Email template hole placement: pin the refused direction of the two scan corrections from PR #25212 #25351

Maintainer-action: write two refused-direction regression rows in packages/formula/src/mail-html-hole-placement.test.ts by hand — done when a PR adding both rows is merged and both rows go red against the classifier with the respective correction of 86330c090 reverted.

Filed by domain:spec seat 2 (seat post #18549), os-bill, session session_01B2EeooonSXrANpkpgUSfFZ, under the maintainer's ruling 甲 6355640946 on #25189. The card goes to the maintainer and is ⛔ not dispatched: an AI session's safety mechanism refused to construct these samples, and the seat does not route around that stop by giving the work to another session.

⛔ Disclosure (ruling #9387, 5323310017). This is a security card. Nothing on GitHub, this card, its comments and the PR that closes it included, carries a sample template, hole, attribute or value string, or a recipe. Each row is named here by class, in the words contract review 6342085020 uses. The test's it(...) title and assertion name the class, as the file's other refused rows do; the sample lives only in the test source.

What is owed

PR #25212 (Fixes #25189) added judgeMailHtmlHoles, the fail-closed classifier behind the email-template-hole-placement authoring rule. Its round-1 contract review found two places where the scan read HTML differently from the HTML tokenizer in the accepting direction. Commit 86330c090 fixed both; round-2 review 6342085020 (①1 and ①2) traced each fix against the tokenizer and passed it. What it found missing (③) is a dedicated test row for either fix:

1. The tag-open state. A character after a tag opener that opens no tag is now reconsumed in the data state, as the specification requires. The row: a hole inside a tag, comment or raw-text element that opens immediately after such a character answers its true class (tag, comment or raw-text-element) and is refused.
2. A script element's escape states. Inside a script element's text the scan now arms an escape flag at a comment opener and disarms it at a comment closer; while it is armed an end tag does not close the element. The row: a hole after an end tag in a script element's double-escaped region still answers raw-text-element and is refused.

Both rows go in the class-by-class table of mail-html-hole-placement.test.ts, in the same shape as its other refused rows. Build each sample from the HTML Living Standard's tokenizer section (the tag-open state; the script data escaped and double-escaped states), not from any shared note.

Acceptance

• Both rows pass at the head of the PR.
• With the respective change of 86330c090 in packages/formula/src/mail-html-hole-placement.ts reverted locally, each row goes red. Record that ablation in the PR body by class only.
• No change to the classifier, the rule, its doors or the renderer. Test-only, so no changeset.

Why it is p3

Review 6342085020 ③ bounds the exposure: both shapes are deliberate-author constructs no mail client renders, each branch is one condition with its invariant stated in a comment, and the neighbouring behaviour is already pinned (the class-by-class rows, the renderer parity pin, the two door tests). The cost is the lack of an alarm if a later edit to one of the two branches loosens it.

Generated by Claude Code