 :root{color-scheme:light;--font:-apple-system,BlinkMacSystemFont,'Segoe UI',system-ui,sans-serif;--text-2xs:11px;--text-xs:12px;--text-sm:14px;--text-base:16px;--text-lg:20px;--text-xl:24px;--text-2xl:32px;--text-3xl:40px;--leading:1.4;--leading-tight:1.15;/* The two light surfaces. Three passes have now taken them down,and this one is the step that ends the near-white field:the first two moved 6-9/255 and then 9-11/255 and each still read as a bright theme,so this one moved 26/255 - 2.5x the previous step. They are still a pair and they still move together:--surface stays a step ABOVE --bg (8/255) or a panel stops reading as raised off the page,and --surface-hover stays 9/255 below --surface so a hover still reads as a hover. Warm-neutral is held exactly - every value keeps R>G>B on the same hue line,and nothing went grey-blue. WHAT ACTUALLY GOT BRIGHTER,and it was never these two:under a panel they are invisible (see .card - the rim gradient is opaque and covers the face),so the panel's own top stop is the brightest pixel on the screen and it is the value that has to carry a dimming pass. It came down with them,to #d6d4d0 from #f0eeea. Measured on the render,every surface a glyph lands on came down 25-27/255,read off the same elements before and after:a dashboard tip-table header #e3e0db -> #c9c6c1,a stat label #e3e1dd -> #cac8c4,a table head on account detail #eae7e3 -> #d1ceca,the menu list at its external-link glyph #eae8e4 -> #d1cfcb,the rail #efedea -> #d5d3d0,the page #e7e5e1 -> #cdcbc7. Whole-screen the change is not a nudge:on the dashboard 86.6% of pixels were brighter than 217/255 before this pass and 0.9% are now,and mean relative luminance fell 0.753 -> 0.578. The limit is not these values either - it is the text scale,which had to come down with them or fail. See --text-muted. */ /* THE PAGE AND THE PANEL MOVE IN OPPOSITE DIRECTIONS,which is the one thing this pass does. Every earlier pass moved the whole screen as a block - the comment above records the last one taking 26/255 off page,rail and panel together - and the result was three greys nobody could rank:page,panel and control sat within ~8/255 of each other,so nothing said which was the content. Here the page goes DOWN 14/255 and the panel goes UP (see --panel-rim),and the step between them is what carries the hierarchy. --surface (rail,fields,controls) does NOT move with either:it is the middle level,and the panel's own face is the gradient,not this token. THE LADDER,measured off the render at 3x with the grain in it:page #bbb9b4,rail and field faces #d5d3d0,panel face #f7f5f2 at its middle and #f3f1ed at its darkest - three steps nobody has to be told the order of. */ --bg:#bfbdb8;--surface:#d5d3d0;--surface-hover:#cccac6;/* THE FACE OF A TEXT FIELD,and it is a step lighter than --surface on purpose. A field that holds words was wearing the same value as the rail and every boxed control,and on the near-white panels of this pass it read as a dark box with small type in it - the numbers said the typed text was fine (--text on --surface is 12.71:1) and the eye said the box was heavy. #e6e4e0 lifts the typed text to 15.4:1 and a placeholder from 6.55 to 7.8,and the field still reads as a field because what says so is the well and the --border-strong edge,not the value of the face. Buttons do NOT take this token:a key's face is --surface and stays it. */ --field-face:#e6e4e0;/* The container edge,as a hairline. It stopped being #d6d3d1 because a panel now carries its own shading:with a full-strength outline over a rim gradient the box reads as an outline and nothing else,which is the flatness this language replaces. A hairline keeps the boundary at the top of a panel where the ambient shadow is invisible,without out-shouting the shading that defines the box everywhere else. Controls do NOT take this token - see --border-strong. */ --border:rgb(0 0 0 / 0.11);--border-strong:#5b5852;/* Row rules and dividers inside a panel. Lighter than the box edge on purpose:the box edge and the row rules have to move together,or the table reads as a grid drawn on paper. */ --border-light:rgb(0 0 0 / 0.075);/* The header row's band. A table's header is a different KIND of row - labels rather than data - and on a near-white face it needs to say so,or the eye reads it as the first record. A band rather than a heavier rule:the rule under it already exists (th carries --border-light),and thickening that alone made the header read as a lid. Token,not a literal,so dark gets its own value from the same rule. */ --th-band:rgb(0 0 0 / 0.045);--text:#12100e;/* BOTH SECONDARY LEVELS,AND THEY ARE NOW ONE LEVEL. This is the decision this pass had to make,so it is written out rather than discovered later. THE ARITHMETIC THAT FORCED IT. --text-dim had been re-based twice because the surfaces under it kept coming down,and each time it was 11-15/255 from --text-muted - the second pass's own comment said that was the floor,"as close as a third step can sit to a second". This pass took the surfaces down 26/255,which is 2.5x what the previous one moved. Against the darkest band a dim glyph actually sits on now (#c3c0bb,the menu list's lowest eighth) a 4.5:1 floor allows a token at L 0.0787 - and #57534e,the muted level BELOW it,is already at 0.0876. There is no value left that is dimmer than muted and still clears the floor. Keeping three tokens would have meant shipping two of them 4-6/255 apart:three tokens rendering as two. So they are merged,deliberately,and the token that survives keeps the name --text-muted because that is the level with the call sites. --text-dim is declared as an alias below so no rule that names it has to change. WHAT THE APP LOSES,exactly. Three call sites stop being a level quieter than secondary text:`.tip--none` (the empty-range line in a card),the More menu's external-link glyph,and a completed project step's name. They now read at the muted level - legible and correctly de-emphasised against --text,but no longer the quietest thing on a panel. The rest of the app is unaffected:--text-muted was always the level carrying card titles,table heads,stat labels and the nav. WHAT IT BUYS is not just the merge. A merged token has one job instead of two,so it can be placed for margin instead of pinned at the floor,and the primary/secondary gap does not have to be spent on the third level. rendered,beside the real glyph,worst of each call site a card's tip-table header row #c9c6c1 5.75:1 a table head on account detail #d1ceca 6.24:1 a card title on the dashboard #cfcdc9 6.17:1 a stat label #cac8c4 5.86:1 the menu list at its external glyph #d1cfcb 6.29:1 the rail's nav labels #d5d3d0 6.55:1 the page #cdcbc7 6.04:1 RE-BASED ONCE MORE,FROM #504c47,and not for a failure:a sweep of every text/background pair on twenty-two rendered screens in light mode found nothing under the 4.5 floor,and this token's worst surface at 5.00:1. What it was,was reported as gray on gray text - which at these numbers is a judgment about how it looks rather than a measurement of what it fails,and on a page that is itself gray,secondary text a hair over the floor reads washed out whether or not it clears it. The step is 9/255,which lifts the worst surface to 5.75:1 and the page to 6.04. It costs hierarchy:the gap to --text goes from 2.23:1 to 1.94. That is the trade - --text is unchanged at 11.72:1 on the page,so the two levels are still 1.94 apart and a reader still reads them as two. */ --text-muted:#474340;--text-dim:var(--text-muted);/* The old dim comment,kept as the record of why the merge happened:it was #6b6560,then #666058,each time because the surface under it moved. #666058 measured 4.84:1 on a card face at a tip row,which was 0.34 of margin,and it was 15/255 above --text-muted. */ --track:#bcb9b7;/* Accent is a TEXT token in every rule that names it as a colour,so it moves with the surfaces. It has now been re-based twice for that reason - #2563eb failed at 4.47:1 on a card before the second pass touched anything,and #1d4ed8 measured 5.23 there after it. This pass took the card's face down another 24/255,which put #1d4ed8 at 4.54:1 - passing,but 0.04 of margin,so it came down again. It is the one place in the palette where a fill value and a text value had to part company:--accent-fill stays #2563eb,because it carries white text and the fills do not flip between themes. #1943b9 on the sub-tab strip,beside the glyph 4.90:1 <- the worst #1943b9 on a card's face,beside the glyph 5.19:1 #1943b9 on the page,beside the glyph 5.09:1 The worst reading