# No certified SD

> own-lab repeats or ±10%±2DL — never batch scatter

Source: https://geodb.io/guide/topic/qaqc_missing_certified_sd/ · Category: QAQC
Sooner or later a standard has no usable certified SD — the whole
certificate reports only expanded uncertainty, one analyte reads `nr`, or
a row sits below detection (`<5 IND IND` — not a certified value at all;
don't QC-evaluate it).

## The sanctioned fallbacks, in order

1. **Derive the SD from your own lab's repeats.** The unanimous vendor
   answer — OREAS: *"SD's should be calculated directly from your own
   laboratory's process."* AMIS wants a minimum of 20 analyses.
2. **OREAS's low-concentration rule:** acceptance at **Certified Value
   ±10% ±2DL** — OREAS's published formula for near-detection grades
   (not CDN's; CDN certificates carry no such rule).
3. **The 5% window** — a standing OREAS certificate column, calculated
   from the certified value.

⛔ **Rules 2 and 3 produce acceptance WINDOWS, not standard deviations.**
Never enter a ±10%±2DL (or 5%) result in an SD field: the engine applies
its 2SD/3SD ladder *on top of* whatever sits there, so a window entered
as a 1SD silently widens your gates 2–3×. A window is the pass/fail test
itself; only a real 1SD (rule 1 — your own lab's repeats) belongs in the
SD field.

Explicitly **not** sanctioned: substituting the confidence interval,
expanded uncertainty, or tolerance limits.

## The trap with no support anywhere

**Substituting the batch's own scatter** — computing an SD from the very
results you are judging — has no support in any source, and it is
mechanically self-defeating: the calculated SD **grows with the very error
being detected**. A uniformly wrong batch produces tight scatter → small
Z → **pass**. This is not hypothetical; it has produced real
massively-biased "Pass" verdicts in production databases (including,
historically, this one — see below).

## In geoDB

For most surfaces geoDB's ladder is: certified 1SD → an SD derived from
the certificate's asymmetric 2SD bounds → a ±10%/±15% bias rule. On the
certificate-detail page, however, a **batch-scatter SD** currently sits in
the ladder for groups with no certified SD — a known, documented gap
scheduled for remediation. So **leaving the SD field blank is not a safe
"bias-only" fallback**: on the page that judges certificates, a blank SD
means your standards get judged against the batch's own scatter — the
very thing forbidden above. A hard bias ceiling (50% by default,
configurable per project) backstops it: no row can "pass" with a huge
bias no matter how the SD came out. If a standard of yours has no
certified SD, the durable fix is to add a proper 1SD to the standard's
record — from your lab's repeats, per the fallback ladder above.