Work the case with the system
Two-way: the assistant brings recommendations, and you can pose a hypothesis, order an investigation, or challenge it back. Everything runs through the confidence cascade and DRLOOP authority — nothing acts on care without you.
Ask CLIN·AI
Ask in plain language — answers are drawn from this patient's record and cite their sources. Verify before acting.
Answers are AI-drafted from this record and cite their sources. The system proposes; the clinician decides — verify against the source before acting.
Safety catches · this patient
The guardrails acted automatically — you confirm each one was right.
The pathway engine proposed a breathing/meditation micro-action. Contraindications (GUARDIA) blocked it — relaxation-induced panic risk — and substituted body-scan & labelling.
AVA drafted a reply that mentioned a penicillin-class option. Allergy screening (MEDI) caught the allergen and the message was rewritten before it reached Marie.
The safety layer acts automatically — this is your doctor-in-the-loop confirmation that each catch was appropriate for Marie. Aggregate guardrail performance (how often they fire, across everyone) lives in the Validation Console.
Assistant recommendation
You decide (DRLOOP)
Early signal supports your hypothesis: on skipped-lunch days, afternoon energy drops ~1.4 points independent of prior sleep. Low-burden, and it targets the worsening meal signal directly.
Held back — low confidence
Autonomy gate
Below the 0.30 gate — the system will not make a recommendation here. It surfaces the gap instead: no recent thyroid panel on file. Consider ordering TSH/FT4 at the next stress test to resolve it.
Pose a hypothesis
You → the system
I think her afternoon crashes are driven by skipped lunch, not sleep — check the interaction of sleep × energy × meals × caffeine over the last 3 weeks.
Order an investigation
Fans out to three actors
Activate a 2-week food + energy journal, prompted at meals.
Collect & correlate meals × energy × sleep × caffeine; flag interactions.
Review reminder at 2 weeks with the correlation read-out.
Active inquiries
2 running