Repository path: workshop/experiments/E-20260808a-set-forms/materials/loci.md · rendered 2026-09-09

The locus classification for «Glogova régen» — rule first, then application

Frozen before the comparator was opened and before any model rendering was generated. The R04/R06 renderings of this chapter were committed first (b583be6 and this commit's parent); Worswick 1900 has not been read at the time of this commit.

Why this exists

RS-20260807c-two-hands §4 found that of the four loci the lead had labelled plain and which two independent hands nevertheless diverged at, three contained a set form the lead had not noticed when it labelled them — an archaic preterite (kérdék), a doubled idiom (Dehogy … dehogy), a proverb-shaped tautology (Ami parancs, parancs). That is an n = 3 post-hoc observation. This manifest turns it into a prospective classification on a fresh span.

The claim under test is not about the lead's introspection. It is the transportable version: set-formhood predicts where two independent hands part, even when nothing about the stretch looks difficult. A translator scanning a source for hard places looks for the MARK class below. If SET behaves like MARK rather than like PLAIN, the advance list has to include it.

The rule (written before the chapter was gone through)

Every locus is a short quotable stretch of the Hungarian and gets exactly one class.

• MARK — the Hungarian carries something English has no slot for, and the difficulty is visible in the meaning: an untranslatable pun or morphological chain, a culture-bound object or office, a code-switch, a name that means something. Admission test: name what English lacks.
• SET — the stretch is a fixed form of the language: archaic or obsolete morphology, a frozen idiom, a proverb, a fossilised politeness formula, a Latin tag in vernacular use, a reduplicative or a frequentative. Its propositional content is ordinary; what is fixed is the wording. Admission test: name the category of fixed form.
• PLAIN — neither. Ordinary narrative or descriptive Hungarian with a compositional meaning.

PLAIN is drawn mechanically, with no selection at all, so that the control class cannot be cherry-picked: the first finite clause of every paragraph that contains no MARK and no SET locus. Ten paragraphs qualify (¶4, ¶5, ¶6, ¶9, ¶10, ¶17, ¶20, ¶24, ¶29, ¶30) and all ten are used.

A locus is not admitted on the strength of the lead's English. Every admission test above is a statement about the Hungarian, and the class was assigned by going through the source in paragraph order.

MARK — 8 loci

SET — 19 loci

PLAIN — 10 loci, drawn mechanically

Controls, assigned now and instantiated after the renderings exist

• REPEAT on P02, S08, M4, P07 — the same rendering is shown on both sides. A seat that calls an identical pair different is confabulating difference, and the block is withheld.
• WRONG on M1, S14, P04, S04 — one side is replaced by a deliberately damaged rendering of that locus. Intents, written now: M1 replace the vehicle with a plain train (deletes the slang); S14 replace the simile with quite on my own (deletes the fixed form and the coyness); P04 replace wept for with forgot (a plain proposition flip); S04 replace the maxim with They ought to pay him more (deletes the proverb shape and reverses the speaker's position). A seat must call these different and name the damaged side as the one that drops something.

What is registered

Stated in full in design.md §Predictions and frozen there. In one line: SET is called different between two independent hands at a rate near MARK and well above PLAIN, on a pair the lead is not in.