# Consecutive outliers

> two same-side 2SD in a row = bias, random 2.3%

Source: https://geodb.io/guide/topic/qaqc_westgard_consecutive_rules/ · Category: QAQC
A single outlier is noise; a **pattern** of them is bias. Trending rules
formalize this, and OREAS endorses them by name on its certificates:
*"Westgard Rules extend the basics of single-rule QC monitoring."*

The core set:

| Rule | Meaning |
|---|---|
| **1-3s** | one result beyond ±3SD → reject |
| **2-2s** | two consecutive beyond ±2SD, same side → systematic error |
| **R-4s** | range between two controls exceeds 4SD |
| **4-1s** | four consecutive beyond ±1SD, same side |
| **10-x** | ten consecutive on the same side of the mean |

The mining-specific one is **2-2s** — Bishi (after Barry Smee): results
rejected when *"two consecutive batches show two CRM results beyond the
two-standard-deviation limit on the same side of the mean."* Under
normality that pattern occurs randomly only **2.3%** of the time, which is
exactly why it detects bias rather than punishing noise.

Whether the long-run trigger is 7, 9 or 10 consecutive same-side results
depends on the rule-set you adopt (Westgard vs Nelson) — a genuine
convention choice. Full multi-rule stacks are not standard in mining;
2-2s is the one with mining-specific backing.

## In geoDB

geoDB does not evaluate trending rules — every verdict is per-result. The
**control charts** on the QC pages plot results in sequence against the
±2SD/±3SD bands, which makes same-side runs easy to spot by eye; for a
formal check, ask Petra to run one over your data.