AX-R05 — COA purity diverges from independent median
The supplier's certificate claims a purity more than 1.5 pp above the median of independent tests on the same lot.
The predicate
Published literally. A rule you cannot read is an accusation you cannot check.
batch.independentN >= 2 && coa.purity - median(batch.independentPurity) > 1.5
Why this rule exists
The most consequential rule in the index, and the one that requires the most care to interpret.
The threshold of 1.5 pp is set well above the combined uncertainty of the two measurements. Inter-lab limits of agreement in this index are approximately ±0.5 pp for the well-behaved compounds and ±0.9 pp for the difficult ones; gradient-length differences add up to 0.5 pp on the acylated incretins; and within-lot vial-to-vial variation contributes a further 0.3 pp. Stacking those pessimistically gives roughly 1.2 pp of explainable difference. Beyond 1.5 pp, method and sampling differences are not a sufficient explanation.
The rule requires at least two independent tests of the same lot, because a single independent result disagreeing with a certificate is a disagreement, not a finding. With two or more, the median is a defensible estimate and the divergence is attributable.
Direction matters and the rule is deliberately one-sided. A certificate that understates purity is not flagged. This asymmetry is a value judgement the index makes explicitly: overstatement misleads a buyer, understatement does not.
Where this rule fires, the supplier is invited to respond and the response is published on the record page verbatim. Fourteen suppliers have responded at least once; the responses are collected in the corrections log.
Incidence
How often the rule fires, where, and when. A rule that never fires is not calibrated; a rule that fires on half the index is measuring the population rather than screening it.
This rule has not fired on any record in the index. That is either a well-behaved dataset or a badly-calibrated rule, and the index cannot tell which from here.