0006 · A typed block speaks placement

Status: accepted, 2026-08-30. Decided by the maintainer. Issue: #81

Phase 0a rewrote a published paper into the syntax and measured the gap that forced this: 634 of 662 figure and table environments in the author’s corpus carry an explicit placement specifier (96%) — [t] alone appears 247 times. A typed block that cannot say [ht] cannot hold the corpus’s normal case, so the block form would go unused and the checked path with it.

The chosen form

A placement field with a bare value:

\figure(fig:cd) {
  placement = !htbp
  ...
}

Zero new grammar: one more row in the §5 field tables, using the bare value kind that already existed.

Rejected

• Brackets after the identifier — \figure(fig:cd)[!htbp] { … }. Reads familiar, but adds a grammar position to the block header, and the header is exactly what the §3 boundary rule keeps minimal.
• A quoted string — placement = "!htbp". Quote noise for a value that never holds a space.

Three rules that carry the semantics

1. Valid bytes are h t b p ! (LaTeX’s own) and H (the float package’s), in any order. Anything else is a hard error naming the byte. Whether float is loaded is a package fact and stays unvalidated, like every other package fact.
2. The value is emitted verbatim inside [ ] — no reordering, no deduplication. The compiler does not improve author intent.
3. An empty field is malformed. Omitting the field is how no-brackets is asked for; writing the field and saying nothing is an error, not a synonym.