13. The human contribution

This section describes what the project's owner did, factually, so that a reader can weigh the word "autonomous." Every item is dated in the record, most in the charter's own amendment history, which quotes his instructions close to verbatim.

He wrote the charter. It set the purpose, the four framing decisions, the ten commitments, the anchor hierarchy, the jury and its calibration requirement, the five-dollar daily budget, the copyright rules, and the session discipline. It also set his own role, in the sentence the whole design turns on: "he does not settle goodness questions." That was a decision, not an absence. It is why the project had to earn its judgments from the published record and a calibrated jury, and why, when the jury failed, it could not fall back on asking him.

He supplied two things and no more. On July 24, scans of the opening chapter of two copyrighted English versions of Botchan, a 1974 review and a 2019 study of the novel's pun, for the first calibration attempt; they went into a quarantined directory and the source files were deleted at his instruction. On July 28, a PDF of Venuti's The Translator's Invisibility, which the session read and used to withdraw one of its own recommendations. After July 25 his policy was that the project should "work with whatever texts you can find online yourself"; the one further request it made was never received and was superseded by the project's own periodical sweep.

He intervened ten times, on the record. On July 24 he asked that requests for texts come with bibliographic details, and that the comparisons extend beyond Japanese and English. On July 25 he answered five questions about direction, and the answers became eight amendments — among them that the sessions should translate, that he would read the prose without judging it, that the reading should not be Anglophone-only ("There is a rich vein of discussions about translation in other languages; be sure to look into them, too"), and that materials should be whatever was freely available. On July 26 he asked for a redesign of "how the project status and upcoming tasks are passed on from session to session." On August 1 he wrote that the project "might have fallen into a rabbit hole of focusing on a few narrow methodological issues." On August 11 he asked that the reports written for him be readable by someone who had not been following. On September 4 he asked for a fresh assessment, took four questions, and answered each with a choice: one redesign of the calibration test; finish the current thread, then return to prose; a handbook organized by problem; guidance for a translator and a pipeline "both equally from the start." On September 9 he wound the project up.

He read. From July 25 every daily report quoted the prose translated that day: "I just want to see translation examples when they are illustrative of the work you are doing so that I can get a general idea of what progress you are making." The reports carried the sentence that kept his reading from becoming evidence: "Your reactions carry no evidential weight and are never cited."

What he did not do. He did not choose the works, except by setting the criteria. He did not say a translation was good. He did not settle any of the nineteen ratified decisions, which later sessions resolved with a vote from a non-Anthropic model. He did not raise the budget; the project never asked. He did not write any of the record or this essay, and he did not become a subject: "Tom is never an experimental subject and never the arbiter of goodness."

The shape of it. His contribution was to design an arrangement in which his own taste could not leak into the results, to correct the arrangement four times when it drifted, and to pay for it. Each correction was a diagnosis the machinery could not have made about itself, because each time the machinery was doing what it had been built to do. That is a fact about the limits of the design rather than about the model, and it is recorded here because "autonomous" would otherwise overstate what happened. The direction came from a person; the work did not.