A DGA reference standard uncertainty sets the smallest difference a monitor test can honestly resolve. If the reference is uncertain by 8% and the acceptance window is ±5%, the test cannot fail an instrument on evidence; it can only describe a disagreement whose cause is unknown. The half-window rule exists to prevent that, and it is the first thing to check before debating a monitor’s accuracy.
The half-window rule
The rule is simple: the reference uncertainty should be no more than half the judgement window. To judge a monitor at ±5%, the reference needs U ≤ 2.5%; for ±8%, U ≤ 4%; for ±3%, U ≤ 1.5%. The logic is in the GUM and in ILAC-G8, which treats acceptance decisions as a guard-banded comparison rather than a bare threshold. When the reference uncertainty is comparable to the window, the pass/fail line becomes a coin toss weighted by someone else’s error.
This is not a theoretical nicety. At U_ref = 5% and a ±5% window, a monitor that is perfectly accurate will fail a meaningful share of tests, and a monitor that is 5% off will pass often enough to survive. Once DGA reference standard uncertainty reaches the size of the window, the test stops measuring the instrument and starts measuring the reference.
Where the budget actually goes
A 5 L gravimetric batch prepared to IEC 60567 has a budget that looks like this:
| Contribution | Relative standard uncertainty | Comment |
|---|---|---|
| Balance | 0.05% | Kilogram-level calibration; not the problem |
| Density at 20 °C | 0.12% | Densitometer, temperature-controlled |
| Syringe volume | 0.60% | Includes pressure and temperature correction |
| Gas certificate | 1.20% | Certified mixture, usually the largest declared term |
| Transfer and dissolution | 2.00% | Not measured in most procedures |
| Homogeneity | 1.50% | Batch-to-aliquot variation |
| Storage drift | 0.80% | Adsorption, permeation, moisture exchange |
Combined in quadrature, these terms land near 3% — which supports a ±5% window with little margin and a ±3% window not at all. The two largest terms are also the two least often measured: transfer loss and homogeneity. Improving the balance or buying a better densitometer does nothing for the total while those remain estimates.
Why two chromatographs cannot referee a ±5% claim
Cross-checking two gas chromatographs is a useful consistency test. It is not a reference. Two instruments calibrated in the same laboratory against the same standards can agree to 5% while sharing a bias of the same size; the comparison measures agreement, not truth. Published laboratory round-robins put the DGA acceptance target near ±15% (IEC 60567:2023), and CIGRE TB 409 found average laboratory accuracy around 12% — a number that includes extraction and analysis, and excludes sampling.
A chromatograph can still serve as a transfer standard. It becomes legitimate when it is anchored to a gravimetric nominal value and a certified gas fraction, and when its own uncertainty is declared. That is a different role from using two chromatographs to define the reference by mutual agreement.
The low-concentration wall
At 1 ppm in a 5 L batch there is about 5 µL of gas in total, while the dead volumes of ordinary apparatus are measured in millilitres. Near the detection limit, GC repeatability of 5–10% is normal. If the homogeneity gate is set at half the judgement window and each check is a single measurement, the batch will fail on the chromatograph’s own noise before the monitor is connected.
The way through is not a better argument; it is a different decision rule. Near the detection limit, use an absolute tolerance rather than a pure percentage. In the measuring range, use the relative window. For homogeneity, use at least three replicate determinations, not one. Without those three changes, a low-concentration disagreement produces “no decision”, not a verdict on the instrument.
Does a lower-uncertainty reference guarantee a better monitor?
No. It guarantees that a disagreement can be attributed. A monitor with a real 4% bias will still fail a ±3% test no matter how good the reference is. What a tight reference removes is the excuse: after the reference is fixed, a repeatable offset belongs to the instrument and has to be corrected by calibration.
Can a certified gas cylinder replace a gas-in-oil standard?
Only for the detector and the calibration curve. A cylinder verifies the chromatograph’s response; it does not exercise extraction, transfer or the oil matrix. The full chain needs oil with gas dissolved in it, which is why a gas-in-oil reference standard exists in the first place.
PAS DGA publishes its monitor specifications with the measurement chain in mind; the DGM-900 is specified for seven gases plus moisture, and the DGA fundamentals hub explains how laboratory and online values should be compared. If you are writing an acceptance test or arbitrating a laboratory-versus-monitor disagreement, contact our team. The assumptions behind the reference itself are covered in the third article in this series.
Sources
JCGM 100:2008 — Guide to the Expression of Uncertainty in Measurement (GUM). ILAC-G8:09/2019 — Guidelines on decision rules and statements of conformity. IEC 60567:2023 — oil sampling and analysis guidance; the ±15% figure is a round-robin acceptance target, not a chromatograph specification. CIGRE Technical Brochure 409, Report on Gas-filled Monitors. ISO 6142-1:2015 — gravimetric preparation of calibration gas mixtures.