How it works

Three CSVs in. A cleaner card out. Every month.

ClearCard onboards in a single afternoon: three exports a facility already has, no EHR integration, no IT project.

Step 1: Inputs

The three files you already have

preference_cards.csv

What each surgeon asks for on every procedure card.

case_usage.csv

What was actually opened and used across past cases.

item_catalog.csv

What each item costs, so every recommendation carries dollars.

Step 2: Composite scoring

Three layers, one swappable pipeline

No single black box. Each layer answers a different question, behind a swappable Scorer interface.

Bayesian probability

The odds a given item goes unused, learned from the surgeon’s own history.

Quantile quantity

The right-sized quantity to keep on the card versus hold in the room.

Trend

Whether usage is drifting over time, so recommendations stay current.

Steps 3–5: Explain, approve, export

Deterministic recommendations a surgeon signs off

Flagged items become plain-language recommendations rendered locally from templates (no LLM, zero hallucination by construction), each citing the surgeon’s own usage, a peer comparison, and a dollar impact.

03

Explain

Local, deterministic templates. Same inputs, same words, every time.

04

Approve

Surgeon approves, tweaks, or rejects each line. Nothing is automatic.

05

Export

The cleaned card exports back; next month’s usage re-enters the loop.

Walk the whole loop yourself.

The demo runs the full pipeline on a synthetic facility.

Open the demo →