council: a composer squeezed to one row says how much of the draft is above

[sanlee-ys]

What a user sees

At the 60x10 floor, in a room that is both gated and short a seat, the compose area gets one row. A three-line draft used to draw there as one row of prose with the typed row breaks turned into spaces, and no marker. It now draws the marker and the tail:

  ╭──────────────────────────────────────────────────────╮
  │  ↑ 2 more above   third line_                        │
  ╰─────────────────────────────────────────────  GATE  ─╯

Nothing else changed. A draft that fits its compose area reaches none of this, which is every room above the floor, and no existing golden moved.

Why the branch was reachable

composerLines flattened newlines to spaces when lay.Prompt == 1. Its comment called that unreachable, and design.md §9.38 repeated the claim as measured. Both were wrong. resolveLayoutIn spends the needs-you strip and the collapsed-seat notice out of the same budget the compose area is measured from, so at MinHeight with a tab bar and both rows on screen the clamp Height - rows - 1 leaves the composer one row while the draft still holds three. A five-seat machine with one vendor not installed draws that room the moment a gate goes up.

The flatten was worse than a clip, which is why no count alone answers it. Clipping drops text and the marker vocabulary describes exactly that. Flattening dropped nothing — it restated the draft, and §7.14 promises the string on screen is the string sent.

The remedy

The marker, not a floor of two rows on the compose area. A floor would close the same gap, and it would take a body row from a room already at its floor and move every golden in the package.

The row is moreAbove — the column overflow marker's own spelling, and the one the multi-row composer already spent a whole row on — then three cells, then the last drawn row, elided from the left. The count is rows not drawn, which is the multi-row path's own arithmetic. The gap is three cells rather than the room's │ because the composer is a box and its sides are that same glyph (§9.44); three cells is needsYouGap's answer to the same question.

The sweep could not see it

TestAMultilineDraftNeverCollapsesSilently built all 588 cells from fiveSeats(), which pins a pending gate and a collapsed seat as absent. That held the compose area two rows taller than a real room's, in every cell, and the test reported the frame correct.

It now varies both, at 2352 cells, and it asserts one more property: two typed rows are never drawn welded into one. Forty cells fail on the old code — heights at MinHeight, three and five seats, gate and notice both present, every width from 60 to 240 once the tabs tier is in play, both glyph sets.

Gates

• go vet ./... — clean.
• go test ./internal/council -timeout 20m — ok, 48.5s.
• go test on every other package — ok.
• go test -race ./internal/council -run over the composer tests — ok.
• Goldens read before acceptance. Only the two new files were written; git status confirms no existing golden moved.

Not verified here: the live Windows Terminal frame. This is an Intel Mac, and the frame is pinned as bytes in both glyph sets instead.

Scope

docs/design.md §9.38 carries the dated 2026-08-17 amendment: what the widened sweep covers, and what the marker shows. STATE.md, the composer's layout and height rules, the editable-composer question and paste handling are all untouched.

🤖 Generated with Claude Code