is the sub-tab strip rather than the card:it is the darkest surface an accent glyph lands on,and it is the number to watch. --accent-hover follows it down to keep the pair apart,which is what makes .link's hover a change at all:.link takes --accent-hover and .link:hover takes --accent,so the hover still lightens by one rung. The pair is held at the same 1.20 luminance ratio it had. */ /* Fourth re-base,and this one was found by an axe sweep rather than by the pass that caused it:the page came down 14/255 under this token in the Daylight pass and #1943b9 landed at 4.21:1 on it - under the floor,on EVERY link in the app,because `a{color:var(--accent)}` is the default. The pass that moved the page re-measured four tokens against it and missed this one;the sweep is what caught it,which is the argument for the sweep. #1a3da8 measures 4.71:1 on the rendered page and 7.24:1 on a panel,and it stays lighter than --accent-hover (5.05:1),so `.link`'s hover still lightens by one rung rather than inverting. */ --accent:#1a3da8;--accent-fill:#2563eb;--accent-hover:#16399f;--accent-subtle:#dbeafe;--on-accent:#ffffff;--on-accent-subtle:#1d4ed8;/* THE BRAND MARKS. atomic8ball.com's own stylesheet carries the palette this comes from:--accent #b2d470 (the lime it draws underlines,borders and list markers with),--light-30 #578303 (the leaf green its headings take) and --dark #42338f (the indigo its primary buttons carry). THE INDIGO IS NOT HERE,and that is a decision rather than an oversight. This app's material language excludes purple by name,and more to the point a control's colour here states what it does to the database - blue reads,green writes,red deletes. A brand fill on a button would say "this one is atomic8ball's" in the same channel that says "this one is safe",so the brand arrives as MARKS:the face of a current selection,the underline under a live tab,the tint behind a chosen chip. Nothing that writes or deletes changes colour. THE LIME IS A FACE,NOT A LINE,on light ground,and that is measured:#b2d470 on the new panel is 1.54:1,which is a fine badge with --mark-ink on it (11.34:1,measured on the rendered chip and the rail's selected pill alike) and an invisible underline. So light draws its lines in the leaf green (4.15:1 on the panel,over the 3:1 a graphical object needs) and takes the lime for faces;dark draws both in the lime,where it measures 11.8:1 on the page. */ --select-fill:#b2d470;--mark-ink:#12100e;--mark:#578303;/* Button fills,by what the button does to the database. Deliberately the same values in both themes:these carry white text,and the dark theme's --success (#22c55e,white on it measures 2.28:1) and --danger (#ef4444,3.76:1) both fail AA as fills. They stay text colours. */ --read-fill:#2563eb;--read-fill-hover:#1d4ed8;/* Rendered is the number that matters,so here is each one beside its token. Measured 2026-09-26 by sampling screenshot pixels of a probe .btn of each kind at 1:1 in three bands,both themes - the fills do not flip,and both themes agree to two decimals:write top 5.26:1 middle 5.71:1 bottom 6.21:1 (lightest px 4.90:1) read top 5.43:1 middle 5.89:1 bottom 6.39:1 (lightest px 5.05:1) delete top 6.85:1 middle 7.35:1 bottom 7.89:1 (lightest px 6.44:1) Every band clears AA,the top band is the binding one,and hover,press and focus-visible all come out stronger still because those fills are darker (write:7.16 / 8.15 / 7.15 at the top). The gap between token and rendered runs the safe way now:the grain multiplies,so the face it paints is *darker* than the token,which keeps the token the conservative number. That is a reversal - `overlay` lightened these mid-tones instead,and on the same probe that put the write top at 4.27:1 with the grain alone and 4.11:1 with the sheen on top,under AA,with the lightest pixel at 3.48:1. The pre-wave grain (0.85 frequency,0.28 opacity) measured 4.54:1 at that band,so the texture as shipped is now stronger than before it existed and more legible with it. If the fill,the grain or the sheen moves again,re-measure:the top band is where it fails first. */ /* WRITE WEARS THE BRAND'S LEAF GREEN. The family was a blue-green - it read as "green" beside the read blue and nothing more. atomic8ball.com's own headings are #578303,and the fill takes that hue:#4f7603 is one step down its own line so white still clears the floor with room (5.35:1 at the token),and the hover is the next step down again. The rendered top band is the reading that fails first here - the sheen and the grain are on it - so it is re-measured with the rest of the pass rather than assumed from the token:white measures 5.29:1 on the rendered top band (#537610),where the blue-green it replaces measured 5.26. */ --write-fill:#4f7603;--write-fill-hover:#436301;--delete-fill:#b91c1c;--delete-fill-hover:#991b1b;--on-kind:#ffffff;/* The label of a key whose write is in flight. NOT a dimmed face:the fill is what says which kind of write is running,and it was the face the old whole-key opacity took away. The ink steps back instead,and the fill stays. White at 55% over these mid-tones is a visible step rather than a disabled-looking grey,which is what the state is - working,not unavailable. */ --on-kind-busy:rgb(255 255 255 / 0.55);/* A ghost's label at rest. Dark overrides all three - these sit on --surface,which inverts,so the fill colours are unreadable there. They are no longer the fill values,and that is a measurement rather than a preference. A ghost paints no face,so its label renders on the PAGE or on a card,which is darker than --surface - the pair --accent-fill/--read-fill is tuned for white text ON the fill,and it was never the right value for these two. Read already measured 4.46:1 on a card before this pass and 4.04 after it;write measured 4.44:1 on the page before and 3.99 after. Both take the accent's own move for read and the next rung of their own ramp for write. Delete is untouched:at #b91c1c it measures 5.14:1 on the new page and it was never near the floor. All three moved again with this pass,for the same reason and by the same measurement:the card's face came down 24/255 under them. Left where they were,read measured 4.54:1,write 4.49 and delete 4.07 on the darkest surface a ghost label lands on (#cac7c3). Each is re-based on its own hue line - value scaled,hue and saturation held - so the three stay blue,green and red rather than drifting to grey. Rendered,beside the real glyph,worst of the surfaces it lands on - and "worst" is the PAGE now,not a card:the Daylight pass took the page below every panel,so a ghost label's hardest ground is the ground itself. Two of the three failed there after that pass and were caught by an axe sweep,not by the pass's own four-token measurement:--btn-hue-read #1a3da8 4.71:1 on the page (#2563eb -> #1d4ed8 -> #1943b9 -> here) --btn-hue-write #365003 4.65:1 on the page (the blue-green -> the brand's leaf green) --btn-hue-delete #8f1414 4.70:1 on the page (#b91c1c -> #9f1818 -> here) Read and delete take the values --accent and --danger-text were just moved to:the three are the same colours as the text tokens beside them by design,and keeping a second,failing copy of each would be the trap this file exists to avoid. On a panel all three clear 7:1. */ --btn-hue-read:#1a3da8;--btn-hue-write:#365003;--btn-hue-delete:#8f1414;/* Ghost hover. The fill tokens are wrong here:a ghost is outline-only,and filling it would put its own hue behind its own hue. These are the pairs a ghost hovers to instead. Read reuses the shipped accent pair. */ --read-subtle:#dbeafe;--on-read-subtle:#1d4ed8;--write-subtle:#eef5dc;--on-write-subtle:#3f5c05;--delete-subtle:#fee2e2;--on-delete-subtle:#b91c1c;/* CE Pending's date groups. Six fills a repeated scheduled date cycles through,so blasts going out the same day read as one band. Same values in both themes and not restated in the dark block,for the reason the kind fills above are not:THESE CARRY WHITE TEXT,and white is what the pair is tuned for. A dark override would break that. They are fills rather than pale tints,and that is measured rather than chosen. A report cell paints --surface,which here is #d5d3d0 - a warm mid-grey,not white - so a pastel chip lands at about 1.05:1 against the row it sits on and disappears. These are the values that clear three ratios at once,and every one of them was computed,not picked:slot fill white on it vs light row vs dark row --group-1 #2563eb 5.17:1 3.46:1 3.38:1 --group-2 #0f766e 5.47:1 3.66:1 3.20:1 --group-3 #b45309 5.02:1 3.36:1 3.48:1 --group-4 #e11d48 4.70:1 