8 The objection🔗ℹ

Every percentage the analysis produces is the analyser inverting the generator. Both were written from the same account of how a printing house behaved, so a high score demonstrates that the simulation is self-consistent and nothing else. McKenzie’s finding from the Cambridge and Bowyer records — the only places the working of a hand-press shop can be checked against its output — is that the patterns are “of such an unpredictable complexity, even for such a small printing shop, that no amount of inference from what we think of as bibliographical evidence could ever have led to their reconstruction.”

His distinction is the useful part: the physical actions of setting, imposing, proofing and distributing type do not change from century to century, and those are what this program models. What varies, and what it assumes away, is “the amount of work done and the relations between those performing it … from day to day”. Of normality in that sense he says flatly: “it doesn’t exist.”

So the honest use of this machine is not to confirm the method but to break it. Only the failures are evidence; the successes are arithmetic.