Peak integration practice
Drop-line versus valley-to-valley, tangent skim, baseline placement, and the reporting threshold — where a purity figure is actually decided.
Chromatographic purity
Uncertainty budget for a purity figure
Nobody publishes one. Here is what it would contain, and what the combined standard uncertainty on a peptide purity result actually looks like.
Every purity figure in this index is printed to two decimal places and none of them is accurate to two decimal places. A number without an uncertainty is an assertion; the purpose of this page is to make the size of the uncertainty explicit, because it determines which comparisons on this site are meaningful and which are noise.
Contributions
| Source | Typical u (pp) | Notes |
|---|---|---|
| Integration — baseline placement | 0.08 | Larger on a drifting baseline; the dominant term on well-resolved peaks |
| Integration — valley/drop-line choice on an unresolved shoulder | 0.05–0.45 | The largest single term when a shoulder is present |
| Injection repeatability | 0.04 | Modern autosamplers; worse on manual injection |
| Detector response non-linearity at high concentration | 0.06 | Reduced by injecting on-scale |
| Relative response factor of impurities ≠ 1 | 0.10–0.30 | Almost never measured; assumed unity. Aromatic impurities respond differently at 214 nm |
| Peaks below the reporting threshold | 0.05 | Systematic and one-directional: purity is overstated |
| Sample-preparation concentration error | 0.03 | Small, because purity is a ratio |
| Column ageing over the sequence | 0.05 | Detected by bracketing system-suitability injections |
| Within-lot vial-to-vial variation | 0.28 (median) | Not a method term, but it limits what a single-vial result says about a lot |
Combination
Combined standard uncertainty (well-resolved main peak, no shoulder):
u_c = √(0.08² + 0.04² + 0.06² + 0.20² + 0.05² + 0.03² + 0.05²)
= √(0.0064 + 0.0016 + 0.0036 + 0.0400 + 0.0025 + 0.0009 + 0.0025)
= √0.0575
= 0.24 pp
Expanded uncertainty, k = 2 (≈95 % confidence):
U = 2 × 0.24 = 0.48 pp
Same calculation with an unresolved shoulder (u = 0.45 pp for the
integration-boundary term):
u_c = √(0.0575 + 0.2025 − 0.0064) = √0.2536 = 0.50 pp
U = 1.01 ppConsequences for this site
- A purity difference of 0.3 pp between two records is inside the expanded uncertainty of either one. The index does not treat such differences as findings and neither should a reader.
- The empirically measured inter-lab limits of agreement — ±0.44 pp for the best-agreeing pair, ±0.52 pp for the widest among the high-volume labs — sit close to the calculated expanded uncertainty. That is reassuring: the model and the observation agree.
- Rule AX-R05's threshold of 1.5 pp for COA divergence was set from this analysis. It is roughly three times the expanded uncertainty of a single well-behaved measurement, which is why the index is willing to call it a finding.
- The relative-response-factor term is the largest and the least addressed. Assuming every impurity responds identically to the parent at 214 nm is convenient and wrong, and correcting it would require an authentic standard of each impurity.
Normative references
Cited because they are the documents the acceptance criteria above are taken from. The index applies them as written and states every deviation.
- ISO/IEC Guide 98-3 (GUM) — Uncertainty of measurement
- Eurachem/CITAC CG 4 — Quantifying uncertainty in analytical measurement
- ICH Q2(R2) — Validation of analytical procedures
Related methods
Method validation to ICH Q2(R2)
The eight validation characteristics, what each one demonstrates, and which of them the reports in this index actually document.
Quality system
System suitability
The injections that establish a chromatographic system was fit to produce the result, and the parameters the index checks before accepting a purity figure.
Quality system