Glossary

Recommendation Confidence

Updated February 28, 2026

Recommendation confidence is a probability-style indicator of how reliable current ranking signals are.

Why it matters

Higher confidence does not mean guaranteed success, only stronger current signal consistency.

Confidence interpretation

Use three practical bands:

• low: early or sparse data, avoid major dose changes
• mid: usable signal with moderate uncertainty
• high: consistent adherence and symptom alignment

Confidence floor behavior

If confidence drops below a stable floor, Unfair slows optimization, favors safer core options, and requests additional journaling.

Symptom log influence

Fresh symptom consistency can raise confidence when aligned with prior trends.

Contradictory symptom reports can lower confidence quickly while preserving safety-first behavior.

Practical action step

Treat confidence as a coaching signal: if low, reduce stack complexity before increasing dosage.

Uncertainty and limits

• Evidence is limited on precise probability mapping for every outcome type.
• Evidence is limited on confidence behavior in users with irregular logging patterns.

Cross-site references

• Personalization weight
• Rationale snippet
• Feedback loop

How this appears in Unfair

You can spot confidence shifts in rationale text, recommendation rank movement, and safety-mode prompts after log cadence changes.

Clinical safety note

Low confidence never replaces clinical judgment; if symptoms are severe, pause optimization regardless of confidence label.