This series has already delivered its bleakest structural finding: the apprenticeship systems that manufactured the industry’s credit judgment are fading away, distributed work has severed the informal transmission that survived them, and most of the industry’s professionals have never operated through a full cycle. That diagnosis assumed, at least, that the raw material of judgment formation still existed — the junior work itself, the thousands of statements spread, files assembled, and first drafts written through which pattern recognition was accidentally installed while something else was officially being accomplished.

Machine intelligence is now absorbing precisely that work, and the industry should be honest about what it is automating. Document intake, statement spreading, file assembly, first-draft memos — the applications with the clearest ROI in the business — are also, to the letter, the historical curriculum of the credit analyst. The efficiency case is unanswerable and this series has made it elsewhere. The formation case has simply not been examined: when the machine does the reps, where does the judgment come from? The industry is about to run that experiment on an entire generation, involuntarily, with the results due — as always — at the worst point of the next cycle.