Source and Versioning of Measurement Data

Why it matters

The source and version of measurement data are the link that explains which work a number represents. A cell that says 8.00 s is not measurement evidence by itself. You must know from which record, which element description and which method this duration was obtained. The recording time, the observation order and the method version also affect later interpretations. Even if the average of a report is calculated correctly, the merging of wrong scopes can invalidate the result. The data chain serves to verify what was calculated before verifying the arithmetic. Approving a duration of unknown source because it looks like the others hides the missing link of this chain.

Completeness and comparability are different. Two records with all fields filled may belong to different start or end definitions. If one runs until the tool is put down and the other until the good part is placed in the box, different work has been measured even under the same element name. The version information makes this change visible; writing a version number alone is not enough. For each version it is explained which work sequence, tool, layout and acceptance condition it belongs to. Equalizing units does not equalize scopes. Averaging all values converted to seconds is therefore not a reliable method.

The number of missing fields and the number of missing records are also different. If both the source and the time of a measurement are unknown, two fields are missing, but there is only one record awaiting review. Adding up field percentages to produce a complete-record rate overlooks the overlap of the gaps. First a status is determined for each record, then the totals are found. The audit table must show which records were set aside and for what reason. Deleting excluded records from the result weakens the data chain. Keeping a gap under an open status and recording its resolution is worth more than delivering a result table that only looks clean.

Making a correction is not an error; the problem is writing the correction over the old record without leaving a trace. The raw measurement must be preserved, and the new value must be tied to the reason and the version that produced it. It must also be possible to know which record set an old report used. This way it can be examined whether a changed result came from a production improvement, from the element definition or from a data correction. The goal of this lesson is not to build a complex information system. It is to build a reproducible decision path with a few clear identifiers, status and version fields. Collecting unnecessary personal data is not a precondition of this path.

Another benefit of the data chain is that it makes visible how the set of records that feed a result changes over time. When new measurements are added the average may change; this change does not prove that the method has improved. The old series may have had a higher share of another product, or the new observations may have been collected under different shift conditions. The delivered result is therefore defined together with the record identifiers and the selection rule. Reaching the same number is not enough; the same scope must be re-selectable with the same rule. Repeated observations receiving separate identifiers also does not create independent information. The difference between a duplicate record and a true repeat must be understood from the source link. During the audit, do not count similar durations as duplicates just because they are similar; different cycles can finish in the same time. For the decision, examine the source, time and element links together. The reason a record is excluded should not be that its duration is long or short, but that it does not meet the acceptance condition written in advance. Removing a suitable but slow cycle without showing it is not data cleaning. Explaining why you classified it and showing its effect separately protects the reliability of the measurement result. In this way the data arrangement stops being merely an archiving job and becomes the basis of the comparison.

Formula