Repository path: workshop/experiments/E-20260808e-low-pole/site-criterion.md · rendered 2026-09-09
Page metadata (front matter)
| type | note |
|---|---|
| id | E-20260808e-site-criterion |
| status | frozen |
| created | 2026-08-08 |
| updated | 2026-08-08 |
| links | workshop/experiments/E-20260808e-low-pole/design.md, wiki/arms/ARM-low-pole.md |
E-20260808e — the census site criterion
Frozen before any published English translation of any census work was opened, and before the
site lists were built. Committed in the same commit as T-malavoglia-i-R06-v1 and
T-malavoglia-i-R21-v1. This file exists so that "we looked where the translator went low" cannot
be the explanation of anything this run finds.
What a site is
A census site is a stretch of source text at which the source is marked BELOW the neutral written literary register of its own language and period. Not informal; not about low subject matter; marked low as language.
A stretch qualifies if it carries at least one of four device classes:
| code | device class | admits |
|---|---|---|
| LEX | popular/colloquial/vulgar lexical stratum | a word or phrase a standard dictionary of the language marks popolare / familiare / spregiativo / populaire / familier / прост. / разг., or that is unmistakably of that stratum in the period |
| IDIOM | popular set phrase | a proverb, saw, or fixed figure belonging to spoken/popular usage rather than to written literary usage |
| MORPH | nonstandard morphology or phonology written into the spelling | patois forms, elisions, dialect endings, eye-dialect |
| SYN | spoken syntax against the written norm | dislocation, anacoluthon, trailing demonstrative, parataxis standing in for subordination, unattributed report |
What each site is typed as, from the source alone
- N — narration, including free indirect discourse. The words are not inside quotation marks, dashes, or a reported-speech frame.
- D — direct speech, inside the source's own quotation marks or dashes, or a proverb the source puts in a character's mouth.
This is the run's registered moderator. RS-20260806g §2's craft claim — that English's
lowering shelf is dialogue-shaped and its raising shelf is not — predicts that published hands can
reach the low pole at D sites and cannot at N sites. The type is assigned from the source's
punctuation and framing, mechanically, before any English is read.
Exclusions, declared in advance
- Proper names, place names, vessel names, and titles fused to names are never sites, even where they are dialectal.
- Subject matter is not register. A neutral sentence about a fishing boat, a pigsty or a drunk is not a site.
- A site must be locatable to a span of at most one sentence, so that the English at that site is identifiable without judgement about the surrounding paragraph.
- No site is admitted because the lead's translator's log mentions it. The lead's list is built and frozen, and the primary is read on an independent list (below).
Who selects
Two lists, and the primary is not the lead's.
- List A — the lead's, built from the source alone, frozen and committed before any English is opened. It is reported for agreement and is not the list the primary is read on. The lead translated one of the three census sources and knew this run's hypothesis while doing so; its site list is therefore evidence about the lead, not about the source.
- List B — an independent census, produced by a non-panel model given the source text, this criterion verbatim, and nothing else: no English, no hypothesis, no mention of register direction beyond the criterion's own wording, no indication that translation is at issue. List B is the site list the primary is read on.
Agreement between A and B is reported as a number, in both directions, and a low agreement is a result about the criterion, not a licence to fall back on List A.