3.14:1 3.72:1 --group-5 #7c3aed 5.70:1 3.81:1 3.07:1 --group-6 #4d7c0f 4.99:1 3.34:1 3.50:1 Hue is what separates one slot from the next;the luminances are close on purpose,because a set that varied in lightness would read as a ranking - slot 1 looking heavier than slot 6 when the only difference is which date it is. Six is what fits before the wheel comes back to a neighbour,and the slots are walked in order so two similar hues are never adjacent. */ --group-1:#2563eb;--group-2:#0f766e;--group-3:#b45309;--group-4:#e11d48;--group-5:#7c3aed;--group-6:#4d7c0f;--on-group:#ffffff;/* The bevel. A light edge on top and a dark one underneath is what makes the surface read as a physical key rather than a rectangle of colour - and it is now the thing that carries a control's depth,because the shadow under it stopped being a drop (see --shade). Deepened from 22/20 to 26/24,and measured on the rendered face rather than chosen. Down a read key's centre column,against a mid-fill of 87/255:the lit top line goes from 131 to 141 (+44 to +54 over that fill) and the shaded bottom line from 64 to 52 (-23 to -35),with a blurred 4px inset under it so the last 2px of face ramp from 79 to 70 into the line instead of meeting it flat. That ramp is the part that matters - it is the difference between a surface curving away and a line drawn across it - and it is why the inset is blurred rather than a second hard row. A wider,brighter pair was tried and rejected at 3x:at 30% plus a 5px blurred highlight the top line measured 172,+85 over the fill and across a 7px ramp,which is a white smudge on the label's own row rather than an edge. */ --bevel-light:rgb(255 255 255 / 0.26);--bevel-dark:rgb(0 0 0 / 0.24);--bevel-press:rgb(0 0 0 / 0.30);/* The blurred falloff under the bottom line,and it is a LIGHTER value than the line for a measured reason:it is 3-4px of face,so it is what the bottom band of a --surface control averages to,and --text-dim is read against that band. Tied to --bevel-dark's 24% the band came out #e2e2e2 and --text-dim measured 4.46:1 there - under the 4.5 floor,and down from the 4.70:1 the same band measured before this pass. At 14% the band is #e4e4e4 and it clears. Those two bands were measured on --surface at #ffffff and the surface has now moved three times,to #d5d3d0. The band is measured on the INTERIOR of a boxed --surface control - inset past its 1px border,which is what made an earlier reading of this note look catastrophic:including the border drags the band to the border's own colour. As it stands,with the merged token,the band measures 4.71:1 on a neutral key,4.70:1 on a pager key,4.43:1 on a header or timer key and 4.36:1 on the credential remove - the last two under the 4.5 floor,and this is the one bucket in the palette where a text token does not clear it. Nothing renders dim or muted text on any of those bands today:the four are the blurred bottom 4px of a key,where no centred label reaches. The merge improved them rather than hurting them - the previous pass measured the same bands at 2.64-3.67,because the token was lighter then - but they are still the first place a new muted glyph on a control face would fail,and it would fail by 0.14. The 1px line keeps the deeper value,because a line is not a band. */ --bevel-shade:rgb(0 0 0 / 0.14);/* A hovered kind lifts its top edge. Its own token because the value is a function of the surface:34% disappears against #1c1917,so dark halves it. */ --bevel-light-hover:rgb(255 255 255 / 0.38);/* The control's edge,as one token list rather than a line repeated in six rules. It was written out verbatim in `.btn`,in each of the three kinds and again in the kind hover - five copies of the same two insets - which is the shape a material drifts apart in. The list is the whole edge:a lit top,a shaded bottom,and the blurred falloff under it. THE CHAMFER WAS TRIED AND REMOVED. For a day this list carried two more insets - `inset 1px 1px 0` and `inset -1px -1px 0` - lighting the key's left edge and shading its right,a chamfer rather than a rim. It is gone,and the reason is that nothing could see it:captured at 6x - the only scale it reads at - the difference is one pale pixel down one side,and at 1x it is not there at all. No contrast requirement rested on it,and it was added for symmetry's sake one day after the press pass,which is the weakest reason in that whole change. Re-adding it needs a reason better than "the edge could be finished". `-hover` is the same edge with the top line lifted. */ --btn-bevel:inset 0 1px 0 var(--bevel-light),inset 0 -1px 0 var(--bevel-dark),inset 0 -4px 4px -3px var(--bevel-shade);--btn-bevel-hover:inset 0 1px 0 var(--bevel-light-hover),inset 0 -1px 0 var(--bevel-dark),inset 0 -4px 4px -3px var(--bevel-shade);/* THE PRESS. A key that is held meets the surface rather than nudging down 1px:the lift collapses,the blur under the top line goes from 2px to 4 with the line's own alpha,and a hairline of light comes back along the bottom edge - which is what a compressed key shows,because its lower bezel is now facing the light. `--press-dim` takes the whole face down 4%,the last bit of the effect and the only part of it that is a percentage rather than a shape. */ --bevel-light-press:rgb(255 255 255 / 0.06);--btn-press:inset 0 4px 7px -2px var(--bevel-press),inset 0 -1px 0 var(--bevel-light-press);--press-dim:brightness(0.96);/* The rest of the button's surface:a two-stop sheen that reads as curvature. Tokens for the same reason the bevel is - a rule must not carry a colour literal - and the sheen's top stop is the one number here bounded by measurement rather than by taste:see the write fill below for what it costs the rendered face. Declared once,in light,and deliberately not overridden in dark. The bevels flip because 26% white smears on #1c1917;this does not need to,because the fills it sits on do not flip either,and the neutral base measures 11.8:1 of label contrast at its lightest band in dark - nothing to fix. */ --sheen-top:rgb(255 255 255 / 0.055);--sheen-bottom:rgb(0 0 0 / 0.10);/* The prominent key's face:the same two-stop sheen at three times the depth,which is what makes it read as a bigger radius rather than a brighter one. It is a VEIL over the kind's fill and not a gradient of its own colour - `--read-fill-hover` still has to show through it,or the family's hover would be dead - and that is also why one token serves all three kinds:the hue is the fill's,and this only bends the light across it. AND THE DEPTH IS IN THE BOTTOM,NOT THE TOP,which is a measurement and not a preference. The first version of this put 16% white at the top - three times --sheen-top,which is what "three times the depth" sounds like - and it broke the same rule --sheen-top itself is bounded by:the write key's top band measured 3.83:1 in light and 3.35:1 in dark,both under the 4.5 floor,because a green fill that carries white ink has almost no headroom to be lifted at all. The veil therefore darkens downward instead:8% white at the top,which is within a point of where the plain key already sits (measured 4.55:1 modelled,and the grain takes it back over 5 rendered),and 0.24 of black at the bottom against the plain sheen's 0.10. The dome reads the same or stronger and the ink keeps its floor - the darkening moves contrast the safe way,exactly as the grain's multiply does. */ --prominent-veil:linear-gradient(180deg,rgb(255 255 255 / 0.08) 0%,rgb(255 255 255 / 0) 45%,rgb(0 0 0 / 0.24) 100%);/* The prominent key's ambient,and it is the one place a fill colour is used outside a face. It is the app's own depth vocabulary at a bigger scale - the same negative-spread ambient `--shade` and the panels use - tinted with the fill it belongs to,so the key reads as resting in its own light rather than as glowing. One per kind,because a read key's ambient in green would say the wrong thing about what pressing it does. */ --halo-read:rgb(37 99 235 / 0.45);--halo-write:rgb(79 118 3 / 0.5);--halo-delete:rgb(185 28 28 / 0.45);--prominent-cast:0 6px 16px -6px;/* The control's own depth,and it is SHADING rather than a cast - which is a measurement,not a word for the same thing. What it replaces:`--cast:0 1px 2px rgb(0 0 0 / 0.18)`. Short offset,tight blur,high alpha - a contact shadow. Down a button's centre column the page under the key read 250/255 and the first pixel below its bottom edge read 215,so a 35/255 drop arrived at once;beside the key there was nothing at all,above it 6/255,and the whole thing was gone within 2.3px. Direction plus a step is the