holding "(green) 5 · (red) 3 · (white) 2" - ai.js:1430, emoji-led source sentiment 8x the check mark U+2713 - SRC_TIER.trusted, dashboard.js:1849 4x direction arrows - dirArrow() 1x U+1F6E1 shield - SRC_TIER.official 1x U+26A0 warning - SRC_TIER.aggregator UNKNOWN is zero. THE CHECK MARK IS THE INTERESTING ONE. It resolves to 'IBM Plex Sans Arabic', which looks like a non-adopting leaf but is the correct answer: U+2713 is in Dingbats, outside the unicode-range of both self-hosted Inter and Vazirmatn (see the @font-face blocks above), and IBM Plex is the only family in the chain that carries the glyph. Naming a face is not the same as failing to adopt one. AND B IS A CONSERVATIVE LOWER BOUND HERE. The resolver classifies a leaf by ONE sampled character, preferring a Latin letter, then an Arabic letter, then simply the first character -- and its Latin test is /[A-Za-z]/, which does not match digits. So a line reading "(green) 5 · (red) 3" is judged on its emoji even though its digits do paint in the contract face. That understates adoption on symbol-led mixed leaves and never overstates it, so it is left alone rather than changed midway through a campaign whose six earlier surfaces were all measured with the same rule. Numeric and mono keep resolving from the global Wave 0 contract. */ #v-deep{ --volurr-font-latin: 'Inter',system-ui,-apple-system,'Segoe UI',Roboto,'Vazirmatn','IBM Plex Sans Arabic',sans-serif; --volurr-font-arabic: 'Vazirmatn','IBM Plex Sans Arabic',sans-serif; --volurr-font-ui: 'Inter','Vazirmatn','IBM Plex Sans Arabic',system-ui,-apple-system,sans-serif; --font-latin: var(--volurr-font-latin); --font-ar: var(--volurr-font-arabic); --font-ku: var(--volurr-font-arabic); --vt-latin: var(--volurr-font-latin); --vt-arabic: var(--volurr-font-arabic); --vt-ui: var(--volurr-font-ui); font-family: var(--volurr-font-ui); } /* Positional, not enumerated. A list of class names rots one sibling at a time - that is how .rv-ancset was missed on relvalue and .rv-fav after it. Every control, by element. */ #v-deep button, #v-deep input, #v-deep select, #v-deep textarea{ font-family: var(--volurr-font-ui); } /* ── #v-portfolio ── WAVE 1 · Candidate B ────────────────────────────────────────────────────────── The VR practice wallet. It owns no stylesheet -- `grep -l v-portfolio app/static/css/*.css` returns nothing -- so the whole migration is this block plus the control neutraliser. Nothing shared is edited. MEASURED, all six cells (en/ar/ku x 375/1366), and every cell is identical: matched coverage 13/13 = 100%, newClip 0, newVClip 0, newOff 0, bidi 0, docOverflow false->false, Sorani 0/13 -> 13/13. Script adoption 9/9 = 100%. ONE NEW WRAP, at en/1366 only: the page subtitle ("Practice wallet - virtual balance for...") goes 1 -> 2 lines in a container computing overflow:visible with auto height, width unchanged. Nothing can be truncated there. INTENTIONAL RESPONSIVE WRAP. RESIDUAL IS TWO LEAVES AND BOTH ARE CORRECT: carries the balance-visibility emoji (portfolio.js:34) and carries the deposit/withdraw arrows and the padlock (portfolio.js:40-42). No UI font should draw those and none does. WHAT WAS NOT MEASURED, AND IT MATTERS. Thirteen geometry-bearing leaves is the REAL mounted state for a visitor with no transactions -- which is what every new user sees -- but it is not the only state. A populated wallet renders `.waltx` rows (portfolio.js:47): `.waltx-ic` an emoji, `.waltx-t` and `.waltx-m` plain text, `.waltx-st` a status chip, and `.waltx-am` a dir="ltr" amount. Those rows inherit from this same root and from the same neutraliser, so the risk is low -- but low is not measured, and populating a wallet to raise the number would mean fabricating transactions to