An inspection result can be numerically accurate and still be the wrong answer for the job. If a part is checked against an obsolete drawing revision, the instrument may record the dimension faithfully, the software may produce a polished report, and the organization may reach the wrong conclusion faster than before. Gleb Tsipursky, a behavioral scientist and CEO of Disaster Avoidance Experts, argues that AI-assisted inspection needs a version passport for every consequential measurement: a compact set of fields that travels with the result and shows which requirement governed it.
Precision without provenance
The failure mode is not a broken instrument or a calculation error. It is a mismatch between the measurement and the specification it is compared with. Connected metrology makes that risk more consequential because results, inspection plans and product definitions can move across systems and teams. Persistent identification efforts, such as those using Quality Information Framework data, aim to preserve the identity of a measured characteristic across product definitions, inspection plans and results. The technical promise is traceability. The management challenge is ensuring that each decision follows the correct version of the requirement.
Tsipursky’s proposed passport is not necessarily a new software product. It can begin as a structured set of fields attached to a measurement: the approved drawing or model revision, the characteristic identifier, the inspection-plan revision, the instrument or method used, the result timestamp, and the identity of the person or system that approved the disposition. These fields do not make the measurement more precise. They make its applicable conditions visible to the next decision-maker.
Why AI raises the stakes
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 remain correct while the conclusion is invalid. Tsipursky connects this risk to broader industrial AI concerns: the National Institute of Standards and Technology’s 2026 smart-manufacturing roadmap highlights integration of heterogeneous data and trustworthy operation as continuing challenges. A version passport addresses a specific part of that problem by keeping the origin and conditions of a measurement attached to the result.
The risk is easy to underestimate because polished outputs can look authoritative. A summary can be clear, a deviation flag can be prominent, and a recommendation can arrive quickly. None of those qualities proves that the underlying reference was current. In a digital handoff, the missing revision may be invisible precisely because the report looks complete.
A small test before platform integration
Quality teams can test the idea before committing to a new platform. Tsipursky recommends selecting a small set of completed inspection records: one legitimate out-of-tolerance result, one conforming part, and one example with mismatched revisions. An AI assistant can then be asked to summarize the evidence, identify the applicable revision and flag missing provenance. Its responses should be compared with an experienced quality engineer’s determination.
The evaluation should not stop at how often the assistant labels a part good or bad. More revealing questions are 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 are the errors that digital handoffs often conceal, because they concern the reference behind the result rather than the arithmetic in front of it.
Ownership and escalation
When records disagree, ownership must be clear. A quality engineer may determine whether a measurement is valid, design engineering may confirm the governing definition, and manufacturing may determine whether production can continue. The escalation route should be settled before an alert appears, rather than negotiated during a shipment dispute. This is a management requirement that sits alongside the metadata design; a passport is useful only if someone is responsible for acting when its fields conflict.
The proposed test is deliberately narrow because it focuses on provenance rather than general model performance. A quality team does not need a large benchmark to learn whether an assistant can distinguish a current revision from an obsolete one. A few records with known answers can expose whether the assistant checks the governing source or merely repeats the language of the report. The results can then inform whether the organization needs better metadata, clearer approval rules, or a different review process. The point is not to certify an AI tool in isolation, but to test whether the surrounding record structure gives the tool a fair basis for judgment.
The reference itself needs scrutiny
The argument differs from the familiar demand for more comparisons between AI and human inspectors. Even perfect agreement between human and automated inspectors can reinforce the wrong decision if both were given an obsolete specification. The reference itself needs scrutiny. Automation can make measurement data travel farther and faster, but a reliable chain of custody must ensure that the drawing, feature, method and result still belong together when the decision is made. For the next generation of connected inspection, the connection between a measurement and its governing version should be impossible to ignore.
A version passport is therefore less a dashboard than a chain-of-custody convention. It does not replace engineering judgment or quality inspection. It gives those judgments a documented reference point when a measurement moves from an instrument to a report, from a report to a review, and from a review to a production decision.