whole definition of something hovering over the page,and it was the wrong shape besides:sat next to a panel whose ambient is wide,soft and negative-spread,the key read as pressed ONTO the page while the panel read as resting IN it. What replaces it is the panel ambient's shape at a control's scale:no offset,negative spread,low alpha,so the page darkens AROUND the box rather than a shadow falling from it. Measured on the same column:12/255 peak below the key and 13/255 above - asymmetry -1,i.e. the same amount all round - a largest single-pixel step anywhere outside the box of 1/255 where the old one stepped 6,and a 5.7px reach against 2.3. There is no step and no direction left for the eye to read as a drop,so the depth has to come from somewhere else,and that is the bevel:--btn-bevel is deepened to carry it,and the two were solved together rather than one compensating for the other. The test it has to pass is the one the surface language sets for itself - does this read as a surface with depth,or as an object hovering over the page - and it passes it in the render:the key's four sides darken by the same amount,so nothing points anywhere. Dark takes its own value. The dark page is #0c0a09,where 16% black moves the mean by 3/255 and reads as nothing;the panel's own ambient goes from .20 to .75 between the themes for the same reason,and this follows it. */ --shade:0 0 10px -2px rgb(0 0 0 / 0.16);/* === The control material:hit area === Every control's smallest side,and it is worth a token for the same reason the radius is:a control that is 26px on one screen and 17px on the next is the same object being two sizes. 24px is WCAG 2.2 AA's target size (2.5.8);44px is the touch floor,and it is applied from the one media query that already knew about it rather than from each family. */ --hit:24px;--hit-touch:44px;/* The focus ring. `currentColor` read white on a filled kind,so the ring drew white on the page - 1.04:1,invisible. --text is dark on light and light on dark,and the ring sits outside the button,on the page. */ --focus-ring:var(--text);/* Fractal noise,desaturated. The feColorMatrix is load-bearing:feTurbulence emits *coloured* noise,and without it every button is tinted rather than textured. Shipped inline so it costs no request. baseFrequency 0.55 on a 160px tile puts the features at roughly 4-6px:past the 1-2px of 0.85,which averaged away to flat fill on a 71px control and measured as invisible,and short of the 8-12px of 0.30,which was large enough to read as blotches - a visible dark patch on the green fill rather than a material. The count did not catch that:76 distinct colours across a 46x5 strip is "plenty of texture" by the number and still looked like a stain,because a count measures variation and not scale. Opacity stays at 0.45;it was solved against the rendered face,not against the frequency. THE RULE,and it is not obvious:the grain belongs on a control whose surface is the kind fill. On --surface it is not texture,it is grime. The same noise at the same opacity reads as a material on a saturated fill and as soiling on white - measured,the header's Edit and Quick Log keys carry --surface,and with the grain on them they read as dirty white keys beside the clean Payment anchor next to them,which has no .btn class and so no grain at all. Three rules that paint --surface reset this image with the `background` shorthand and that reset is correct,not a bug:see .detail-header__action-btn,.ce-cred__remove and .timer__copy. The same applies to every box with no surface of its own - .icon-btn,.sub-tab,.back-button,.menu-group__item all declare `background:none` for the same reason,and restoring the image there paints a grey mottled square around a bare glyph. If a control's surface is white and you are about to give it the grain,the control is what needs changing,not this rule. */ --grain:url("data:image/svg+xml,%3Csvg xmlns='http://www.w3.org/2000/svg' width='160' height='160'%3E%3Cfilter id='g'%3E%3CfeTurbulence type='fractalNoise' baseFrequency='0.55' numOctaves='3' stitchTiles='stitch'/%3E%3CfeColorMatrix type='saturate' values='0'/%3E%3C/filter%3E%3Crect width='160' height='160' filter='url(%23g)' opacity='0.45'/%3E%3C/svg%3E");--nav-text:#504c47;--danger:#dc2626;--danger-text:#8f1414;--warn:#b45309;/* Warn as words,the third of the three. #b45309 fails the way the other two do and by the same amount:on rendered pixels it is 4.29:1 on the page a dashboard stat or the instructions panel sits on,and 4.26:1 in the timer's card,against a 4.5 floor. #9b4808 renders at 5.40:1 and 5.36:1 on those same two crops - sampled off the pixels,not computed from the tokens - and at 5.31:1 against #edebe7,the darkest surface in this palette - the card foot --danger-text renders on,where warn never lands. Each figure is the modal backdrop of the element's own crop,so grain is in it. It is the FILL darkened and nothing else:hue 25.97 and saturation 90% are #b45309's to the digit,only the lightness moves (37% -> 32%). Warn is the one of the three whose identity is genuinely at risk from this pass,because amber is what turns brown first,so the value was taken off the fill's own hue line rather than from a ramp:#78350f,the amber-900 rung,measures 7.62:1 and drops the hue 4.3 degrees to read as brown,and #92400e measures 5.96:1 - headroom this palette has no surface to spend. The floor is real rather than cautious. #aa4e08,one lightness step up from this one,measures 4.64:1:it passes,and it is 0.14 above the line,which is the trap --text-dim documents - a token inside 0.4 of the floor pins every surface it renders on,and these surfaces moved once already. 0.81 puts warn inside 0.12 of --danger-text (5.43:1 worst) instead of making it the token the next surface pass is bounded by. That prediction came true and this pass paid it. The card came down 24/255 under it and #9b4808 measured 4.62:1 on the new card,inside the 0.4 band the paragraph above calls a trap,so it moved. It moved the same way as before - its own hue line,value scaled,hue and saturation held at 25.97 and 90% - and it is #833d07 now,at 4.73:1 worst beside the real glyph. It stays a clear step inside --danger-text (4.76) rather than outside it. */ --warn-text:#743405;--success:#15803d;/* Success as words,the way red already has one. The account header's type badge paints this colour on a 15% tint of itself,and it measured 3.52:1 there - #15803d on the #cddcd1 the tint renders as,under the 4.5 floor and the same defect --danger-text exists for. #166534 fixed it at 5.01:1. THE CLAIM THAT IT WAS STABLE WAS WRONG,and the surface pass is what showed it. The tint is `color-mix(...,15%,transparent)` - transparent,not a solid - so it composites over whatever is behind the badge and moves with the surface after all. Taking a card down 10/255 took the badge label from 5.01 to 4.67 on the probe face and 4.55 on the real header,which is 0.05 above the floor and not a margin. It is now #14532d,the rung the previous comment named and rejected:5.97:1 on the probe face and 5.80:1 beside the real glyph,which restores the half a point this token was created with. The objection to it was that a label far below its own tint stops reading as the tint's colour - but the tint came down 10/255 with the card,so the label coming down with it keeps the relationship rather than breaking it. The 7.13:1 this measures on white is not the number that matters here. It moved once more with this pass,for the third time and by the same mechanism:the badge's tint is a 15% color-mix over the card,so it fell with the card again (5.80 -> 5.04) and came down its own hue line to #134d2a,which restores it to 5.04:1 beside the real glyph on the darker edge of the tint and 5.15:1 on the lighter. Still the same rung of green. */ --success-text:#134d2a;/* THE RAIL'S STATUS LIGHT IS AN INDICATOR,AND IT IS NOT A TEXT TOKEN. The AI glyph at the rail's foot says up,down or undecided by its colour,and the point of it is to be SEEN from across the room - so it takes the fill strength of each hue rather than the text strength,and it has its own tokens so that the next time --danger-text or --success-text is re-based for a 4.5:1 text ground,the light does not dim with them. It did once:the Daylight pass darkened --danger-text and took the outage light with it. What binds a mark is 3:1,not 4.5,and on the light rail's face (#d5d3d0) these are the vivid ends that clear it:#15803d at 3.10:1 and #b91c1c at 3.99:1,against the 6.1:1 and 4.0:1 of the darker text tones they replace. A luminous