A mark, not the same mark twice

[tannevaled]

The bullet rule asked for one repeated mark, and that was the wrong
question twice over.

it read • • - - as a table of four rows	the bullets were not all the same character, and a reader sees two lists there. go-pdfkit/document found this: it could not make a list of what extract had already called a table.
it read a first column of A and A as an ornament	because A is one character and it repeated. A label is data. On a 6-page technical paper that silently cost a table.

The question is whether the column carries information.

• a mark of one or two characters with no letter and no digit does not:
  - carries exactly as little as •;
• a letter or a digit does: A/B are labels, 1./2. are a numbering
  somebody may want, 90/91 are years. Taking those would be deciding that
  a table of dates is a list.

Measured

The 6-page paper now reports 15 tables where it reported 14, and the extra
one is real. The 213-page bullet-list document is unchanged at 40, which is
the control this rule was written against in the first place.

A spot check of the paper's first page, which is the reason the rule matters:

-- p1 9x4
  | Système      | Configuration réelle          | Prix estimé 2026 | À retenir
  | HGX B300     | 8 boîtiers B300, 2,1 To HB... | 300–500 k$       | seul déployable en salle a...
  | GH200 NVL32  | 32 superpuces GH200, 4,6 T... | ≈1,5–1,9 M$      | génération sortie ; 2 à 3 ...
  ...

A dead guard went with it

seen reported "found no mark at all", and dropEmptyColumns has already
taken away the columns nothing was drawn in — so nothing could reach that
state. The invariant is named where the loop ends instead of being tested
where it cannot happen.

100% of statements, no uncovered blocks, 9 mutations caught.

🤖 Generated with Claude Code