Repository path: workshop/experiments/E-20260728j-classb-marking/dole/sites.md · rendered 2026-09-09

Stage 2 — Dole 1896 at the nine Class B sites

Read only after Stage 1 was committed at e923abc. Comparator: Nathan Haskell Dole, "Jeli, the Shepherd", in Under the Shadow of Etna (Boston: Joseph Knight, 1896), Project Gutenberg #37979 — public domain. 12,862 words, 309 sentences by fetch_dole.py's splitter. Raw candidate sentences: candidates.json.

Grades apply the frozen §3 definition as amended. AMBIGUOUS where the anchor is present but the clause structure does not correspond; per amendment B2 it counts against the lead's prediction.

Three things about this table that are not in the grades

B3 — the mark can be evaded by removing the clause. Verga's come fa un cagnuolo di latte is a clause with a finite verb; Dole compresses it to a prepositional phrase, "a whine like a sucking puppy". There is then no verb to be present or past. Neither the frozen policy nor the §3 definition anticipated a third option beside carrying and flattening, and it is coded AMBIGUOUS on the letter of amendment B2 — which is also the coding that costs the lead's prediction, so nothing turns on the judgment call.

B4 — Dole marks the passage, at a site the inventory does not contain. Dole's sentence opens "because the disease is one of those clear and evident ones" — a present tense where the Italian has an imperfect (la malattia **era**). He declines the mark at B4 itself and creates one, seven words earlier, where the source has none. RS-20260727-jeli-fid §4 says the frozen inventory is "a floor, not a census"; this is the first evidence that it is a floor for the comparator too, and it bears directly on the B6 reading below.

B9 — the m2 closed list is too narrow, and the defect is reported rather than patched. Dole's "must be grown tall" is the same construction as the lead's "must have grown up", and the frozen list contains must have and not must be, so the mechanical rule grades Dole UNMARKED at a site where the lead is MARKED. B9 is not in P4's denominator, so no reported figure moves; the list stays as frozen and this line records why the cell is wrong.

The anchors failed at exactly the two sites that falsify the prediction

The anchors were frozen in forced/renderings.md §6 before Dole was opened, on the critic's amendment B2, precisely so the lead could not rescue P4 by declaring an inconvenient site non-corresponding. Seven of nine hit. The two that missed are B6 and B7 — the only two sites in the study where Dole marks a conditional, i.e. the two cells that falsify both P4 and P5.

• B6's anchor required sulphate. Dole writes quinine. A realia substitution, which is strategy three of the seven on cultural-mediation's own list.
• B7's anchor required one of help / shift / manage / take care / care of. Dole writes "do for himself".

Applied mechanically, the procedure the critic prescribed to stop the lead protecting its own prediction would have protected it — it would have returned NO-COUNTERPART at both sites and left P4 and P5 confirmed. The sites were found by reading the surrounding paragraph, and they are coded here against the lead's prediction because they plainly correspond.

The mechanism is general and is not bad luck. An anchor written in the lead's own target vocabulary is a list of the words the lead chose. At a site where the comparator did something different, the wording diverges most — so the anchor is least likely to match exactly where disagreement lives. An anchor set built this way is biased toward finding agreement, and it is the same shape of defect as a control that can only confirm. Method note (bdi).