Repository path: workshop/experiments/E-20260804h-affect-halves/purposes.md · rendered 2026-09-09

The two purpose specifications — frozen before rendering

R11 §1 freeze. These clauses are committed before a word of either arm is written. They may not be added to, deleted or reworded once translating has begun (R11 §1); a clause discovered to be missing is recorded in the log as missing.

Source unit: Anton Chekhov, «Хамелеон» (1884), whole, 1,019 words, copy-text pss-1974 (materials/versions/pss-1974.txt).

R11's binding constraint, and why it matters more here than anywhere it has been used before. A purpose specification names a reader and a use, and may not be written in the vocabulary of wiki/goodness-senses.md. This experiment is about affect, whose definition reads "produces in its reader an experience comparable to what the source produces in its reader." Writing either half of that sentence into either spec would build the result. Neither spec below contains the words experience, effect, comparable, humour, funny, land, or the id or definitional wording of any sense on that page — the mechanical check is materials/check_specs.py and it is run before the freeze, with the S084 caveat carried: the check searches literal strings and cannot stop a paraphrase (R11 §The purpose specification), so the pre-run critic is asked for this specifically.

The check fired once, before the freeze, and the amendment is recorded here rather than made silently. STAGE clause 1 first read "a public evening of comic short stories". That is affect's subject matter handed to one arm and withheld from the other — an asymmetric prime on the exact property the run measures. It now reads "short prose". The check ran non-zero, the clause changed, and the run re-ran it to zero before anything was translated.

Purpose STAGE (arm A)

1. The reader is an audience: forty people in a room at a public evening of short prose, hearing this piece read aloud, once, straight through, in under nine minutes.
2. They have nothing in front of them — no page, no programme note, no glossary, no footnote, and no opportunity to stop, re-read or ask.
3. They know nothing about Russia beyond what any English-speaking adult knows, and nothing whatever about the 1880s, provincial police ranks, or the Russian language.
4. The performer will read exactly what is written and will not gloss, explain or ad-lib.
5. The evening's organiser has asked for a piece that holds a room that has been drinking, and will drop it from the programme if it does not.

Purpose SEMINAR (arm B)

1. The reader is a second-year undergraduate in a survey course on nineteenth-century Russian literature, reading on the page, alone, at their own pace, able to stop, re-read and go back.
2. They have no Russian at all and will never read this story in Russian.
3. They will be expected, in a seminar the following week, to say what the Russian text is doing at any point their tutor puts a finger on — what its words are up to, sentence by sentence — and this translation is the only access they have.
4. The tutor has forbidden footnotes, endnotes, glossaries and translator's introductions for this text: whatever the student is to know has to be in the running English.
5. The course reads this piece next to Chekhov's other Осколки sketches of 1884, so the student needs to be able to tell what is characteristic of this one.

The rule that governs both, and the deviation this run declares

Both arms are honest translations of the whole story, rendered straight through, once (R11 §4). Neither is a straw man, neither is a crib, and neither may add or drop propositional content. The purposes differ; the standard of workmanship does not.

Declared deviation from R11 §1. The copy-text was extracted and read once, in full, before these specs were written — the extraction is materials/extract.py and the read is on the record in this session's transcript. The strictest reading of §1 wants the specs frozen before the source is read at all. What that risks is precisely stated: the specs may have been chosen because this source visibly affords the divergence the experiment is about (speaking names, an untranslatable idiom, a servile plural, a cascade of diminutives). It is declared here, carried into the design's limits, and it is the reason the design registers a localization control (N1, N2) rather than resting on the forked sites alone.

Arm order (R11 §6). STAGE is written first, SEMINAR second. The second arm is written by a translator who has already rendered this source once, and the carryover is a known confound of this regime, not something this run measures away. Its direction is stated rather than hidden: the carryover advantage goes to SEMINAR, which the design predicts will lose the first of its two registered comparisons and win the second — conservative on one conjunct, anti-conservative on the other.