When Buyer and Supplier Measurements Disagree

Conflicting inspection reports do not prove that one party measured badly. They may concern different physical items, drawing revisions, locations, datum frames, restraints, fitting algorithms or environmental conditions. The first job is to preserve both records and determine whether the numbers describe the same measurand before anyone argues about which value is correct.

The existing explanation of machining tolerances for track rollers provides the product context. This article addresses only the joint comparison after results disagree. It does not appoint a master instrument, calculate product conformity or assign blame.

Preserve both reports and the disputed part identity

Keep the original supplier and buyer reports unaltered, including attachments, raw exports and revision metadata. Do not retype values into a shared spreadsheet and discard the sources. A comparison record can quote them, but it should link every entry back to the original file and identify any later corrected version.

Confirm part, serial, heat or lot identity and determine whether both reports address the same physical item. Identify the feature and characteristic, drawing or specification revision, units, nominal and tolerance. If either report uses an internal characteristic code, map it to the controlled drawing rather than assuming similar labels mean the same feature.

Record measurement dates and the part's condition before and after each inspection. Handling, cleaning, coating, temperature exposure, assembly, disassembly or rework can change the condition presented to the second party. Preserve photographs and a location map where available. A part that changed cannot serve as an unchanged comparison artifact.

Create a custody and condition timeline when the part moved between organizations. Record packaging, receipt observations, storage, identification marks and every authorized preparation step. If the feature was cleaned differently or allowed to stabilize for a different time, keep that fact visible. The timeline does not prove that a change caused the disagreement, but it prevents the joint team from assuming an identical physical state without evidence.

Name who measured, the report status and whether each document is preliminary, corrected or approved. Apply quarantine or another material status only through the authorized nonconformance procedure. This evidence-preservation step does not accept or reject the part or lot.

If the drawing revisions differ or the reports cannot be tied to the same part and feature, classify them as non-comparable until the identity gap is resolved. Editing one report to match the other's labels would hide the most useful finding.

Confirm that both sides measured the same quantity

Define the feature precisely and write the measurand in operational terms. Include the datum reference frame and datum precedence, free-state or restrained condition, measurement location and section, and whether the coating or surface treatment is included. A nominal feature name such as “bore diameter” may conceal several different quantities.

Distinguish a point or local two-point size from a feature derived from multiple sampled points. A fitted circle, cylinder or plane depends on sampling and an evaluation algorithm. Two calibrated systems can return different values when one reports a two-point bore diameter and the other fits a circle. The question is not which machine is more sophisticated; it is whether both values represent the same required measurand.

Record reference temperature and compensation, the reporting resolution and any profile filter. Confirm that both parties measured for the same purpose. A setup check, process-control value and final conformance result may use related data but different definitions or decision rules.

A NIST explanation of measurement-method disagreement describes how techniques can interact differently with a sample or effectively measure different quantities. Its lesson is to understand each measurement model and error source, not to presume that either result is wrong.

NIST Technical Note 1297 Appendix D4 also explains that a measurand defined by a standard method should identify that method and that uncertainty depends partly on implementation. If the two reports define different measurands, do not average them or treat similar units as proof of equivalence.

Compare how each result was produced

Create parallel process maps. Record each instrument and ID, current calibration or verification status, range and resolution. Calibration is one input. NIST's calibration policies note that uncertainty also arises from the instrument, method, operator and physical environment, and that a report does not guarantee future performance.

Map the fixture, support and clamping force, contact or non-contact principle, probe qualification and stylus. Record alignment, datum simulation and any local re-alignment. Capture the number and distribution of points, feature-fitting method, filter and outlier rules. A CMM result is not automatically the true value; its model and setup remain part of the evidence.

Identify the operator, work instruction, software and version. Record environment, stabilization and part-temperature evidence, repeat measurements and every applied correction. For each uncertainty statement, retain the value, coverage factor or stated coverage information and the measurand and conditions it belongs to. Do not assume uncertainty statements from two reports are comparable because both use a plus-or-minus format.

Ask what contributors and coverage each uncertainty statement includes. One report may address a calibrated instrument under controlled conditions, while another includes fixture, temperature or repeatability effects for the executed process. This does not make either estimate superior. It shows why the two values must be read with their models, units and scope before they are used in a joint interpretation.

