Handle repeated tests for the same sample

Repetition can be meaningful

Preserve them before consolidation so an apparent duplicate does not erase evidence the reviewer needs.

Our public material-register example shows a synthetic report beside downloadable output. Use it to inspect how values and source references travel together. The self-service ParseMill trial is for invoice PDFs; supplier test reports and engineering review need a separate fit discussion.

Extend the observation model with reported replicate, run or condition identifiers when available. If the source supplies no such label, retain a source-row locator and an explicit repetition status rather than inventing a scientific replicate number.

Use a repetition-aware row model

These synthetic rows contain distinct reported contexts and one unresolved repetition. The two equal 2.40 values are not automatically duplicate records. Keep their separate locations until the source or issuer establishes whether they are independent observations or repeated copies of one result.

W3C: RDF Data Cube Vocabulary models observations through dimensions and qualifying attributes. Replicate or condition context can be part of the identity needed to distinguish observations.

Review the repeated rows

1. Capture every reported row with its source location, raw result, unit and method. Preserve explicit replicate or retest wording.
2. Check whether repetition reflects another test, a changed condition, a revised result or a duplicate source copy. Record the evidence for the classification.
3. Create a stable internal observation key without overwriting the reported sample identifier. Keep unresolved repeated rows visible.
4. If analysis requires an average or selected result, obtain a documented rule appropriate to the intended task and create a separate derived view. Retain all raw observations.

Microsoft: Table.Distinct removes rows according to chosen comparison columns. A sample-test-result criterion can erase legitimate repeats, so do not use it as an unreviewed cleanup rule.

Avoid invented summaries

A mean may combine a retest with original replicates even when the source treats them differently. Selecting the last row may mistake page order for an authorized replacement. Neither operation is a neutral transcription choice. Keep summary decisions outside the raw extraction layer.

If the source itself reports an average, store it as a source-reported summary with its stated population. Do not replace the underlying observations unless the register’s declared scope intentionally contains summaries only.

Verify a faithful register

For the fixture, all four rows remain identifiable, the first two retain their replicate labels, the third remains a retest and the fourth remains unresolved. No automatic average or deduplication changes the evidence. The recipient can then apply an appropriate analytical rule, with its assumptions documented separately from the source observations.

Use repeated tests and retests as acceptance cases when discussing a supplier-report pilot. Agree the required observation structure and inspect whether each source row remains distinct. Any averaging or technical interpretation belongs in the separate reviewed analysis.