Choose measurement equipment by the decision it must support, not the number of digits it displays. Compare the uncertainty of the complete method with the gap between a result and the relevant decision boundary. When the evidence cannot separate the possible outcomes, plan further measurement or report the conclusion as unresolved.
An uncertainty budget records the contributions to that uncertainty, their evidence, and how they combine. It helps a buyer ask whether a proposed measurement method will be useful near the decisions that matter. A promising instrument specification is only one input to that assessment.
Start with the gap your decision must resolve
Hypothetical worked scenario
Consider a sample shaft for a non-safety-critical product-development report. The customer has specified a diameter tolerance interval of 19.980 to 20.020 millimetres, written mm. These dimensions and all later report values are invented. They are neither calibration results nor operating limits.
- Quantity
- Diameter at the drawing's specified section and orientation.
- Decision
- Can the report support an in-tolerance statement under the agreed rule?
- Owner
- Quality lead, with customer requirements and metrology support.
- Unresolved path
- Withhold the statement and assign further review.
Metrology is the science of measurement. Its specialists help evaluate the method and uncertainty. The quantity being measured is called the measurand. Define its location, orientation, and relevant reference conditions. A diameter at one section does not establish every dimension of a shaft. A method that answers a different dimensional question cannot become suitable merely by reporting a smaller uncertainty.
The full tolerance width is 20.020 minus 19.980, or 0.040 mm. But a result close to either limit has less room for uncertainty. Buying equipment against the full width alone can miss the difficult reporting decisions near an edge.
Separate uncertainty from error and tolerance
Uncertainty describes the spread of values reasonably attributable to the measured quantity from the available evidence. It is not a known error to subtract, permission to ignore a fault, or the variation among different manufactured parts.
| Term | Meaning for this decision |
|---|---|
| Uncertainty | How widely supported values may be dispersed. |
| Error | Difference between a measured and reference value. |
| Resolution | Smallest change the indication can distinguish. |
| Tolerance | Permissible values specified for the component. |
| Margin | Distance from this result to the relevant boundary. |
JCGM 100:2008, the uncertainty guide, distinguishes uncertainty from error and defines standard uncertainty through a standard deviation. Correcting a recognized effect does not remove uncertainty about the correction.
A fine display increment can help resolve small changes, but it does not account for contact geometry, temperature, or the reference used. Repeatedly obtaining the same indication also leaves those other contributions to assess. Keep a known data or instrument fault in a separate investigation; an uncertainty budget is not a repair waiver.
Build a budget for the complete method
Use the shaft reporting question to request evidence from equipment suppliers and the measurement specialist. A contribution is an effect whose incomplete knowledge influences the result. The budget should show which contributions have been evaluated and which remain unknown.
| Contribution | Evidence to request | Review before combination |
|---|---|---|
| Instrument specification | Applicable range and stated limits. | Meaning of the quoted bound. |
| Reference and checks | Relevant reference report and comparison records. | Uncertainty of the reference and corrections. |
| Setup and contact | Method, alignment, and contact conditions. | Influence on the specified diameter. |
| Environment | Temperature evidence and applicable correction model. | Residual influence and shared effects. |
| Conversion and reporting | Scaling, units, and rounding method. | Contribution at the reported output. |
| Repeatability | Relevant repeated observations. | Single reading or mean being reported. |
For each row, attach the evidence reference, units, evaluation method, assumed distribution, standard uncertainty, and responsible reviewer. For this shaft result, express evaluated output contributions as standard uncertainties in mm, retaining the original input units and conversion. Record how an input change affects the reported diameter. The table is an evidence request, not a completed numerical budget or an assertion that every row is independent.
For procurement, ask the supplier to explain performance under the proposed method and environment. A specification obtained under different conditions cannot silently stand in for the whole workplace method. A buyer can compare unresolved evidence and support needs before commissioning a detailed evaluation.
Combine contributions only on a common basis
Standard uncertainty expresses uncertainty as a standard deviation. Expanded uncertainty uses a coverage factor to multiply the combined standard uncertainty for reporting. A quoted maximum error, a standard uncertainty, and an expanded uncertainty are different inputs. Do not enter their printed magnitudes into the same calculation unchanged.
- Interpret: Identify what each quoted number represents.
- Convert: Evaluate standard uncertainties on the result's unit basis.
- Relate: Account for model sensitivity and correlated effects.