NIST's paper on the calibration concept emphasizes measurand, conditions of validity, uncertainty and subsequent use. Those limits explain why two in-date instruments can disagree in service. They do not prove a calibration failure or identify root cause.

Log deviations from each procedure. The first material process difference may explain a testable hypothesis, but it remains a hypothesis until a controlled comparison isolates it.

Run a controlled joint comparison

Hold shared conditions constant

Name a comparison owner and observers from both parties. Use the same identified part where it remains stable, or a suitable reference artifact when the physical question and protocol justify it. Agree the measurand, drawing requirement, units and reference conditions before measuring.

Freeze the protocol and data sheet before the first joint run. Where practical, conceal prior numeric results from the operator until the observation is recorded, while keeping the necessary setup information available. Use independent witnesses or controlled electronic capture when the dispute warrants it. The purpose is to reduce memory and transcription effects, not to turn the exercise into an adversarial test.

Control handling and stabilization and define stop conditions if the part changes. Agree fixture and datum methods, sampling and fitting rules, instrument checks, measurement order and replication. Preserve all raw data and uncertainty fields. Record deviations as they occur instead of correcting the protocol retrospectively.

NIST proficiency-testing material provides a useful protocol model by identifying artifact handling, environmental and reference conditions, measurement instructions, units, timing, uncertainty components and method deviations. It should inform a joint shop-floor comparison without turning that exercise into an accreditation proficiency test.

Test one material difference at a time

After a shared-condition run, change one suspected material factor where feasible: fixture, datum realization, sampling pattern, algorithm, compensation or another documented difference. Hold other conditions constant and replicate enough to observe the effect under the agreed plan. Changing several factors together can show that the process matters without proving which factor caused the difference.

Do not calculate the arithmetic average of buyer and supplier results and call it a reference value. NIST research on reference values in dimensional intercomparisons explains that no single approach is best in all circumstances and that uncertainty, distribution, outlier and correlation assumptions matter. Neither party's instrument becomes a master merely by agreement.

Protocol state	Identity and measurand	Controlled comparison	Next action
Normal: comparable	The same stable part, feature, revision, measurand, units and purpose are confirmed.	Shared conditions, method fields, replication, raw data, uncertainty and deviations are complete.	Record observed agreement or difference and its limited scope without declaring either report true.
Missing: protocol incomplete	Part identity, datum, location, restraint, method-defined measurand or requirement is absent.	Environment, fixture, algorithm, uncertainty, raw data or another required field is missing.	Request the missing evidence before drawing a correlation conclusion.
Conflict: comparison cannot isolate the issue	The reports define different quantities, the physical part changed, or one pass/fail label is proposed as the reference.	Several factors changed together, methods cannot be aligned, or original data would be overwritten.	Preserve both records, stop the comparison and route redesign or technical escalation to the joint owner.

The protocol can establish comparability or expose why it is unavailable. It does not require numerical agreement and does not create a product disposition.

Record what the comparison resolved

Report the original difference, repeat results, conditions held constant, factor changed and observed effect. Place uncertainty information beside each result. State whether the methods became comparable and distinguish an established observation from a suspected cause.

For example, a controlled run may show that a fixture or fitting algorithm shifts reported values under the tested condition. That does not prove a universal root cause, recalibrate either system or decide whether the product conforms. Scope the conclusion to the part range, feature, methods and conditions actually studied.

List remaining disagreement, open evidence, owners and due dates. Technical escalation may involve metrology, design, supplier quality or an independent laboratory under an approved plan. Keep product-impact assessment separate from measurement correlation.

JCGM 106 addresses the use of uncertainty in conformity assessment, while the ILAC guidance index lists decision-rule guidance. Neither supplies the contract-specific rule for this dispute. The drawing, contract and authorized quality process still control acceptance or rejection.

Retain original and corrected reports, the joint protocol, raw data, deviations and approvals. Never overwrite the earlier values or convert comparison into lot release. This article does not establish a measurement capability for KTSU. Its useful result is a neutral technical record of what matched, what differed, what changed under control and which decisions remain open.