Repository path: workshop/experiments/E-20260731e-option-census/census-rule.md · rendered 2026-09-09
Page metadata (front matter)
| type | note |
|---|---|
| id | census-rule-20260731e |
| status | frozen |
| created | 2026-07-31 |
| updated | 2026-07-31 |
| links | workshop/experiments/E-20260731e-option-census/design.md, wiki/arms/ARM-option-census.md, wiki/findings/results/RS-20260730i-candidate-reach.md, framework/closure.md |
The option-census instrument — frozen 2026-07-31 (S073) BEFORE any source text was opened for translation
Why this is a separate file and why it is committed first. Standing critic disposition 2
(workshop/experiments/README.md): "The instrument is frozen before the design that uses it, not
authored alongside it." This session's whole question is whether the option census — the object
every published framework-reach figure is computed over — is a property of the decision or of the
translator. An instrument written after the sites were chosen could be shaped to the answer.
Nothing below was written with any locus of the source in view. The source work had been chosen (Sienkiewicz, «Janko muzykant», 1884) and its Polish text fetched; no span had been selected and no word had been translated when this file was committed.
⚠ AMENDED 2026-07-31, after the pre-run critic's BLOCKING finding 1, and the amendment records a fact rather than repairing a wording. The critic observed that this file names Sienkiewicz as the source work while
design.mdand every item use Reymont's «Śmierć», and refused to let the run proceed on the contradiction. The contradiction is real and its cause is the standing contamination gate.The sequence, verifiable in git and stated so it need not be taken on trust:
8c12508froze this file, naming Janko.0c338b1froze 308 words of Janko as the gate's probe unit. The gate then returned 18 tokens and 7 shared 12-grams against Curtin 1893, and the work was abandoned (workshop/translations/janko/R06-v1/unitA.md).f1b07abrecorded that and froze the replacement span, Reymont's «Śmierć», whose gate came back clean at 8 tokens.This file's rules were not edited when the work changed, and are not edited now. §§1–5 are byte-identical to
8c12508. That is the point the critic was protecting: the instrument predates the Reymont span by two commits, so it cannot have been shaped to it. The naming sentence above is left standing rather than corrected, because rewriting it to say "Reymont" would destroy the only evidence that the instrument was frozen before either work was settled on.
1. What a census site is
A census site is a locus in the source at which the translator, at the moment of choosing, weighed more than one English rendering. It is recorded then, in the same pass as the translation, and is never reconstructed afterwards from the finished text.
Each site records, verbatim:
| field | content |
|---|---|
id |
S##, assigned in source order |
span |
the smallest Polish substring that carries the choice |
context |
the full source sentence containing span |
options |
every distinct English rendering actually weighed, each as a short phrase |
chosen |
which of options went into the translation |
class |
CONSTRAINED, OPEN or MIXED, by §2, assigned at the moment of decision |
options is a set, not a ranking. A rendering goes in if the translator would have been prepared to
write it; it stays out if it was never in play. The chosen rendering is always a member.
2. The class rule
Assigned at the moment of decision, before any panel sees anything.
CONSTRAINED— the locus is a proper name, a numeral, or a common noun/verb/adjective with a single standard English equivalent in a general Polish–English dictionary and no cultural qualifier attached to it.OPEN— the locus is a culture-bound item, a dialect or sociolect form, or a figure whose rendering requires choosing among the eight-wide handling set ofCL-20260726-handling-set.MIXED— neither cleanly.
3. The two item sets the design draws from
- Primary items: the first 24 census sites in source order. Stated as a rule so that the item set cannot be selected after the options are known. If the span yields fewer than 24 sites the span is extended forward in the source until it does, and the extension is declared.
- Control items: 8 constructed loci, 4
CONSTRAINEDand 4OPEN, chosen for being as unambiguous in their class as the passage allows, and drawn from outside the decision census. The 4CONSTRAINEDcontrol loci need not have been decisions at all — that is the point of them: a locus with one live rendering is invisible to a decision log by construction, and the control needs the low end of the scale to exist.
Control items are declared before dispatch and are put to the independent pre-run critic for endorsement or contest sight-unseen of the lead's rationale, on the S070/S072 precedent.
4. What the instrument does not do
It records no quality judgment of any kind, about any option, by anyone. options is a list of
things a translator would have been willing to write; nothing in the census says which is better,
and chosen records what happened rather than what was right. No sentence produced under this
instrument is evaluative, so none of it needs internal-judgment-only — and any sentence that turns
out to be evaluative carries the flag.
5. The known defect, declared in advance
The lead cannot be an unprimed subject. It wrote none of the fourteen candidates in the sense of
inventing them, but it has read framework/traceability-inventory.md many times and is the author
of every census this project holds. The lead's own census is therefore the maximally primed
reading and is reported as a descriptive reference point only. It is never a condition of the
primary comparison, which is between three independent seats under three frozen preambles.