Glossary
Recommendation Engine
Updated February 28, 2026
The recommendation engine is the path from your inputs to ranked suggestions, through filters and guardrails.
Why it matters
It helps explain why a suggestion appears, disappears, or pauses over time.
Input -> filters -> ranking -> guardrails
- Inputs: goals, constraints, adherence history, medication context
- Filters: safety checks, duplicate risks, contraindication rules
- Ranking: weighted scoring across evidence and behavior signals
- Guardrails: consistency caps and emergency-style safety overrides
What can be overridden
- timing windows and personal preferences can be adjusted by the user
- override fields are applied only within defined safety limits
- hard safety flags remain locked until conditions are resolved
Trust boundary
Recommendations are behavioral guidance, not diagnosis.
They are designed to be adjusted with clinical context when uncertainty is high.
Practical action step
Before trusting a top suggestion, confirm input quality in current stack, goals, and adverse logs for the last 7 days.
Uncertainty and limits
- Evidence is limited on exact ranking behavior in uncommon comorbidity profiles.
- Evidence is limited on how rapidly ranking should shift under abrupt context changes.
Cross-site references
How this appears in Unfair
The engine output drives both what you see first and why the app requests extra logs before certain escalations.
Clinical safety note
If severe risk flags are present, no override should force continuation; clinician review takes precedence.