

This is arithmetic on stated assumptions, not a forecast, a staffing recommendation or a treatment plan. Raising arrivals alone would not automatically alter linked beds, rosters or medicines. Clinical and operations leads must supply a defensible demand model, review the result and make any real changes in the hospital's systems.

10. Scenario: a batch recall

The authorised pharmacy team, after supplier notice RC-42 arrives.

Later, a separate supplier notice recalls batch AB-42 of a fictional discharge medicine. The trace goes through dispensing events, rather than assuming everyone on a ward received the batch.

Pharmacist: "Find every patient dispensed AB-42 in this imported trace."

Eunomia, recall and impact: "D-104 links AB-42 to Nora's reference; D-118 links it to patient P-118. Source: dispensing export DP-1800 and recall notice RC-42. This is every match in that export, not a guarantee that the export is complete."

Pharmacy reconciles the full dispensing record, checks stock and follows the recall procedure. The authorised team contacts patients and records the outcome in its own system. Eunomia does not send recall messages or change prescriptions in this example.

11. Scenario: governance and receipts

The agent and pharmacy vault owner, when a new kind needs approval.

Strict mode is switched on by the pharmacy vault owner. It is off by default.

Agent: "Create a cold_chain_incident for F-02."

Eunomia: "That custom kind is not approved. The write is refused and a proposal is filed."

Mira, the vault owner: "I will review cold_chain_incident, its properties and links in Memory spaces, Kinds and links."

Mira approves it in her browser session, then the agent retries the write. The agent's API token cannot approve its own proposal. Readiness shows outstanding proposals; coverage shows whether a question has an approved term or metric behind it.

Numbers use certified metrics through metric_eval, with dated source receipts. Certified means the calculation has been reviewed. It is not external certification or a clinical endorsement.