green (#16a34a measures 2.03:1 there) is not available on a light rail at all - the ground decides how bright "bright" can be. */ --status-up:#15803d;--status-down:#b91c1c;--shadow-md:0 1px 3px rgb(0 0 0 / 0.12);/* === Surface and material language === Four levels of shading,each a token set so no rule carries a literal. Level 0 is the page and gets nothing. Level 1 is a panel. Level 2 is a control,and the button's material is the control's - see .btn. Level 3 is an overlay,which is a panel that also has to sit above the page. The shape of every level is the same three ideas,and each was chosen against a render rather than by taste:- the rim is a THREE-stop vertical gradient,not a two-stop wash. A two-stop gradient over a 300px card puts a veil on the whole face;three stops put the curvature at the edges and leave the middle clean. - the bevel is 2-4px and BLURRED. A hard `inset 0 3px 0` reads as a painted stripe;a blurred one reads as a rounded edge catching light. The top is lit,the bottom is shaded,so the panel reads as a surface that is in the page and not hovering above it. That is the whole test:altitude shows as a shadow under a box with an even face,curvature shows as a light top and a dark bottom. - the ambient is wide,low-opacity and NEGATIVE-spread,so it darkens around the panel rather than casting from it. */ /* This gradient is the bright thing on screen. It is the PANEL that is the whitest object in the app,not --surface:the face is this gradient,it is opaque,so the background-color under it is never visible (see .card) - a light pass that moved --surface alone would change nothing anybody could see. Every surface pass therefore has to spend its brightness here. It goes UP this pass,which no previous pass did. The panel is now the brightest thing the app owns and the page is the darkest,and the whole hierarchy is that one step. The reason it could not be done before is in the previous version of this comment:while the page,the rail and the panel all sat inside a 26/255 band,lifting the panel alone would have put it above the rail and every control face. --surface keeps its value here,so the rail stays the middle level and the panel is free to be the top one. THE CONSTRAINT REVERSED AND THAT IS WORTH KNOWING. It used to bind against --text-muted on the darkest band of the panel - the darkest eighth of a menu list rendered #c3c0bb and the merged token cleared 4.5:1 there by 0.18. On a near-white face that reading has room to spare,so what binds here is the other direction:grain and the bevel's bottom shading that read as material at #d6d4d0 can read as dirt on white,and the row rules have to stay visible against a face this light. Both were checked on the render,not chosen. The gradient's span is 10/255 top to bottom,the same order as the 9/255 span it replaces,so the panel still reads as curvature and not as a flat box. */ --panel-rim:linear-gradient(180deg,#ffffff 0%,#fbfaf7 42%,#f5f3ef 100%);/* The bottom shade comes down to a whisper:at 0.13 black on a near-white face the shading that read as a rounded edge at #d6d4d0 read as a grey smudge along the bottom of every card. The top light and the box's own shadow carry the curvature now. */ --panel-bevel:inset 0 1px 0 rgb(255 255 255 / 1),inset 0 -4px 7px -3px rgb(0 0 0 / 0.05),inset 0 0 0 1px rgb(255 255 255 / 0.55);/* Up,because the panel is now the brightest thing on a darker ground:the altitude between the two is what reads as a sheet rather than a rectangle. */ --panel-lift:0 20px 44px -28px rgb(0 0 0 / 0.45),0 2px 5px -3px rgb(0 0 0 / 0.12);/* Level 3 is level 1 plus this. There is no level-1 shadow only on the modal because a modal is the one surface that genuinely is above the page;it is still a soft wide one rather than a tight drop,which is the shape that was making the old modal read as hovering. */ --overlay-lift:0 24px 56px -24px rgb(0 0 0 / 0.30),0 6px 16px -8px rgb(0 0 0 / 0.10);/* Level 2's field. A control's interior IS the surface - measured,the well's fill is one distinct colour across candidates - so only the edges shade:a blurred top recess and a lit bottom line,which is what reads as a field cut into a panel rather than a rectangle drawn on it. */ --well:inset 0 3px 5px -2px rgb(0 0 0 / 0.13),inset 0 -2px 0 -1px rgb(255 255 255 / 1);/* Grain,per surface size. One frequency,four opacities,and the frequency is the number that matters:0.55 is the shipped control grain and reads as blotches on a large face,0.85 was too fine to survive a control,and 1.25 on a panel is even paper tooth. Measured at 3x;at 1x this is sub-pixel and reads as a slightly darker panel and nothing more. */ --grain-panel:url("data:image/svg+xml,%3Csvg xmlns='http://www.w3.org/2000/svg' width='160' height='160'%3E%3Cfilter id='g'%3E%3CfeTurbulence type='fractalNoise' baseFrequency='1.25' numOctaves='3' stitchTiles='stitch'/%3E%3CfeColorMatrix type='saturate' values='0'/%3E%3C/filter%3E%3Crect width='160' height='160' filter='url(%23g)' opacity='0.18'/%3E%3C/svg%3E");--grain-table:url("data:image/svg+xml,%3Csvg xmlns='http://www.w3.org/2000/svg' width='160' height='160'%3E%3Cfilter id='g'%3E%3CfeTurbulence type='fractalNoise' baseFrequency='1.25' numOctaves='3' stitchTiles='stitch'/%3E%3CfeColorMatrix type='saturate' values='0'/%3E%3C/filter%3E%3Crect width='160' height='160' filter='url(%23g)' opacity='0.12'/%3E%3C/svg%3E");--grain-menu:url("data:image/svg+xml,%3Csvg xmlns='http://www.w3.org/2000/svg' width='160' height='160'%3E%3Cfilter id='g'%3E%3CfeTurbulence type='fractalNoise' baseFrequency='1.25' numOctaves='3' stitchTiles='stitch'/%3E%3CfeColorMatrix type='saturate' values='0'/%3E%3C/filter%3E%3Crect width='160' height='160' filter='url(%23g)' opacity='0.14'/%3E%3C/svg%3E");--grain-page:url("data:image/svg+xml,%3Csvg xmlns='http://www.w3.org/2000/svg' width='160' height='160'%3E%3Cfilter id='g'%3E%3CfeTurbulence type='fractalNoise' baseFrequency='1.60' numOctaves='3' stitchTiles='stitch'/%3E%3CfeColorMatrix type='saturate' values='0'/%3E%3C/filter%3E%3Crect width='160' height='160' filter='url(%23g)' opacity='0.12'/%3E%3C/svg%3E");/* A placeholder that renders inside a panel,so it is a surface and it moves with the panels. It is not a text token and no ratio governs it - it is here because leaving it would have inverted it:at #e7e5e4 on a card the second pass moved the card's face to ~#e6e3de,so the base would have come out LIGHTER than the thing it is drawn on and the loading state would have read as an empty card. THAT INVERSION IS EXACTLY WHAT THIS PASS WOULD HAVE CAUSED,in the other direction:the card went up 25/255 and these were left where three dimming passes had put them,which would have made every loading state a grey slab on a white panel. They move with the card,as they always have. What the shimmer needs is the DELTA between the two,which is held at 17/255;the base sits 21/255 under the gradient's middle stop,close to the 16 it held before. */ --skeleton-base:#e6e4e1;--skeleton-shine:#f7f6f3;--backdrop:rgb(0 0 0 / 0.5);--space-1:4px;--space-2:8px;--space-3:12px;--space-4:16px;--space-5:20px;--space-6:24px;--space-8:32px;/* A container sits one step softer than the controls on it,and no further. The whole scale is a two-family system with a 4px gap between the families:6 is every control and 4 is the small parts inside one,8/10 are boxes. 10px on a container against the 6px control is the step that reads as a second family at 3x without going pill-shaped,and it is the outermost number the app uses - nothing is rounder than 10px. 4px is badges and skeletons now,not controls. It used to be chips too,and the chip took 20px instead - a pill,rounder than every container in the app and therefore the roundest thing on any screen,which is the hierarchy upside down. The chip is --radius-md with everything else that is pressed,and the badge is what is left at 4px because nothing presses it. --radius-lg is the box that sits INSIDE a panel. It is deliberately only 2px inside --radius-xl rather than equal to it:two nested boxes at the same radius read as one box with a line drawn on it,and a nested box rounder than its container is the shape that looks like a mistake. */ --radius-sm:4px;--radius-md:6px;--radius-lg:8px;--radius-xl:10px;/* Measured,not chosen:the phone bar is 53px of tab plus the 1px top border,and the tabs' own box model is what sets that (6px/4px/4px padding under a 28px icon chip and a 10px label). It was 48px while the bar was 54px,so body's reserved padding came up 6px short and the last line of a page sat behind the bar. */ --tab-height:54px;/* Data pages share one container width so tables and lists line up page to page. The login and empty states set their own narrower text measure rather than reading one from here. */ --page-max:1120px;/* How much of a URL column a report shows before it cuts. In `ch`,so it scales with the table's own font size rather than pinning a pixel width to one report's typography. 