Glossary
Correlation Metadata
Updated February 22, 2026
Correlation metadata tracks timing overlap between doses, logs, and outcomes without treating overlap as proof of causation.
Why it matters
It helps you see likely windows where effects tend to cluster, while keeping direct cause claims out of routine recommendations.
What on-device correlation does and does not claim
Unfair can align dose timestamps, symptoms, and trend windows to identify temporal co-occurrence.
It does not infer that a correlated ingredient is the sole cause of an outcome.
Privacy-safe data and retention
- Retained locally: timing metadata, dose logs, and stack IDs.
- Optional cloud sync: enabled only with explicit user settings.
- Identity fields are separated from raw event timing where possible.
Default retention is typically short-term and controlled by account settings.
Known limitations
- Missing sensors and manual logging gaps reduce confidence.
- Delayed logging can produce false lead/lag patterns.
- Self-report bias can hide subtle effect shifts.
Cross-site references
Uncertainty
- Evidence is limited for causal claims from observational timestamp pairings.
- Evidence is limited on missing-data recovery for long gaps without active logging.
How this appears in Unfair
Unfair uses correlation metadata to build trend confidence bands and recency-aware ranking without automatic causal labeling.
Clinical safety note
If you notice severe or escalating symptoms, do not use correlation inference to self-diagnose; seek clinician guidance and share raw logs.