← Back to ARTHUR

Settings

Site-level configuration for Menomonee Falls Community Cardiology Providers

Billing

Monthly Auto Billing
Automatically generate EPIC charge capture and progress notes for eligible patients each month.

Expert Mode

Enable Expert ModeiExpert Mode grants designated providers two extra powers:
Custom goal & range — drag the PAD goal and acceptable-range lines on a patient's chart to set a custom target. Custom values are labeled CUSTOM and override AAG algorithm-derived targets.
Manual Override — on an AAG(3)/AAG(4) alert, dictate a free-text intervention. ARTHUR structures it and records it like an algorithm intervention (cooldown + post-burst effectiveness tracking), bypassing the adversarial review.
Allow designated providers to manually set custom PAD goals and acceptable ranges per patient. Custom values are labeled CUSTOM in the database and override AAG algorithm-derived targets.

ARTHUR Model Tuning

Model role
Each ARTHUR model is tuned and DOJO-trained in isolation — pick which one to dial in.
Model and sampling parameters for AI-drafted patient email replies (arthur-clinical:patient-facing). Custom parameters apply only to reply drafts — the safety-reply classifiers keep the tight modelfile defaults. Use the testbed below to try settings before saving.
Use custom parameters
Off = replies use the modelfile defaults shown in gray.
Model
Ollama tag the email path calls. Overrides the bridge's requested model.
Temperature
Higher = more varied, less canned prose. Try 0.7–0.9 for warmth.
Top P
Nucleus sampling. The current 0.33 is very restrictive; 0.9 is typical for prose.
Top K
Repeat penalty
Max reply tokens
Hard cap on draft length (num_predict).
Testbed — draft a reply with the settings above (nothing is sent or queued)