Text in a housekeeping destination

How the text is hidden

The same \* ignorable marker as the technique above, but on a destination the specifications name: \*\cs, \*\bkmkstart, \*\picprop, \*\wgrffmtfilter, footnote and endnote separators, embedded-object and XML plumbing. Every word processor writes these and nobody types them; what they carry is identifiers rather than prose. They are not drawn on the page and they are not removed from the file.

Why a model still reads it

Mechanically identical to any other ignorable destination: readers skip it, token-stream extractors keep it. So a payload written into a recognised destination reaches a model exactly as one written into an invented destination does.

What we do about it

A destination word on the parser's spec list sets rtfSpecDestination on the run, and the pack rule keyed on that flag grades it informational with action disclose — the band is the only thing that changes. Nothing is skipped: the words remain in text.canonical, pinned by test, precisely so a recognised destination does not become a hiding place. Severity is a floor — the shared instruction-shape matcher promotes a planted instruction out of the quiet band, and that bypass is written as a test. The technique is not on the CERTIFIABLE_DISCLOSURES allowlist, so it still withdraws certified while leaving the verdict clean.

How often it fires

25% of 195 real RTF files from four national open-data portals (CA, UK, GR, AU) — 196 of 201 parse, measured 2026-08-31.

This is an alert-volume number and nothing else. It says how often the alarm sounds on documents as found — not how often it is right, and not whether what it found was harmless. Documents as found may themselves carry concealment. Read it against the population named above rather than as a property of documents in general.