verify, which this campaign does not do. Numeric and mono keep resolving from the global Wave 0 contract. */ #v-portfolio{ --volurr-font-latin: 'Inter',system-ui,-apple-system,'Segoe UI',Roboto,'Vazirmatn','IBM Plex Sans Arabic',sans-serif; --volurr-font-arabic: 'Vazirmatn','IBM Plex Sans Arabic',sans-serif; --volurr-font-ui: 'Inter','Vazirmatn','IBM Plex Sans Arabic',system-ui,-apple-system,sans-serif; --font-latin: var(--volurr-font-latin); --font-ar: var(--volurr-font-arabic); --font-ku: var(--volurr-font-arabic); --vt-latin: var(--volurr-font-latin); --vt-arabic: var(--volurr-font-arabic); --vt-ui: var(--volurr-font-ui); font-family: var(--volurr-font-ui); } /* Positional, not enumerated. A list of class names rots one sibling at a time - that is how .rv-ancset was missed on relvalue and .rv-fav after it. Every control, by element. */ #v-portfolio button, #v-portfolio input, #v-portfolio select, #v-portfolio textarea{ font-family: var(--volurr-font-ui); } /* ═══════════════════════════════════════════════════════════════════════════════════════════════ AR/KU TRACKING NORMALIZATION — the Phase 1H law, applied beyond the calendar pilot ─────────────────────────────────────────────────────────────────────────────────────────────── WHY. `letter-spacing` and `text-transform` are Latin devices. In Arabic script positive tracking is inserted between every glyph pair and breaks the cursive join -- the word stops being one connected form and becomes a row of detached letters. `uppercase` is a no-op on a unicase script and a corruption risk on any Latin fragment sharing the element. news-calendar.css:605 established this for #v-calendar; these are the surfaces it could not reach. MEASURED with tools/typo/ar_tracking_audit.js, ar AND ku, 1366, per selector, counting pure / mixed / latin / inside-[dir=ltr]. A selector is reset ONLY when it is 100% pure Arabic-script in BOTH languages. Everything else is left alone and listed in the report. surface selector pure ar / pure ku reports .nsl 6 / 6 ← the defect found in Run 1, now measured portfolio .cpdemo 1 / 1 portfolio .walc-bll 1 / 1 market .mov-alive 9 / 9 market .mov-pt 6 / 6 market .mov-pill.down 6 / 6 market .mov-pill.up 5 / 5 market .mov-aval.mut 3 / 3 market .mov-pst.down 2 / 2 market .mov-age.warn 1 / 1 market .mov-tab.on 1 / 1 market .mov-pst.amb 1 / 1 news .ngt-dom 24 / 24 news .ngt-why 24 / 24 news .ngt-tab 2 / 2 news .ngt-break-tag 1 / 1 dash .tab 5 / 5 dash .dn 3 / 3 dash .tab.on 1 / 1 DELIBERATELY EXCLUDED, and why each: · every BARE TAG group -- i, u, b, span, h3, h2, em, p. The Phase 1H law forbids resetting a tag globally ("exactly the class of change that turns a fix into a regression somewhere nobody looked"), and dash alone has eleven of them. They need their owning component named first. · market `.mov-tab` -- pure 5 in Arabic but carries a LATIN leaf in Kurdish. Pure in one language is not pure. · news `.ngt-tag` (33 pure, 24 latin), `.ngt-mi`, `.ngt-lang` -- genuinely mixed; each needs its own before/after test. · dash `.chip`, `.state`, `.tag`, `.kwin` and friends -- they appear pure in Kurdish but do not appear at all in the Arabic pass. Absent is not the same as clean; excluded until measured in both. ENGLISH IS THE CONTROL ARM. The html[lang="ar"] / html[lang="ku"] gate means an English reader sees byte-identical output. Any measured English change is a FAIL of this rule, not a side effect. `:not([dir="ltr"])` protects any Latin island that appears inside these elements later. SCOPE. Every selector is prefixed with its surface root, so `.tab`, `.dn`, `.chip` and the other generic class names cannot reach a page they were not measured on. No shared stylesheet is edited: this lives in the contract