Why AI in engineering must cite its methods

Engineering has always had a rule for trusting calculation: show the method. A bearing capacity is not a number; it is a number derived by a named correlation, under stated assumptions, from identified data. Any colleague can retrace it. Professional liability, peer review and the entire culture of checking rest on that retraceability.

Then AI arrived, producing answers without workings, and the industry split. Some rejected it outright. Some used it quietly and hoped. Both responses accept a false premise: that AI outputs cannot carry workings. They can, and they must.

The citation standard

When Strats3D estimates bearing capacity from SPT data, the correlation is named on the output — the same published relationships an engineer would cite by hand, applied to identified boreholes, with the screening-level status stated. When it scores settlement risk or flags an SPT reversal, the flag links to the data that raised it. The engineer checks the AI the way they would check a graduate: by reading the workings, not by trusting the tone.

What this makes possible

• Outputs can enter reports, because they can survive review.
• Disagreement becomes productive: you can dispute a correlation choice, not a vibe.
• Junior engineers learn from the tool, because the tool shows its reasoning.
• The liability chain stays intact: the engineer signs, informed rather than replaced.

AI that cannot cite its method is asking engineering to abandon its oldest quality system. AI that can cite its method strengthens that system. The choice between them is not close.