/* fit.css — the CSS half of the fit-to-box pass (B-073). * * WHY * `.slide-content` is `overflow: hidden` (base.css:220), so a slide with one line * too many does not scroll and does not warn — it silently loses the bottom of * its content, and the deck looks finished. Static character budgets cannot * prevent that: the same content occupies different space across 16 presets, two * scripts (keep-all makes Korean break at completely different points than * English at equal character counts — B-047) and every viewport. So the fit is * measured at runtime and applied here. * * HOW THE SCALE REACHES THE TYPE * fit.js sets `--fit-scale` / `--fit-space` INLINE ON EACH `.slide`, and the six * type steps plus the slide padding are re-declared below at `.slide` scope so * they resolve per slide. Declaring them at `:root` cannot work — a custom * property substitutes at its declaration site, so a `:root` step would keep * reading the root's scale no matter what a slide sets. Same trap the dark-theme * aliases hit (tokens.css:139-145). * * WHY NOT `transform: scale()` * Because it would look right on screen and ship clipped: the PDF path force- * resets transforms on slide content (`body.pdf-capture .slide .slide-content > * * { transform: none !important }`, base.css:408-412). Scaling the type system * survives every export path. * * THE FLOORS ARE IN THE CSS, NOT IN THE SEARCH * Each clamp keeps B-048's decided minimum as its literal MIN term, so no scale * fit.js could ever compute will drive text below the projector/A4 legibility * floor. The JS only has to notice that it has stopped gaining room. * * @media screen ONLY * print.css pins larger fixed sizes at `:root` for A4 (`--small-size: 0.95rem`, * `--body-size: 1.1rem`, print.css:10-12). A `.slide`-scoped declaration * out-specifies that, so if this block were not screen-only it would silently * delete the print floor. * * Opt-in: inert until carries `data-fit`. */ @media screen { body[data-fit] .slide { /* B-096: multiply the AUTHORED ladder, never restate it. These used to spell out tokens.css's own terms — `clamp(2rem, calc(6vw * var(--fit-scale,1)), calc(4.5rem * …))` — which is correct arithmetic against the DEFAULT ladder and a silent deletion of any other one. A preset's `type.titleSize` lands on `:root` (0,1,0); this rule is (0,2,1) and won on every slide, so both art-direction bundles rendered their h2 at 48px where broadsheet authored 34.4px and signal 57.6px — two decks whose whole point is a different type voice, reading identically. `--fit-scale` was `unset` throughout, so this was not the pass working: it was the inert layer overwriting. The `-base` aliases in tokens.css hold whatever won at `:root` (see the note there for why a second name is unavoidable), and both factors still default to 1, so a slide fit.js never touched renders as authored. The `max(min(...))` wrapper is the B-048 floor, kept. Multiplying the whole clamp would scale its floor too, and the gate caught that immediately: small resolved to 10.6px at the 0.78 bottom, under the 13.6px projector/A4 floor those literals exist to hold. The floor term is `min(base, LITERAL)` rather than the literal alone so that scale 1 stays an exact no-op even for a preset that authored a step BELOW the shared floor — clamping such a deck up would be this same bug wearing the opposite sign. */ /* B-112: the floors are CANVAS pixels now, like the ladder they guard. They used to be rem, which is the window's idea of a pixel — on a scaled canvas that is a different unit from the one the type is written in, so a floor of `0.95rem` would have bitten at a different point on every screen. Rewritten at the canvas equivalent of the same physical minimums: at a 1440 window the canvas scale is 0.75, so a 20px canvas floor renders 15px, which is B-048's 15.2px projector/A4 floor within a rounding error. */ --title-size: max(min(var(--title-size-base), 40px), calc(var(--title-size-base) * var(--fit-scale, 1))); --h2-size: max(min(var(--h2-size-base), 32px), calc(var(--h2-size-base) * var(--fit-scale, 1))); --h3-size: max(min(var(--h3-size-base), 26px), calc(var(--h3-size-base) * var(--fit-scale, 1))); --subtitle-size: max(min(var(--subtitle-size-base), 24px), calc(var(--subtitle-size-base) * var(--fit-scale, 1))); /* B-048 floors, restated on the canvas: body 20px, small 18px. */ --body-size: max(min(var(--body-size-base), 20px), calc(var(--body-size-base) * var(--fit-scale, 1))); --small-size: max(min(var(--small-size-base), 18px), calc(var(--small-size-base) * var(--fit-scale, 1))); /* Rhythm shrinks on its own axis: on component-dense slides the space between blocks is where the room actually is, and type scaling alone recovers little (measured ~19% of the excess on a 10-row stack). This one had the same defect for the same reason — it restated tokens.css:53 `clamp(2rem, 5vw, 6rem)` and an earlier draft mis-typed the restatement as `clamp(1.5rem, 4vw, 6rem)`, changing the padding on every data-fit deck even at scale 1 (.slide-content clientHeight 888 -> 926 at 1920x1080). Deriving removes the class of bug rather than the instance: there is nothing left to mis-copy. */ --slide-padding: max(min(var(--slide-padding-base), 2rem), calc(var(--slide-padding-base) * var(--fit-space, 1))); } /* B-096: the chart box is the one region the type ladder cannot reach — `.chart-box` is sized in vh, so scaling a slide moves its heading and its rhythm and leaves the plot exactly where it was. That fix is real and it is NOT here: it lives in charts.css, inside `.chart-box`'s own rule, where the space factor is multiplied into the height directly. B-105 moved it. The version that lived here was `body[data-fit] .slide .chart-box { height: … }` — (0,3,1) on a FINAL property, which silently beat every `.chart-box` rule an author writes in deck.css, the file index.html:49 promises is theirs. The rule two blocks up states the principle it violated: multiply the authored value, never restate it. A custom property can be declared here safely because the author's own declaration substitutes at ITS site; a final property cannot. Do not re-add a `.chart-box` height rule at this scope. If a future reserved region needs the same treatment, give it a `--…-base` custom property and multiply it in that component's own stylesheet, the way charts.css does. `ci-validate-fit.mjs` pins both halves: that the fit pass still reaches the box, and that a deck.css height still wins. */ /* Deck furniture must not resize slide to slide, or the folio and eyebrow visibly breathe as the reader flips. Reset the scale for the motif layer's own pseudo-elements — the counter value is scale-immune but the string it renders is not. */ body[data-fit] .slide::after { --fit-scale: 1; --fit-space: 1; } /* The honest failure state. When the search bottoms out at the legibility floor and the content still does not fit, the slide says so instead of quietly cropping: authors get a visible marker, and the CI probe / slide-reviewer can name the slide. Kept subtle enough not to wreck a screenshot, and dropped entirely from every capture path. */ body[data-fit] .slide[data-fit-state="overflow"]::before { outline: 2px dashed var(--negative); outline-offset: -6px; } body.pdf-capture .slide[data-fit-state="overflow"]::before { outline: none; } }