file rather than in four peer-owned files. NOT CHANGED: font-family, font-size, font-weight, line-height, colour, spacing, layout, or any font-variant-numeric. Two properties only. */ html[lang="ar"] #v-portfolio :is(.cpdemo, .walc-bll):not([dir="ltr"]), html[lang="ku"] #v-portfolio :is(.cpdemo, .walc-bll):not([dir="ltr"]), html[lang="ar"] #v-market :is(.mov-alive, .mov-pt, .mov-pill, .mov-aval, .mov-pst, .mov-age, .mov-tab.on):not([dir="ltr"]), html[lang="ku"] #v-market :is(.mov-alive, .mov-pt, .mov-pill, .mov-aval, .mov-pst, .mov-age, .mov-tab.on):not([dir="ltr"]), html[lang="ar"] #v-news :is(.ngt-dom, .ngt-why, .ngt-tab, .ngt-break-tag):not([dir="ltr"]), html[lang="ku"] #v-news :is(.ngt-dom, .ngt-why, .ngt-tab, .ngt-break-tag):not([dir="ltr"]), html[lang="ar"] #v-dash :is(.tab, .dn):not([dir="ltr"]), html[lang="ku"] #v-dash :is(.tab, .dn):not([dir="ltr"]){ letter-spacing: 0; text-transform: none; } /* ── #v-ai ── WAVE 1 · Candidate B ───────────────────────────────────────────────────────────────── The AGI assistant surface. Owns no stylesheet; migration is this block plus the neutraliser. MEASURED, all six cells (en/ar/ku x 375/1366): newClip 0, newVClip 0, newLClamp 0, newOff 0, bidi 0, docOverflow false->false, Sorani 0/13 -> 13/13, script adoption 100% in every cell. All-visible adoption is 100% in four cells and 94.2-94.4% in the two English ones, where the residual is three symbol leaves. ONE CELL NEEDED THE SAME-STATE GATE. ai/ar/375 first read matched 15/22 = 68.2% -- the page grew while it was being measured, so the delta covered two thirds of it and was not evidence. Re-run with --settle=30 it reported STABLE at 15 leaves and then matched 22/22 = 100%, all deltas zero. The number was never wrong; the coverage behind it was too small to read. One new wrap at en/1366, in a container computing overflow:visible. Nothing truncates. Canvas: ai.js declares zero getContext and zero ctx.font, so the two canvases this surface hosts come from shared components and paint no text. No canvas typography path here. */ #v-ai{ --volurr-font-latin: 'Inter',system-ui,-apple-system,'Segoe UI',Roboto,'Vazirmatn','IBM Plex Sans Arabic',sans-serif; --volurr-font-arabic: 'Vazirmatn','IBM Plex Sans Arabic',sans-serif; --volurr-font-ui: 'Inter','Vazirmatn','IBM Plex Sans Arabic',system-ui,-apple-system,sans-serif; --font-latin: var(--volurr-font-latin); --font-ar: var(--volurr-font-arabic); --font-ku: var(--volurr-font-arabic); --vt-latin: var(--volurr-font-latin); --vt-arabic: var(--volurr-font-arabic); --vt-ui: var(--volurr-font-ui); font-family: var(--volurr-font-ui); } /* Positional, not enumerated. A list of class names rots one sibling at a time - that is how .rv-ancset was missed on relvalue and .rv-fav after it. Every control, by element. */ #v-ai button, #v-ai input, #v-ai select, #v-ai textarea{ font-family: var(--volurr-font-ui); } /* ── #v-feedback ── WAVE 1 · Candidate B ─────────────────────────────────────────────────────────── The improvement centre. Owns no stylesheet; migration is this block plus the neutraliser. MEASURED, all six cells (en/ar/ku x 375/1366), and every one identical with MATCHED COVERAGE 82/82 = 100%: newClip 0, newVClip 0, newLClamp 0, newOff 0, bidi 0, docOverflow false->false, Sorani 0/13 -> 13/13. Script adoption 69/70 = 98.57%, all-visible 81/82 = 98.78%, residual ONE. Two and three new wraps in the English cells only, both in overflow:visible auto-height containers, so nothing can be truncated. The Arabic and Kurdish cells wrap nothing new at all. Twenty of the 102 DOM leaves carry no box and are excluded from geometry -- they cannot clip. */ #v-feedback{ --volurr-font-latin: 'Inter',system-ui,-apple-system,'Segoe UI',Roboto,'Vazirmatn','IBM