Repository path: workshop/regimes/R11-declared-purpose.md · rendered 2026-09-09

R11 — declared-purpose target

v1.0 (frozen 2026-08-01, S084.) An R06 (lead, source-only, single pass) whose declared purpose — the reader and the use the translation is made for — is written out as numbered clauses and frozen before the source is read for translation. Every logged decision records which numbered clause it acted on. Two R11 runs over the same source unit under two different purposes form a matched pair.

Why it exists

wiki/goodness-senses.md has said since run one that purpose-fit is a sense and that "evaluations under this sense must state the purpose they assume". In eighty-three sessions no translation in this project has ever stated one. Purpose has been an unstated constant — a competent general rendering — across every artifact in workshop/translations/, which is why the sense has no derivation and why the one place the project reached for it (E-20260727-log-decision-coding class C14) recorded a stretch.

R11 is the minimum apparatus that makes purpose a variable. It stands to purpose-fit exactly as R10 stands to naturalness: R10 turns a naturalness anchor into a target the translator can aim at and a reader can check; R11 turns a declared skopos into one.

The pairing is the point, and for a sharper reason than R10's. A single R11 run measures nothing — the translator's own report that it served a purpose is internal-judgment-only. The matched pair is what makes the typological question answerable: with source, translator, session and length held fixed and only the purpose varying, every divergence between the two texts is a divergence purpose produced, and each one can be asked whether any other sense already names it.

The purpose specification — the binding constraint

A purpose specification names a reader and a use. It may not be written in the vocabulary of wiki/goodness-senses.md.

This is the regime's one non-obvious rule and it is load-bearing. A spec that says "maximise accuracy, tolerate translationese" has assumed the answer to the only interesting question about purpose-fit — whether purpose is a re-weighting of the other senses — by writing the re-weighting into the instructions. A spec written as who is reading this, with what open in front of them, and what for leaves the question open, because the translator must derive the re-weighting (if that is all there is) rather than be handed it.

Concretely, a purpose clause may not contain the ids or the definitional wording of any sense on that page. The specs are checked against that rule before they are frozen, and the check is recorded.

What the check is worth, established the first time the regime was used (S084, E-20260801g §9 A5). The mechanical check searches literal strings and cannot stop a paraphrase. The pre-run critic on the first R11 pair found a frozen clause — "the English is not required to be … mistakable for something written in English" — that contained no listed term and was nevertheless naturalness restated as a licence, and would have manufactured a trivial two-sided trade at every site. The check is hygiene on wording; the guarantee has to come from a reader who is not the author. A first R11 use of any spec pair should put it in front of an independent critic before the freeze holds, and that is the recommendation this regime carries rather than a stronger claim for the script.

Roles

• Translator — the lead agent, as a labeled subject (charter §3, A4). No panel call, no API cost.
• Judge — never the lead (charter §5). A single R11 run is not evidence about its own quality.

Human involvement

None.

Procedure

1. The purpose specifications are written and committed before the source is read for translation, as numbered clauses naming a reader and a use, under the constraint above. The commit is the freeze. A clause may not be added, deleted or reworded once translating has begun; a clause discovered to be missing is recorded in the log as missing.
2. R06 §1 unchanged. Source only. The translator reads no published rendering of the same passage into any target language before the translations and their logs are frozen. Priming, if any, is declared.
3. The translator records, before translating, which clauses have an OPPORTUNITY in this source unit — whether the source supplies any material on which the clause could be satisfied or violated at all. Frozen with the specs, per R10 §6, and for the same reason: a clause absent from the result for want of material is not a clause the translator failed to serve.
4. Translate once, straight through, per arm. No return pass, R06 §2.
5. Every logged decision carries a purpose code, assigned at the moment the decision is made, per R10 §5's vocabulary: - A (aimed) — taken in order to satisfy one or more numbered clauses; the clause ids are recorded. - X (contrary) — a clause bore and was not followed, because something else outweighed it; ids and reason recorded. - N (none) — no clause bore on the decision at all. A decision is logged when the translator was aware of two or more live renderings at the moment of writing. The log under-counts total choices by construction, exactly as R07 §5 and R10 §5 do, and code rates are rates over contested sites only.
6. Arm order is recorded and its cost is declared. The second arm is written by a translator who has already rendered the same source once, so the pair is not two independent renderings — note (bhb): across sessions the lead matches itself at up to 37 contiguous tokens, above its record against any published human translation. The carryover is a known confound of this regime and is stated on every artifact, not measured away.
7. Freeze before anything else happens. Both translations, both logs and the opportunity list are committed before any coding, comparison or evaluation is designed.
8. Contamination declared with a one-line basis, measured where a freely reachable published rendering exists (CLAUDE.md, standing rule), on a gate unit translated before the pair's locus is selected (note (bcd)).
9. File under workshop/translations/<work>/R11-<purpose-tag>-v<N>/.

Outputs

Both translations, both decision logs with purpose codes, the frozen purpose specs, the frozen opportunity list, and the contamination measurement.

What R11 is not

• It is not a claim that either purpose is the right one. purpose-fit says a scholar's crib and a general reading edition differ legitimately; R11 executes a declared purpose, it does not endorse one.
• It is not calibrated. Tier D has not passed. An R11 translation is a labeled subject.
• It does not measure whether the purpose was served. Only a reader who was not the translator can say that. What R11 makes available is the decision record — which is descriptive, is what the charter licenses the lead to produce about its own work, and is the only thing this project's use of it depends on.

Change log

• 2026-08-01 (S084) — created and frozen at v1.0, for the matched pair T-julgast-R11crib-v1 / T-julgast-R11read-v1 (E-20260801g-purpose-row).