30ch is measured,not chosen:cepending is the widest of these tables at six columns,and 30ch is the point where Client + Email + Blast + URL + Scheduled + Size,with their padding,come to the content column. Uncapped,the URL column alone measured 859px and the table ran past the page. */ --link-max:30ch;/* How wide a cell of running text gets before it wraps. The pull preview's Description is the only one:task descriptions are paragraphs,and a paragraph has no natural width to shrink to. Chosen for the measure rather than fitted to a data sample,which is the difference between this and --link-max above. 40ch is a short reading line rather than a wide one on purpose - this column sits last on a table whose other six columns are a date,an id and four short names,and a description given half the width would leave the identifying columns squeezed. */ --prose-max:40ch;/* The task row's two action buttons,sized once so they are one pair rather than two controls that happen to sit next to each other. Inline is the width "Delete?" needs at --text-xs,which is the widest the pair ever gets - Delete reserves its word in both states,see .task-row__btn--confirm. */ --row-btn-inline:5rem;--row-btn-block:1.75rem;} /* === The phone type scale === Every size in the app comes from these four tokens,so a phone changes THEM rather than restating a size at each rule. That is what a token scale is for and it is why there is nowhere else for a size to be written. Measured off this stylesheet:`--text-sm` (14px) is the body size and carries 61 rules - table cells,log notes,list rows,form values,button labels - and `--text-xs` (12px) carries 49,and those 49 are not all labels. `.tip-table__when`,`.menu-item__desc`,`.task-row__meta` and `.contact-card__dept` are things a person reads,and they were 12px. On a phone held at arm's length 14px body text is a strain,and the platform baselines agree with the complaint:iOS body text is 17px and Android's is 16px. So on a phone the body size is 16px and the small size is 13px. Nothing below is re-declared - every rule already says `var(--text-sm)` and simply gets the phone value,including rules added later by someone who never reads this block. `--leading` travels with the size. 1.4 is a desktop measure and 1.5 is where running text sits;a size change without its leading is half a change,and on a narrow column the leading is what stops the eye losing its place on the way back to the left margin. `--text-base` is not raised. It is already 16px,it is what `html`/`body` read from,and 16px is the value the iOS zoom guard needs on a focused field - moving it would take the guard with it. The page's own `--text-2xs` stays at 12px:it has one caller,a hint under a form field,and it is the floor rather than a size anything reads at length. */ @media (max-width:560px){:root{--text-xs:13px;--text-sm:16px;--leading:1.5;}} /* Dark repeats every colour token rather than overriding a few,so a themed page can never inherit a leftover light value. The selector carries both the attribute and :root so it outranks the :root block above on specificity and not merely on source order. Loaded from localStorage,applied in app.ml,and toggled from the More menu. */ :root[data-theme="dark"]{color-scheme:dark;/* One notch deeper this pass. Dark does NOT get the light pass's split:its panel already sits a clear step above its page (--panel-rim's top stop is #2c2724 against a #0c0a09 page),so the hierarchy light had to build is already here. All this does is widen that step a little,and it is the only surface value dark moves. --panel-rim is deliberately untouched:its lightest band is what --text-muted is measured against (5.18:1),and lighting that band would spend the ratio to buy nothing. Measured after the move:page #090807,rail #1c1917,panel top band #35312f,middle #1f1b19,bottom #171413. The panel's step above the rail is its top band rather than its whole face,which is the shape dark has always had and this pass did not change. */ --bg:#090807;--surface:#1c1917;/* hover was #292524,which was also the divider colour - so hovering a table row erased its own rule. The four greys are now four distinct steps. */ --surface-hover:#262220;/* Dark keeps the field on --surface,and the light-side move is deliberately not mirrored:on a dark panel the field is already darker than its ground,which is the well,and its text measures 11.74:1 against it. Lifting it here would put the field ABOVE the panel in dark and turn the well into a plate. */ --field-face:var(--surface);--border:rgb(255 255 255 / 0.11);--border-strong:#7a746c;--border-light:rgb(255 255 255 / 0.075);--th-band:rgb(255 255 255 / 0.05);--text:#e7e5e4;/* Lifted from #a8a29e,which was a real step and still read thin everywhere it landed. It measured 6.93:1 on --surface,5.85:1 at the top band of the panel rim,and 4.56:1 in the tightest place it goes - --accent-subtle,which is over the 4.5 floor by 0.06 and had no margin left for a surface to move under it. #d6d3d1 measures 11.74:1 on --surface,9.91:1 at the rim's top band and 7.72:1 on --accent-subtle. It is not a new invention:--nav-text already ships this exact value as the rail's dark label colour,so the two levels meet rather than one being invented to match the other. What it costs,stated because it is real:dark's muted level is now 1.19:1 below --text,where light's is 2.25:1. The two levels sit closer together here than they do in light,and what separates them on screen is size and weight rather than luminance. If muted text starts reading as full text on a screen where both appear - a report table is the place to look - the step back to is about #cfcac6,which is 10.76:1 and 7.08:1 at those worst bands. Dark only. Light is untouched at 5.70:1 on --surface. That 5.70 was against the page this pass moved. The muted token renders on the new page at 4.99:1 - sampled off the render,grain included - which is still clear of the 4.5 floor but is the light side's tightest reading now,and the number to watch the next time a surface under it moves. On the panel it went the other way and has room to spare:9.00:1 at the middle band,8.68:1 at the darkest. */ --text-muted:#d6d3d1;/* #8a847e measured 4.73:1 on --surface here,and the panel's rim gradient drops that to 4.05:1 - under the floor,on the plain panel,before any grain. #9e978f measures 6.06:1 on --surface and 5.18:1 at the worst band of the treated card. Lightened rather than darkened,and that direction is not a preference:on a dark surface every added layer takes luminance away. */ --text-dim:#9e978f;--track:#6b665e;--accent:#3b82f6;--accent-fill:#2563eb;--accent-hover:#60a5fa;--accent-subtle:#1e3a5f;--on-accent:#ffffff;/* Lighter than --accent-hover (#60a5fa),which on this tint measured 4.52:1 - less legible than the unselected labels beside it,so the selected tab read as the dimmest thing in the nav. */ --on-accent-subtle:#93c5fd;/* THE BRAND MARKS ARE THE SITE'S VIOLET IN DARK,and the lime in light. That is the split asked for,and it is the one that makes sense of a palette with two leads:the lime is what reads as bright on a light ground,the violet is what reads on a dark one. The indigo the site's own buttons carry (#42338f) is too dark to be a face here - it measures 1.65:1 against this panel,and a selected chip has to look selected - so the violet one rung up is the face and a lighter violet is the line:face #5b46c9 white ink 6.64:1,and 2.48:1 against the panel it sits on line #8b7ae8 4.75:1 on the panel,5.67:1 on the page - over the 3:1 a graphical object needs,which the face's violet is not,so the underline does not take the face value. The face moved one step deeper than the site's --dark-accent,which is where it started:#654edb beside the read blues read as a second blue rather than as the other lead,and the deeper value is the site's own direction anyway (its primary is #42338f,which is too dark to be a face here at 1.65:1 against this panel). Four values were rendered on the screens and compared before this one was chosen;the cost is that the face sits 0.4 closer to the panel than #654edb did,and the white label is what