v0.19.1 — the class the umbrella had been catching for us, twice

[sshlg]

B-126's board row shipped in v0.19.0 with nine cells against the eight its header declares, and the umbrella's validator refused the re-pin.

That is the second time in three releases. B-124 did the same in v0.18.0 — also caught by sshlg-skills rather than here.

A class seen twice becomes a script

check_ledger_tables_keep_their_shape() counts the cells of every row in docs/evidence/*.md against the header its table declares, splitting on an unescaped | exactly as a reader does. The member now refuses what a sibling had been catching on its behalf.

It found a pre-existing defect on its first run

Older than either of mine: verification.md rows 015 and 016 each quote a shell pipeline inside backticks — … \| grep -c '^def check_' — and that bare pipe had been splitting both rows into five cells since they were written. Nobody had seen them because nothing counted.

The real cause was not markdown

Worth recording, because the two breakages look like the same careless paste and are not:

• v0.18.0 — a grep pattern pasted into a cell, pipes and all.
• v0.19.0 — a Python edit where adjacent string literals were written across lines without parentheses, so only the first fragment bound to the variable. The replacement matched that fragment alone and left an orphaned tail behind, which happened to carry a bare |. The row was malformed because the edit that wrote it was silently partial — the pipe was a symptom.

Evidence

Check	Result
npm test	rc=0 — 14 checks = 10 named + 4 per-skill
Plant (one bare pipe into B-126)	refused — row has 9 cells against the 8
Pre-existing rows 015, 016	escaped; both now 4 cells

Ledger TG-1 … TG-3.

🤖 Generated with Claude Code