The note reads at 16px and 40em, measured against the editor people arrive from

[lilseyi]

The reading measure landed in #508 and the owner said it still read wrong. It did, and the measure was only half of why.

1 — 40em, not 36em

36em drew 522px of text. Measured against Obsidian in a pane of the same width — 1010px — Obsidian draws 582px where we drew 522px. A tenth narrow, in the one place a reader has something to compare this to.

The old number took the low end of the 60-75 character band, on the reasoning that erring short is the cheaper error. That holds in isolation and loses to the comparison: a measure is comfortable across a range, and inside that range being conspicuously narrower than the editor somebody used yesterday is a cost the band does not price.

2 — 16px on a pointer, not 14.5px

The real cause. The pointer layout drew the note at 14.5px while a phone drew it at 16px, and nothing in the code argued for the smaller number. Because the measure is a multiple of the note's own type, that one value was shrinking the column as well as the glyphs — two defects from one number.

Measured from the owner's own screenshots, our console beside Obsidian at comparable width:

	pane	text	% of pane	gutters	ink on a line
before	1116	468	42%	319/329	12px
Obsidian	1016	564	56%	214/238	14px

Small text lost in a wide pane, which is what it was. Raising the type fixes both without touching the measure again — 40em resolves against 16px instead of 14.5px, so the column widens on its own.

Measured, before and after

Chromium on the built export:

viewport	before	after
desktop 1440x900	522px (36em at 14.5px), gutters 319	640px (40em at 16px), gutters 260
phone 390x844	342px, gutters 0	unchanged — already 16px, and the measure cannot bind inside 24pt gutters either way

The behaviour model is unchanged and was already right: the cap is a true maximum, so the gutters absorb every pixel of a narrower pane and collapse to zero before the text does. Verified across twelve viewport widths — the column holds until the pane can no longer fit it, which is what Obsidian does and what the owner identified as the thing to match.

Self-review

• A test named for the opposite of what it proved. the measure changes with the density, not with a second breakpoint asserted 16px against 16px once this branch landed. Renamed to what it holds, with the reason the size stopped branching and a note that re-splitting it is a decision rather than a tidy-up — the measure is a multiple of the type, so a second size moves the column too.
• Comments that had become false. Three passages justified the measure's unit by the two densities' sizes differing. The faces still differ — Times New Roman on the wrapper, a sans in the note — which is what made ch ambiguous and is why the indirection stays; their sizes agreeing today is not something it may lean on. Corrected rather than deleted.
• A side effect I am not fixing here and am not hiding: --lp-leading still branches, 1.75 on a pointer. At 14.5px that was a 25.4px line pitch; at 16px it is 28px, so desktop lines are about 10% airier than before this branch. That is a consequence of the size change, it is legible, and picking a new leading is its own decision with its own comparison to make. Flagged rather than folded in silently.
• The editor bundle changed only for --lp-measure; --lp-size reaches the document through themeVars at runtime, so the size commit reproduced the bundle byte for byte.

Checks

• apps/mobile: 281 suites, 5,340 passed, lint 0 errors, tsc --noEmit clean.
• apps/mcp: ALL PASS, untouched.
• pnpm test:e2e:webkit:chromium: 43 passed, including the reading-measure spec at both viewports and the case that clicks the margin beside a line to prove the caret still lands in the note. I ran the chromium project; a chromium pass is never a WebKit result — CI's Editor in WebKit job is the WebKit one.
• The e2e bounds (30-42em, 50-95 characters) already admitted 40em and are deliberately wider than the design number; the comments quoting 36 are updated.

🤖 Generated with Claude Code

https://claude.ai/code/session_012pS75o4hy4RDLUegCyrLcx

---

Generated by Claude Code