Required CSV columns
PlateSequenceName
BatchName
WellId
ChemistryId
Concentration
Optional: CompletionState, Units, LocalCompletionTime, Errors
How are outliers detected?
Three independent statistical tests run on each replicate group (≥ 3 replicates required). A replicate is recommended for discard when it accumulates enough flags (configurable via Consensus signals).
- Modified Z-Score
- Uses the median and MAD — robust to non-normal distributions. Flags a replicate if |0.6745 × (value − median) / MAD| exceeds the threshold.
- IQR Fence
- Flags values outside Q1 − multiplier × IQR or Q3 + multiplier × IQR.
- Leave-One-Out CV
- Tests whether removing each replicate reduces group CV below threshold. Only the best-improvement replicate is a candidate.
The cleaned mean/SD is computed after the recommended discard. Groups are REVIEW when no single outlier is found but CV still exceeds threshold.