The problem
An engineering database held decades of technical drawings, identifying each piece of equipment its own way. The classification that says whether a component is export-controlled lived somewhere else entirely, tracking that same equipment under a different reference. Nothing connected the two records directly.
That database has also gone through several generations, and each one behaves differently: some link equipment straight to their drawings, others don’t store that link at all and it has to be reconstructed from lower-level technical files, and even the amount of descriptive information available for a given device varies widely from one generation to the next. A single method couldn’t cover all of them, and checking every drawing by hand, one by one, against every classified component was not realistic across tens of thousands of drawings and every aircraft programme.
What we did
Because no two generations of the database stored their data the same way, we built a distinct extraction approach for each one, then combined several independent ways of tracing a drawing back to its underlying equipment and cross-checked them against each other. Matches were treated as leads, not verdicts — every one of them went through manual review before being counted, and the rules were deliberately set to flag more rather than risk missing genuine cases.
In practice, that meant screening tens of thousands of technical drawings across every generation of the database, and confirming close to 800 drawings as carrying export-controlled content. The result is one way of tracking which drawings need protecting, applied consistently across every generation and handed over as the reference for how newly created drawings should be treated.
What made it hard
Automated matching is only useful if you know where to trust it and where not to. Some equipment couldn’t be traced to any programme with confidence, some references were incomplete, and a keyword that flags a genuine case can just as easily flag an unrelated one. Every one of those judgment calls was made in the direction of caution, because in export control an unreviewed gap is a much bigger risk than a false alarm.