Scripts for shorts 12, 13 and 14: the day the reviewers found nine th…

[peopleworks]

…ings

Scripts only. Nothing is narrated — that spends credits, and the numbers in these want checking against the repository first, which is the whole point of the first one.

12 — the accusation my own code refused to make. A clean bibliography that merely listed further reading was announced as "1 source contradiction", over a note saying the document disagrees with itself, on the page a teacher carries to a committee. The field that draws the line exists and carries a comment explaining exactly why; the line printing the headline never asked it. Second in the series where the tool comes off badly, and worse than the eleventh: that one said nothing, this one said too much, in the direction that accuses.

The figure is five days and one release, verified: the headline entered on 2026-08-05 in 1b2b1b8 and went out in v0.4.0 the same day. I had written "months" in the previous commit message and in PR #53 without checking. The PR body now carries the correction rather than a quiet edit.

13 — a report that swears nothing was uploaded, fetching a pixel. Escaping the angle bracket blocks HTML and does nothing to Markdown's own image syntax, so the promise on the last line was broken by the document being analysed.

14 — two reviewers, two questions, no overlap. Nine defects between them and not one in common: the claims reviewer found no escaping bug, the attacker found no overclaim. The most transferable thing this project has learned this week, and the only one of the three aimed at developers rather than teachers.

What this changes

Why

Checklist

• dotnet test passes, with no new build warnings
• One idea per PR

If you touched the rule packs (rules.en.json / rules.es.json):

• The rule carries a suggestion — what the writer should do instead
• There's a test in tests/SignsOfAI.Core.Tests covering it
• CleanHuman in ScoringTests still scores exactly 0 (the new rule doesn't fire on ordinary prose)
• Any regex is bounded — no unbounded .*; this runs on every keystroke in the browser
• Spanish rules are derived from Spanish AI output, not translated from the English pack

If you touched the articles in Docs/Blog:

• Re-ran the CLI and updated the scores the articles quote about themselves