Plex Sans Arabic',sans-serif; --volurr-font-arabic: 'Vazirmatn','IBM Plex Sans Arabic',sans-serif; --volurr-font-ui: 'Inter','Vazirmatn','IBM Plex Sans Arabic',system-ui,-apple-system,sans-serif; --font-latin: var(--volurr-font-latin); --font-ar: var(--volurr-font-arabic); --font-ku: var(--volurr-font-arabic); --vt-latin: var(--volurr-font-latin); --vt-arabic: var(--volurr-font-arabic); --vt-ui: var(--volurr-font-ui); font-family: var(--volurr-font-ui); } /* Positional, not enumerated. A list of class names rots one sibling at a time - that is how .rv-ancset was missed on relvalue and .rv-fav after it. Every control, by element. */ #v-feedback button, #v-feedback input, #v-feedback select, #v-feedback textarea{ font-family: var(--volurr-font-ui); } /* ═══════════════════════════════════════════════════════════════════════════════════════════════ INTRINSIC-FIT LAW — labels wrap rather than lose text (owner decision, Master Run 3) ─────────────────────────────────────────────────────────────────────────────────────────────── THE OWNER'S RULING. Preserve full text. Do not shrink type, reduce weight, tighten tracking, shorten translations, or revert to Cairo. The layout adapts to the typography, not the reverse. THE SHARED STRUCTURAL CAUSE, and it is one sentence. Each of these labels declares `white-space:nowrap` while its container declares `min-width:0` -- "never wrap, and always be allowed to be narrower than your text". Those two together can only ever resolve as truncation. Cairo's narrower Latin hid the contradiction; Inter exposes it. The bug was always there. MEASURED before choosing, with tools/typo/fit_probe.js (intrinsic width of every real string, in every language, at 375/390/768/1024/1366/1920): mmap .mi-cl-n "Precious Metals" box 78px needs 87.4px clips at 390 and 1366 market .mov-drv-l "REAL YIELD 10Y" box 80px needs 85.6px clips at 768 and 1920 market .mov-drv-l "LABOR MARKET" box 80px needs 84.8px market .mov-drv-l "INFLATION EXP" box 80px needs 82.5px watchlist .wr-nm "Dollar (DXY)" box 78px needs 82.0px clips at 375 WHY WRAPPING AND NOT WIDTH. Widening was tried first, on evidence, and costs columns: mmap's cluster grid is `repeat(auto-fit,minmax(90px,1fr))` inside a panel that is `grid-column: span 4` of twelve, so at 1366 it has only 320.7px and yields three 101.5px cards. Raising the floor to fit 87.4px resolves to TWO columns, not three -- trading a text defect for a density defect, which the brief forbids. Every container here is column-flex or a centred row, so the cost of wrapping is vertical, and vertical is the one resource that is free. `overflow-wrap:anywhere` rather than `break-word`: it also lowers the min-content contribution, so a grid track can still shrink instead of forcing the page to scroll. That matters at 375. NOT INCLUDED, DELIBERATELY. `.mov-drv-s` (driver descriptions, over by +99 to +226px) and `.mi-rib-note` (`max-width:54vw`) are long prose that is MEANT to ellipsis and always did. Making those wrap would add three or four lines to every card. A label and a sentence are not the same thing, and one law should not pretend they are. This rule is exactly what was simulated before it was written -- same four declarations, same three selectors -- so the verification tests what was proved rather than something adjacent. */ /* SPECIFICITY, checked rather than assumed -- and this nearly shipped broken. index.html loads typography-contract.css at position 12, but market.css at 18 and market-live.css at 20, so they come LATER in the cascade. `#v-market .mov-drv-l` (one id, one class) TIES with market-live.css:144 and loses on order, and the rule would have done nothing on market. The simulation could not reveal this: it injected a