- Combine: Apply the justified model and retain its assumptions.
- Report: State the coverage basis and conditions of use.
NIST Technical Note 1297, section 5, combines standard uncertainties with covariances where appropriate. Covariance describes how effects vary together.
For an approximately linear model with uncorrelated inputs, the root-sum-of-squares calculation squares the standard-uncertainty contributions in output units, adds them, and takes the square root. Correlated inputs require covariance terms. Sharing a temperature influence or a reference can make contributions related. Counting the same effect in two rows can also inflate a budget.
Use metrology review for the model and assumptions. No component combination is calculated here. A coverage factor of two does not automatically establish 95 percent coverage in every situation. Do not invent a probability from the notation on a report.
Compare the reported interval with the decision gap
Suppose the invented shaft report gives a diameter of 20.016 mm and expanded uncertainty U of 0.006 mm. For this arithmetic exercise, use the symmetric interval formed by subtracting and adding U. No calibration, coverage probability, or actual method performance is claimed.
- Reported interval
- 20.016 minus 0.006 to 20.016 plus 0.006 equals 20.010 to 20.022 mm.
- Upper margin
- 20.020 minus 20.016 equals 0.004 mm.
- Lower margin
- 20.016 minus 19.980 equals 0.036 mm; the smaller upper margin governs.
- Comparison
- U of 0.006 mm exceeds that margin by 0.002 mm.
- Boundary crossing
- 20.022 minus 20.020 equals 0.002 mm beyond the upper tolerance limit.
For this hypothetical example, assume the quality lead and customer agreed in advance: the reported interval must lie within both tolerance limits, including endpoints, before an in-tolerance statement is considered. This report does not meet that criterion. It does not follow that the shaft is definitely outside tolerance.
At this unchanged central value, the interval would need U no larger than 0.004 mm to stay within both limits under that criterion. That is a decision-specific constraint, not a universal equipment accuracy ratio. A different result or reporting rule changes the available margin.
U is a supplied quantity for this exercise, not a calculation from the worksheet. A real report without a stated method, coverage factor, and coverage basis leaves the evidence incomplete. JCGM 106:2012 on conformity assessment distinguishes tolerance limits from acceptance limits and connects acceptance to a stated decision rule. The example above is an editorial reporting exercise, not automatic conformity approval.
Improve the contribution that limits the decision
Ask what would actually reduce the uncertainty relevant to the shaft report. Purchasing a finer display may leave setup or environmental contributions unchanged. Repeating readings can help evaluate repeatability, but it cannot remove every shared effect.
| Budget finding | Question before spending |
|---|---|
| Reference evidence dominates | Would a more suitable reference improve this comparison? |
| Setup contribution dominates | Would an approved fixture or method change help? |
| Environment remains unknown | What characterization is needed before a capability claim? |
| Display is already adequate | Would extra digits change the actual decision? |
The worksheet does not prove which contribution dominates; that requires the evaluated budget. Ask for a comparison of plausible method changes, their supporting evidence, and the resulting uncertainty under consistent assumptions. Record unknown effects instead of using optimistic supplier values to complete a purchase justification.
Keep the tolerance fixed while comparing methods. Widening a customer's requirement is a separate agreement, not an uncertainty reduction. A guard band, an offset between an acceptance limit and a tolerance limit, also needs an agreed decision rule; it is not a universal allowance to apply after seeing a result.
Record an outcome when evidence is inconclusive
In the hypothetical workflow, report record DIM-05 links the drawing revision, measured quantity, result, uncertainty statement, budget, and decision rule. The quality record system retains the reviewed report. The quality lead owns its status; the measurement specialist owns unresolved method questions.
- Supported: Submit the evidence for authorized quality review.
- Inconclusive: Withhold the statement and assign further measurement.
- Changed method: Preserve the earlier result and review the new budget.
- Closed: Record the approved disposition and its scope.
For the invented report, leave the in-tolerance statement unresolved and ask the specialist whether a better-supported method is practical. Record who will review that answer. Quality acceptance remains with the authorized process and people.
Ordinary software can calculate interval endpoints and compare them with the approved rule. Optional AI can organize source documents and open budget questions for inspection. It cannot certify the method, fill unknown contributions, or approve the component.
Before buying equipment, bring one difficult reporting decision, its drawing, and the budget worksheet to the quality and measurement owners. Agree what evidence would change the decision and what happens if that evidence remains insufficient.