An EMR is judged on the twenty minutes between patients

Ambulatory

What is this patient here for, and what did we agree?

The outpatient day, in the order it happens: the schedule, check-in opening the encounter, the chart, the scribe drafting alongside the consultation, and the coding surface at the end of it. The draft and the chart are visibly different things on this screen, because the moment they look alike is the moment one is mistaken for the other.

• A draft note is marked as a draft everywhere it appears, including on the way out
• Proposed codes shown with what they were derived from, accepted one at a time
• The problem list, allergies and the core data set as structured resources rather than prose
• Nothing on this screen commits to the record without an explicit signature

Inpatient

Who can go home, and where is the ward stuck?

A bed board over the real ward and bed hierarchy, and a flow view showing readiness for discharge and where the bottleneck actually is. Assignment comes from the deterministic rules engine, so the board shows what the rules permit rather than what a model thought would be tidy.

• Admission, transfer and discharge as movements through a real location hierarchy
• Census computed from the assignment ledger, not from a query that has drifted
• Discharge-readiness flags from an agent, acted on by a person
• Bottleneck flags that name the step, not the ward

Orders and results

Has it come back, and has anybody seen it?

Order entry and the results that answer it, bound together. The unanswered order is the interesting object on this screen — a result that arrived and was never seen is the failure mode this surface exists to make visible.

• Every result tied to the request that asked for it, never filed loose against a patient
• Critical values routed to a named clinician, flagged and never interpreted
• Observations, reports and imaging pointers in one trail per encounter
• Acknowledgement recorded, so “nobody saw it” is a question with an answer

Pharmacy

Is this safe to dispense, and do we actually have it?

Formulary, inventory, dispensing, the administration record and retail point of sale. The safety checks on this screen are deterministic engine output rather than model output, and they block rather than warn — a warning in a busy pharmacy is a thing that gets dismissed.

• Interaction, allergy and dose-range checks that hard-block, with the reason shown
• Lot and expiry tracked, and dispensing that takes the oldest-expiring stock first
• Controlled substances honouring their schedule, including the ones that never refill
• Reconciliation across transitions, presented for a pharmacist to sign

Revenue cycle

Why has this not been paid?

Billing, coverage, claims, prior authorisation and the denial trail, with the agent queues that draft into them. The charge on this screen traces back to the encounter that produced it, so a disputed line is answered from the record rather than from memory.

• Every charge traceable to the clinical event behind it, pharmacy charges included
• Denials categorised from the remittance, with a drafted appeal a human sends
• Charge-integrity flags raised before the bill goes out rather than after it bounces
• Prior-authorisation status tracked as state rather than chased by phone

The review queue

What did the agents do while I was on the ward?

One queue for the whole roster. Everything Tier A waiting on a signature, and everything Tier B that has already acted and can be overridden, in one place — because a review surface per agent is a review surface nobody opens.

• Drafts awaiting signature, with what the agent was given and what the critic said
• Completed Tier-B actions with a one-click reversal and the state they changed
• Every override stored, attributable, and reportable as a measure of the roster
• Cost and latency per invocation visible, so the agents are accountable as a line item