Fiction Is a Column

This continues The Column a Sentence Goes In, which mapped the belief-revision rules of EPISTEME onto Sen's desk one by one. That entry ended on why none of the repositories is named in lore. This one is about the repositories that are not EPISTEME, and about the one column of the four that is not Sen's.

The equivalent forms

EPISTEME is not the only one of my repositories that keeps this discipline, and Sen's familiarity with its equivalent forms came from reading the others against it. Voting, a proportional allocation engine, is the count in Sen's three uniforms: exact arithmetic, conservation asserted at every phase boundary, a loud stop preferred to a silent error. careful-memory, a belief store for agents, has one decision record that says its list of evidence types deliberately contains no entry for a model's own output, so that a system can never gain confidence by reading its own memories back; that is Marginalia again, from the other side. And The Archipelago, a simulated civilisation built for virtual anthropology, labels every section of every publication it produces with an epistemic category — observation, metric, hypothesis, method and provenance, interpretation, limitation — and never generates the interpretation, leaving it as a labelled placeholder for a named researcher to fill. That became the sentence about how Sen reads a survey report, section by section, looking for the signature on any interpretation. An unsigned interpretation is a sentence that has wandered into the wrong column.

Two sortings, one shelf

The four kinds sort sentences. EPISTEME also sorts the beliefs those sentences become, by lifecycle state: unknown when no evidence stands on either side, disputed when evidence stands on both, undecided when what there is does not settle it, active when it does, and a fifth state, revised, which the code keeps in its vocabulary while nothing yet produces it, because removing a public word silently is itself the kind of change the methodology forbids. Sen's desk has both halves of this already, in two papers rather than one: the states with no evidence are the marked absences, kept in five kinds; the states with evidence are the gradings, kept in four. What I had not noticed until today is that the site itself runs a third sorting of the same shape, at the scale of whole pages. The four canon statuses — canon, valid-for-its-author, overlay, chartered — are the four kinds applied to documents instead of sentences: lore and the seasons are what the record has verified, the Codex is what a named source holds, an overlay is one readership's preferred reading and says so, and a chartered work is told by another hand and binds nothing. The Codex is where the column for told things lives, which is why Pinya belongs there and why the question of whether she may speculate was never really open.

Pinya

The fourth kind is the one that decided Ariel Pinya. It would have been easy to read a methodology this strict as a prohibition on fiction, and the opposite is true: fiction is a category in it, admitted on the same terms as the others, provided it is labelled. Pinya speculates and conjectures freely, and is right to, because she writes fiction and a story says on its cover what it is. The desk shelves her Called In stories without quarrel for exactly that reason. The same shelf holds an author who grounds nothing and says so and an archivist who grounds everything and says so, and the Archive is built so that neither is mistaken for the other. Her one refusal, that the wrongness itself is never explained, is not the methodology's; it is the record's shadow-shaped rule, which she arrived at by another road. The two rules meet at that point and do not overlap.

The same instrument

The lesson for the writing is the one I keep relearning in different clothes. A character who never speculates is not a temperament I described; it is a discipline I had already written down in code, with tests, in a repository that refuses plausible completion as a matter of policy. The story's instrument for keeping the columns apart is the Codex. The engineering's is a decision record with a test behind it. They are the same instrument, and the day one of them drifts, the other is where to look.