A digital inspection result can carry an accurate measurement and still be wrong for the job. Suppose a part is checked against an obsolete drawing revision: the instrument records the dimension faithfully, the software produces a polished report, and everyone receives the wrong conclusion with impressive speed. Gleb Tsipursky, a behavioral scientist and CEO of Disaster Avoidance Experts, argues in Metrology News that manufacturers adopting AI-assisted inspection should give every consequential measurement a version passport.
Connected metrology makes that failure mode more important, he writes. An October 26 Metrology News account of persistent identification describes how Quality Information Framework data can preserve the identity of a measured characteristic across product definitions, inspection plans and results. The technical promise is traceability; the management challenge is making sure each decision follows the correct version of the requirement, not merely that the data survives the transfer.
What travels with a measurement
The passport does not have to be a new software product. It can begin as a set of fields that travel alongside a result:
- the approved drawing or model revision;
- the characteristic identifier;
- the inspection-plan revision;
- the instrument or method;
- the result timestamp;
- the identity of the person or system that approved the disposition.
Each field answers a question a later decision-maker would otherwise have to guess: which requirement governed the check, which plan was executed, which method produced the number, when it was produced, and who accepted the outcome. Without those answers, a measurement is a value without terms.
Why AI makes the gap harder to see
AI can summarize an inspection report, point to an apparent deviation or recommend which cases need review. But a tool that receives results without version context may compare an observation with the wrong specification. The arithmetic can be correct while the conclusion is invalid, and the fluency of the summary makes the error harder to notice rather than easier.
A mismatch can cut in either direction. Measured against an older or looser revision, a genuine deviation may pass; measured against a tighter revision, a conforming part may be flagged for review. Either way the report looks normal, because nothing on its face records which requirement was applied.
Tsipursky places this in a wider context: the National Institute of Standards and Technology's 2026 smart-manufacturing roadmap highlights integration of heterogeneous data and trustworthy operation as continuing challenges for industrial AI. A version passport addresses a narrow slice of that problem by making the origin and applicable conditions of a measurement visible to the next decision-maker, whether that decision-maker is an engineer or a model. It is not offered as a complete answer to data integration.
A small test before a platform decision
Quality teams can try the idea before integrating a new platform. The suggested pilot draws on a small set of completed inspection records that deliberately includes one legitimate out-of-tolerance result, one conforming part and one example with mismatched revisions. An AI assistant is asked to summarize the evidence, identify the applicable revision and flag missing provenance, and its responses are compared with an experienced quality engineer's determination.
The evaluation should not stop at how often the assistant labels a part good or bad. Three failure modes matter more, and digital handoffs often conceal all of them:
- whether it catches a revision mismatch, a comparison error;
- whether it invents an unstated tolerance, a fabricated requirement;
- whether it can point to the approved source behind a conclusion, a provenance failure.
Raw label accuracy can look acceptable while every one of those errors is present, because a wrong verdict and a right verdict become indistinguishable once the version context has been dropped from the record.
The practical appeal of the idea is that it reads as a documentation and governance change rather than a systems replacement: the fields can be added to existing inspection records, and existing quality roles can review them. That is also why the author proposes testing on historical records instead of waiting for a new platform to be selected.
Ownership and the escalation route
Records will disagree, and the article recommends assigning clear ownership before they do. A quality engineer may determine whether a measurement is valid, design engineering confirms the governing definition, and manufacturing decides whether production can continue. The escalation route should be settled before the alert appears rather than negotiated during a shipment dispute.
The approver field keeps that accountability attached to the result as it moves. When a model recommends a disposition, the named role that accepted it remains visible instead of dissolving into the toolchain.
Not another AI-versus-human scoreboard
The proposal differs from the familiar demand for more comparisons between AI and human inspectors. Even perfect agreement between the two can reinforce the wrong decision if both were given an obsolete specification. The reference itself needs scrutiny, not only the agent reading it.
Automation can make measurement data travel farther and faster. A reliable chain of custody ensures that the drawing, the feature, the method and the result still belong together when the decision is made — the connection that the next generation of connected inspection should make impossible to ignore.