carries it. The ink flips with the face:near-black on the lime,white on the violet. Everything that WRITES stays green in both themes - the read/write/delete rule is what a control's colour states,and the brand does not outrank it. */ --select-fill:#5b46c9;--mark-ink:#ffffff;--mark:#8b7ae8;/* The fills do not flip - see the light block. Only the bevel does:a 26% white highlight is far too hot against #1c1917 and reads as a smear. Deepened alongside light for the same reason it was:with the cast gone the bevel is what carries the key's depth,and on a dark page it is the only thing that can - the ambient below moves the mean by a few 255ths at most. */ --bevel-light:rgb(255 255 255 / 0.12);--bevel-dark:rgb(0 0 0 / 0.42);--bevel-press:rgb(0 0 0 / 0.48);--bevel-light-hover:rgb(255 255 255 / 0.20);--bevel-shade:rgb(0 0 0 / 0.34);/* The control ambient,dark. Nine times the light alpha rather than scaled by taste:on #0c0a09 a 16% black measures a 3/255 delta and is not there. The panel's own ambient makes the same move in the same direction (.20 -> .75),and at .55 this measures 11/255 below a key and 10/255 above - still directionless,which is the property that matters. */ --shade:0 0 10px -2px rgb(0 0 0 / 0.55);/* The ghost labels,which are the one place a kind's colour is its own text rather than its fill. The light values are fill colours and read against #1c1917 at 3.38:1 / 3.49:1 / 2.70:1 - all under AA - so they are overridden here rather than inherited. */ /* Lifted a rung,and the sweep is why:#3b82f6 measured 4.46:1 on the ground a ghost label actually lands on in this theme (a table row inside a panel,#241e1e),and it computes to 3.50:1 against the light band at the top of --panel-rim,which is where a first-row label sits. #60a5fa measures 5.06:1 on that band and 6.69:1 on --surface. It is the value --accent-hover already carries in this theme,which is the same relationship light has between its ghost read and its accent. */ --btn-hue-read:#60a5fa;--btn-hue-write:#9dc25a;--btn-hue-delete:#f87171;/* Ghost hover,dark. Read is the shipped accent pair;write and delete are built the same way - a deep tinted well with a light label. */ --read-subtle:#1e3a5f;--on-read-subtle:#93c5fd;--write-subtle:#1e2a0b;--on-write-subtle:#b6cf7d;--delete-subtle:#3b1414;--on-delete-subtle:#fca5a5;--nav-text:#d6d3d1;--danger:#ef4444;--danger-text:#f87171;--warn:#f59e0b;/* Restated for the same reason --success-text is,and with the same non-decision:#9b4808 is a light-theme value and would inherit onto this theme,where it renders at 3.15:1 on the page and 2.80:1 in the timer's card - under the floor,and worse than the defect this sweep set out to fix,which is the reason --success-text's comment below gives. Measured with the line deleted rather than reasoned about. #f59e0b needs no help here:it measures 8.22:1 on the worst surface it renders on and 7.78:1 anywhere in this palette,so this token carries the fill's own value and the dark theme renders exactly as it did before the sweep,which is verified rather than asserted. */ --warn-text:#f59e0b;--success:#22c55e;/* Dark has to restate it or the light value inherits,and #166534 on this theme's tint measures 2.27:1 - lower than anything else in the file,the badge's own defect made three times worse. It needs no colour of its own:#22c55e is already 7.11:1 on the badge face here,which is the same pair the light theme fails at,so this token exists only to stop the inheritance and the dark badge renders exactly as it did. */ --success-text:#22c55e;/* The rail's status light,dark side. It already had the bright end here,so these two values are what it has always shown in this theme;they are named so that light and dark agree on what the light IS,and so a future re-base of the text tokens cannot reach it. On the dark rail's face (#1c1917) the green measures 7.46:1 and the red 5.10:1. */ --status-up:#22c55e;--status-down:#f87171;--shadow-md:0 1px 3px rgb(0 0 0 / 0.4);/* The material language,dark. Same four levels and the same three ideas as the light block;what changes is which of them carries the depth. EVERY dark surface gets NO GRAIN. It is a measured no-op:on #1c1917 a multiply grain moved the mean by 1-2/255 and two candidates were indistinguishable at 3x. The only blends that lighten a dark surface are `screen` and `overlay`,and both read as static rather than as paper - there is no dark-paper metaphor for them to land on - and `screen` at even 8% takes --text-dim to 4.28:1. So the bevel and the sheen do all of it here,and the 3x crop says that is enough:the top bevel highlight reads clearly against #211d1b. The tokens are set to `none` rather than left out,so a themed page cannot inherit the light grain. */ --panel-rim:linear-gradient(180deg,#2c2724 0%,#211d1b 42%,#171413 100%);--panel-bevel:inset 0 2px 4px -1px rgb(255 255 255 / 0.115),inset 0 -4px 7px -3px rgb(0 0 0 / 0.58);--panel-lift:0 16px 36px -18px rgb(0 0 0 / 0.75),0 4px 12px -6px rgb(0 0 0 / 0.40);--overlay-lift:0 24px 56px -24px rgb(0 0 0 / 0.85),0 6px 16px -8px rgb(0 0 0 / 0.50);--well:inset 0 3px 5px -2px rgb(0 0 0 / 0.58),inset 0 -2px 0 -1px rgb(255 255 255 / 0.115);--grain-panel:none;--grain-table:none;--grain-menu:none;--grain-page:none;--skeleton-base:#292524;--skeleton-shine:#44403c;--backdrop:rgb(0 0 0 / 0.6);} .link{color:var(--accent-hover);text-decoration:underline;text-underline-offset:2px;} .link:hover{color:var(--accent);} /* The icon control's HIT AREA and RADIUS,and nothing else - this is not a face and cannot be one. Two of the three controls that carry it (the queue row's edit,the activity feed's pencil) are bare glyphs on `background:none`,and the third (the stickies row's delete) is a red key because Delete gives it a fill. So the material cannot live here:it lives in .btn and in the bare set's depth reset,and this rule is the square,the target and the hover well. The box model here used to be dead. `padding` and the two minimums are one class,and `.btn` declares `padding:9px 18px` at the same specificity later in the file - so on the stickies delete,which has no family class of its own,`.btn` won and the "square icon button" rendered 56x38 with 18px of padding round a 20px glyph. It is written at two classes now so the square is the control's and not an accident of which other class it happens to carry. */ .btn.icon-btn,.icon-btn{display:inline-flex;align-items:center;justify-content:center;min-width:var(--hit);min-height:var(--hit);padding:var(--space-1);border-radius:var(--radius-md);cursor:pointer;} /* The muted glyph is the bare families' and is deliberately not in the rule above:that one is two classes and would land on the kind-filled icon button,which carries --on-kind for the red fill under it. */ .task-row__btn{color:var(--text-muted);} .icon-btn:hover{color:var(--text);background-color:var(--surface-hover);} /* === Icons === Tabler outlines on a 24x24 grid,stroked with currentColor and never filled,so an icon takes the colour of whatever contains it and carries no colour of its own. Width and height come from the size Icon.view renders;these rules only place the box. */ .icon{display:inline-block;vertical-align:middle;flex-shrink:0;} .icon--unknown{width:1em;height:1em;} *,*::before,*::after{box-sizing:border-box;margin:0;padding:0;} /* The page is level 0,and level 0 gets no shading. It gets grain and nothing else - the weakest of the four,and the one that does least:at 1x it is sub-pixel and reads as a slightly warmer background rather than as texture. It is here because it costs nothing measurable once --text-dim is re-based. `background-blend-mode` blends an element's OWN layers,so the colour has to be on the same element or the grain paints over the page instead of into it;that is why this is a background-color/image pair and not `background`. */ html,body{font-family:var(--font);background-color:var(--bg);background-image:var(--grain-page);background-blend-mode:multiply;color:var(--text);-webkit-font-smoothing:antialiased;font-size:var(--text-base);line-height:var(--leading);} /* Floor at the narrowest phone this supports. Below that the page scrolls sideways rather than folding the tables and toolbars into a column two characters wide. */ body{padding-bottom:calc(var(--tab-height) + env(safe-area-inset-bottom,0px));min-width:320px;} /* Selecting text is not an action and has no kind,so the highlight is the brand's face:lime with near-black ink in light,violet with white in dark,the same pair the chosen chip and the rail's live tab use. The browser's own