A four-year-old, 14 kg, casualty at 11 p.m. Bronchopneumonia. The intern writes the amoxicillin order: 250 mg three times a day.

The senior, looking over his shoulder during a brief handover pause, asks: "What dose are you using?"

"Forty per kg per day."

"That's 560 mg per day. Three times a day is about 186 mg per dose. You wrote 250."

The intern corrects it. It is a minor error. The drug is amoxicillin, the patient is robust, the dose he wrote was a third over instead of a tenth under. He moves on, embarrassed.

Most pediatric dosing errors look like this. Not catastrophic. Slightly wrong. Hard to notice unless someone checks.

The published rates are worse than residents are taught to expect. Kaushal and colleagues (JAMA, 2001) audited a paediatric inpatient ward and found a medication error rate of 5.7 per 100 orders, with 57% being potential adverse drug events — errors that could have caused harm if not caught. The actual harm rate was lower (1.1 ADEs per 100 admissions), because most errors got caught by a pharmacist or nurse or, as above, the senior glancing at the order.

The robust catch system is mostly people. In a busy casualty without a pharmacist embedded in the workflow, the catch fails.

Medows handles pediatric dosing by being patient-grounded. When the intern types the amoxicillin order, the patient's weight is on the screen. The standard dose range for amoxicillin in pediatric bronchopneumonia (40–50 mg/kg/day divided three times) auto-populates as a recommended range below the dose field, derived from the AIIMS pediatric formulary. If the typed dose falls outside, the line greys with a soft note: "Outside typical range for this weight."

This is not a hard stop. It is a soft signal — the way a pharmacist looks up from their desk when an unusual dose lands. The intern reads it, adjusts if needed, moves on. The patient gets the right dose. The dose-checking is adjacent to where the order is being written, not a separate workflow.

The 5.7-per-100 number from 2001 is, for what it is worth, almost certainly the same today in the average busy unit. The catches happen because of the doctors and the pharmacists, not because the technology improved.

If the AI consult already knows the patient is 14 kg, it should know what the dose range is. That is the bar.