A digital inspection result can contain an accurate measurement and still be wrong for the job. If a part is checked against an obsolete drawing revision, the instrument records the dimension faithfully, the software produces a polished report, and the wrong conclusion travels quickly. Writing in Metrology News, behavioral scientist Gleb Tsipursky argues that manufacturers adopting AI-assisted inspection should give every consequential measurement a "version passport."
The passport, as he describes it, need not be a new software product. It can begin as a set of fields that travel with a measurement: the approved drawing or model revision, the characteristic identifier, the inspection-plan revision, the instrument or method, the result timestamp, and the identity of the person or system that approved the disposition.
Why version context matters for AI
Tsipursky's concern is that AI tools 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. In that case, he writes, the arithmetic can be correct while the conclusion is invalid.
The argument builds on connected metrology. The article cites an October 26 Metrology News account of persistent identification, which described 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, Tsipursky says, is ensuring that each decision follows the correct version of the requirement.
He also points to the National Institute of Standards and Technology's 2026 smart-manufacturing roadmap, which he says highlights the integration of heterogeneous data and trustworthy operation as continuing challenges for industrial AI. A version passport, in his framing, targets a specific part of that problem by making the origin and applicable conditions of a measurement visible to the next decision-maker.
A small test before a platform
Quality teams can try the idea before integrating a new platform, the article suggests. It proposes selecting a small set of completed inspection records that include one legitimate out-of-tolerance result, one conforming part and one example with mismatched revisions, then asking an AI assistant to summarize the evidence. The assistant should be required to identify the applicable revision and flag missing provenance, with its responses compared against an experienced quality engineer's determination.
Rather than asking only how often the assistant labels a part good or bad, Tsipursky recommends recording whether it catches a revision mismatch, whether it invents an unstated tolerance, and whether it can point to the approved source behind a conclusion. Those, he argues, are the errors that digital handoffs often conceal.
Ownership and escalation
When records disagree, the article calls for clear ownership: a quality engineer may determine whether a measurement is valid, design engineering confirms the governing definition, and manufacturing determines whether production can continue. The escalation route, Tsipursky writes, should be settled before an alert appears rather than negotiated during a shipment dispute.
He distinguishes his argument from the familiar demand for more AI-versus-human comparisons. Even perfect agreement between human and automated inspectors, he writes, can reinforce the wrong decision if both were given an obsolete specification, so the reference itself needs scrutiny. Automation can make measurement data travel farther and faster; a reliable chain of custody, he concludes, keeps the drawing, feature, method and result together when the decision is made.
Tsipursky is a behavioral scientist, chief executive of Disaster Avoidance Experts and author of The Psychology of AI Adoption at Work: From Resistance to Results (Georgetown University Press, 2026). The piece is an argument for provenance discipline in connected inspection rather than a report of a deployed system.