id(web): world presenter still mis-measures ⬛/⭕, narrow-base skin tones, composed Hangul jamo and long graphemes — #1431's folded-in classes survived its fix, and the 16-unit symbol cap has no Rust twin #1581

Summary

The web world presenter (pkgs/id/web/src/world-surface.ts) still diverges from the Rust validator for four grapheme classes. CLOSED #1431 fixed emoji-modifier and tag sequences (PR #1542), but two classes that were folded into #1431 as evidence (comment 6446198995) were not touched by that fix, and two further classes in the same function were found this run. In every case a screen the Rust side validates and the SSH presenter renders (world_presentation.rs:219 parse, :350 render_ansi, both measuring with UnicodeWidthStr::width under unicode-width 0.2.2, pinned in pkgs/id/Cargo.lock) is either refused whole by the web presenter — renderWorldSurface returns false, so world.ts:175-176 falls back to dumping the raw frame text — or, worse, accepted at the wrong column count, producing a misaligned grid.

Verified on main 601af5a (2026-10-27, lane-B creator run)

Method: TS probe on the laptop driving the verbatim world-surface.ts (sha256 3ee9823c identical both sides; Bun 1.4.2, minimal DOM stub, renderWorldSurface called directly) against a Rust probe crate on the laptop using the pinned unicode-width =0.2.2 + unicode-segmentation (the exact crates world_presentation.rs uses). Controls behaved: x Rust 1 / TS renders at 1; U+1F600 Rust 2 / renders at 2, refused at 1; the #1542 controls U+1F44D U+1F3FD and the England tag flag render at 2.

1. ⬛ U+2B1B and ⭕ U+2B55 — wide in Rust, missing from WIDE_RANGES

• Rust: both are 1 grapheme, width 2.
• TS: refused at columns=2, wrongly accepted at columns=1 (WIDE_RANGES, world-surface.ts:60-127, has no U+2Bxx entry; codePointWidth falls through to 1).
• This class was reported on id(web): world presenter rejects whole screen when a cell has a skin-tone emoji or tag flag — TS grapheme width sums modifiers instead of measuring the cluster #1431 (evidence comment 6446198995) and survived its fix.

2. Narrow base + skin-tone modifier — ☝🏻 U+261D U+1F3FB

• Rust: 1 grapheme, width 2 (unicode-width measures an emoji-modifier sequence at 2 even when the base alone is narrow).
• TS: refused at columns=2, accepted at columns=1. PR fix(id/web): measure emoji-modifier and tag sequences as their base — world presenter (#1431) #1542's modifier rule (world-surface.ts:162-167) strips the modifier and measures the base alone, so a narrow base measures 1 — correct for the bases id(web): world presenter rejects whole screen when a cell has a skin-tone emoji or tag flag — TS grapheme width sums modifiers instead of measuring the cluster #1431 tested, wrong for narrow bases.
• Also reported on id(web): world presenter rejects whole screen when a cell has a skin-tone emoji or tag flag — TS grapheme width sums modifiers instead of measuring the cluster #1431 and survived its fix.

3. Composed Hangul jamo clusters — U+1100 U+1161 (LV) and U+1100 U+1161 U+11A8 (LVT)

• Rust: each is 1 grapheme, width 2 (a jamo syllable measures as the precomposed syllable U+AC00 does — and the precomposed form renders correctly in TS).
• TS: refused at both columns=2 and columns=1 — codePointWidth sums choseong 2 + jungseong 1 + jongseong 1. Any decomposed-Hangul screen (NFD text, jamo input methods) falls back whole.
• New this run; the id(web): world presenter rejects whole screen when a cell has a skin-tone emoji or tag flag — TS grapheme width sums modifiers instead of measuring the cluster #1431-era sweep tested jamo only as single code points, which match.

4. One grapheme longer than 16 UTF-16 units, and a lone spacing mark

• e + 17× U+0301 (combining acute): Rust 1 grapheme, width 1. TS refuses at columns=1 — pinned to the cell.symbol.length > 16 check (world-surface.ts, cell validation in renderWorldSurface): the same grapheme at UTF-16 length 15 and 16 renders, at 17 and 18 it is refused. The Rust validator imposes no symbol-length limit — only exactly-one-grapheme and width 1–2 — so the cap is a TS-only rule that rejects Rust-valid cells.
• Lone U+0903 (Devanagari spacing combining mark, category Mc): Rust width 1; TS width 0 via the /^\p{M}$/u rule (world-surface.ts:133,142), so the cell is refused at both column counts. (Inside a cluster after a base consonant both sides agree at the base's width; only the standalone mark diverges.)

Why it matters

The presenter is the web half of a two-presenter contract: the same validated ScreenSurface must render on SSH and web. Every divergence either drops the whole screen to a raw-text dump (classes 1–4, at the correct column count) or silently misaligns it (classes 1–2, at the wrong count — a guest that disagrees with the Rust validator, or a future Rust change, would pass the web check it should fail). ⬛/⭕ are ordinary emoji a world program can plausibly draw as tiles; decomposed Hangul and accented Latin are ordinary text.

Acceptance criteria

• renderWorldSurface agrees with UnicodeWidthStr::width (unicode-width 0.2.2) for: U+2B1B, U+2B55 (2, and refused at 1); U+261D U+1F3FB (2, refused at 1); composed jamo LV and LVT clusters (2); a lone U+0903 cell (1).
• The symbol-length rule is reconciled with the Rust side: either the TS cap is removed/raised so a 1-grapheme, width-1 symbol of 17+ UTF-16 units renders, or an explicit matching cap is added to world_presentation.rs with a comment naming the shared bound — the two validators must not disagree silently.
• A parity test in world-surface.test.ts covers each class above at its correct column count (renders) and at the wrong one (refused), in the style of the existing PR #759's world-surface width check mis-measures astral-plane, ZWJ and ⚠️ glyphs (JS width 1 vs Rust unicode-width 2) — must not close #724 as-is #761/id(web): world presenter rejects whole screen when a cell has a skin-tone emoji or tag flag — TS grapheme width sums modifiers instead of measuring the cluster #1431 tests.
• Existing presenter tests still pass; the fix(id/web): measure emoji-modifier and tag sequences as their base — world presenter (#1431) #1542 modifier/tag controls (U+1F44D U+1F3FD, England flag) still render at 2.

Notes

• Follow-up remainder of CLOSED id(web): world presenter rejects whole screen when a cell has a skin-tone emoji or tag flag — TS grapheme width sums modifiers instead of measuring the cluster #1431 / PR fix(id/web): measure emoji-modifier and tag sequences as their base — world presenter (#1431) #1542 (classes 1–2 were its evidence comment 6446198995); classes 3–4 are new findings in the same function from this run's sweep. Distinct from OPEN id(e2e): no Playwright coverage of the world surface — presenter, input, records, resize untested in a browser (#1076 slice 3 gap) #1322 (Playwright e2e coverage of the presenter — that is test-infrastructure, not a width defect).
• Probe artifacts: hidden_files/probe-2026-10-27-laneB-surface2/ in the goal workspace (probe.ts, boundary.ts, main.rs, out.log, rust-out.log); laptop copies under ~/git/fleet/probe-2026-10-27-laneB-surface2/.