Lists (enumerations)

The printed Elucidário runs many lists together as ordinary paragraphs: catalogues of maps ("1. Mappa geographico… 2. Geo-Hydrographic survey… 3. …"), the parish-by-parish register of levadas ("Achadas da Cruz: Moinhos, Pereiras…; Arco da Calheta: Rabaçal…;"), outlines of reports ("A Ribeira de Machico…; B Ribeiras de São Jorge… a) Plano geral, b) …"), chronologies of laws separated by semicolons. The digital edition restores them as real lists without touching the text. Code: elucidario/stages/lists.py.

How a list is found

Candidates (code). Blocks with two or more consecutive markers (1. 2. / 1.º 2.º / 1.ª / a) b) / I II — legal and page references such as "art. 3.°", "n.° 2", "§ 1.°", "tomo 1.°" are ignored), three or more semicolons, blocks that begin a numbered item next to another numbered block, and the blocks already typed as list items.
Decision (Claude Opus). For each candidate the model reads the Portuguese and every translation and decides whether it is a list. Ordinary series inside a sentence ("irrigated by the levadas of A, B and C") and ordinals in prose ("the 2nd, 3rd and 5th donatários") stay prose. A list is cut into segments — intro, items (with the printed marker, a "Label:" when there is one, and one level of sub-items), outro — given as verbatim anchors in every language, so all languages share one structure. Blocks with strong list signals that were judged prose get a second opinion at higher effort.
Offsets. Code turns the anchors into character offsets in exactly the text the site renders.

Files

data/13_lists/<lang>.jsonl — one row per list block: article, block, hash of the segmented text, ordered, cont (continues the previous block's list), segs = [start, end, kind, level, marker_end, label_end].
data/13_lists/index.json — the registry: which articles contain lists, and in which blocks. The same count is stored on each article in data/04_structured/articles.jsonl as formatting.lists.

A segmentation is used only while its hash matches the current text: a re-translated block silently falls back to a paragraph until it is re-segmented.

New languages and re-translations

Only the registry is processed: lists.align_submit("<lang>") then lists.align_collect("<lang>") transfer the Portuguese segmentation to the new translation (Claude Sonnet returns the matching anchors). Any later change to how lists work also needs only the articles in the registry.

Display

Ordered lists keep the printed markers ("2.°", "a)", "I") in a hanging column rather than renumbering, labels are bold, sub-items nest one level, and a list split over two blocks continues as one list (within a chapter).