blue was the only unthemed colour left on screen. Both ink pairs are the ones already measured on those faces - 11.34:1 and 5.73:1. */ ::selection{background:var(--select-fill);color:var(--mark-ink);} /* The nav is not rendered for a signed-out visitor,so the room reserved for it goes too - otherwise a wide screen shows an empty rail column beside the login box,and a phone shows a strip of nothing under it. */ body:not(:has(.tab-bar)){padding-bottom:0;padding-left:0;} body:not(:has(.tab-bar)) .login-page{min-height:100dvh;} a{color:var(--accent);} :focus-visible{outline:2px solid var(--accent-hover);outline-offset:2px;} .page__title{font-size:var(--text-xl);font-weight:600;line-height:var(--leading-tight);margin-bottom:var(--space-6);} .app-content{padding:var(--space-4);max-width:var(--page-max);margin-inline:auto;} /* A report is a table,and a table is the one thing here that wants width rather than a text measure. --page-max holds the reading pages to 1120px,which is right for them and wrong for this:cepending's six columns were being laid out in a 1088px content box on a window with several hundred pixels spare on each side,so Scheduled and Size fell off the right edge while the space to show them sat empty. Reports take the window;everything else keeps the measure. */ .app-content:has(.report-view){max-width:none;} /* The bar is an edge,not a surface:it gets a hairline and a defined top edge and no material of its own. Grain on it was measured and reads as a dirty strip,because there is no face for texture to establish itself on. */ .tab-bar{position:fixed;bottom:0;left:0;right:0;background:var(--surface);border-top:1px solid var(--border-light);z-index:100;padding-bottom:env(safe-area-inset-bottom);} .tab-bar ul{display:flex;list-style:none;} .tab-bar li{flex:1;} /* Scoped through .tab-bar on purpose:the tab is a button,so it also carries .btn,and .btn is later in this file. At one class each,source order handed every shared property to .btn - the padding,the border,the display,the font-size - and the bar grew from 54px to 66px under a 48px reservation. Two classes outrank one,so the tab's own box model holds against .btn while the kind's colour still wins on source order (the .btn--ghost.btn--read pair is also two classes,and it is later). The same scoping is on the desktop rule below:a media query does not add specificity,so it needed it too. */ .tab-bar .tab-btn{transition:color 200ms ease;background:none;border:none;color:var(--nav-text);cursor:pointer;padding:6px 4px 4px;font-size:10px;display:flex;flex-direction:column;align-items:center;gap:4px;font-family:inherit;width:100%;} /* The bottom bar's selected tab has no pill - its chip is behind the icon,not the row - so this label sits on the bar itself and wants the accent that moves away from a plain surface. --accent on its own measures 5.52:1 light and 4.75:1 dark,which left the selected tab dimmer than the unselected ones once those took --nav-text. */ .tab.active .tab-btn{color:var(--mark-ink);} .tab-icon{width:28px;height:28px;border-radius:var(--radius-sm);display:flex;align-items:center;justify-content:center;} .tab.active .tab-icon{background:var(--select-fill);color:var(--mark-ink);} /* A hovered tab is not the selected tab. Every tab is a ghost read button,so the ghost hover rule painted it --read-subtle - and --read-subtle IS --accent-subtle,which is the fill the selected tab carries. Measured at 1280 in both themes before this rule:the active tab and a hovered inactive one were rgb(219,234,254) with an rgb(29,78,216) label in light,and rgb(30,58,95) with an rgb(147,197,253) label in dark - the same fill,the same label,on two different tabs. In the rail that is two identical pills and no way to tell where you are. A nav cannot use the ghost's hover,because its selected state already owns it. Hover takes a well instead:--bg. In light that is the move --surface-hover already makes - the two are 1/255 apart - and in dark it puts the hover pill below the rail's face rather than above it. So in both themes the hover well is darker than the rail while the selected pill is the lighter,bluer thing sitting in it,and the two differ in direction and hue rather than only in degree. The label stays --btn-hue-read,the colour it already rests at,so hover changes the fill and nothing else - which is what every other .btn hover in the app does. Why not --surface-hover for the fill:--btn-hue-read on it measures 5.04:1 in light but 4.29:1 in dark,under the 4.5 floor,and the dark rail's unselected labels have only 0.25 of margin to begin with (4.75:1 on #1c1917) - no fill lighter than the rail's own face can carry them. --bg measures 5.09:1 light and 5.37:1 dark under the same label. The ghost's --on-read-subtle is not an option either:it was solved against --read-subtle and measures 4.09:1 light and 4.29:1 dark on either candidate. Five classes,so it outranks .btn--ghost.btn--read:is(:hover,:focus-visible) at three wherever either sits in the file,and :not(.active) is what keeps it off the selected tab in every state - including while that tab is hovered,which is the case that has to stay --accent-subtle. */ .tab-bar .tab:not(.active) .tab-btn:is(:hover,:focus-visible){background-color:var(--bg);color:var(--btn-hue-read);} /* The theme toggle's row at the rail's foot. The phone bar is five tabs and nothing else - the toggle is reachable there from the More menu - so this is not rendered below the rail's own breakpoint. It is the same class the rail and the bar are gated on,so the two can never disagree about which shape the nav is in. */ .rail-foot{display:none;} /* === Desktop rail === The bottom bar is right for a thumb and wrong for a window:across a wide screen the same five tabs stretch to fill the whole width,which is what made the app read as a phone layout opened on a desktop. From tablet up the bar becomes a fixed left rail and the page moves beside it. The markup is untouched - same nav,same buttons,same order - so nothing has to know which shape it is in. AT REST THE RAIL IS ICONS ONLY,and it opens to its labels while it is hovered or while focus is inside it. Two decisions carry that:- IT EXPANDS OVER THE PAGE,and the page is inset by the collapsed width. That was the whole of it,and the reasoning was that pushing the content would re-lay-out every card,table and toolbar on the screen once per pass across the nav - the expensive version,and it moves the thing being read. Overlaying cost the left 138px of content for as long as the pointer was on the rail,which is what an icon rail conventionally does. IT DOES NOT SURVIVE A FULL-WIDTH TABLE. The 138px was free while every page sat in a centred 1120px column with slack beside it. A report now takes the window - see .app-content:has(.report-view) - so the open rail came down on the Client column and part of Email. There is no inset that fixes that:the content occupies all of the space the rail would open into. So the page moves with it,which is the cost the note above refused. The refusal was about the screen the pointer is NOT on;here it is the screen being read,and being covered is worse than being re-laid-out. Measured at 1898 wide:at rest rail 78 / table left 94;open rail 216 / table left 232 / body padding-left 216. Overlap 0 in both states,and back to 94 on the way out. Both edges run on the same 160ms ease so they travel together. - :focus-within OPENS IT TOO,and that is not a courtesy. A keyboard user cannot hover,so with :hover alone the rail is five icons whose labels no one navigating by keyboard can ever bring back. The phone bar is untouched by all of this:it lives outside this query,where the labels are the entire point. */ @media (min-width:768px){:root{/* Sized to the TABS,not to the foot,and that is a correction rather than a tightening. It was 216px,which is what "Switch to dark mode" needs - and that label is not one of the five things this rail is for. Measured at 14px,the natural widths are:widest tab Accounts 12 + 28 + 12 + 51 + 12 = 115 foot,theme Switch to dark mode 181 foot,status DeepSeek is up 151 With the ul's 12px of side padding and the 1px border drawn inside the width,the tabs need 12 + 115 + 12 + 1 = 140. The foot's labels are hidden now (see .rail-foot .tab-label),which is what makes 140 possible at all,and their sentences moved to the buttons' accessible names. 144 rather than the exact 140:label widths are font-rendered,and a platform a few pixels wider than this one would clip the end of "Accounts" against the rail's own overflow. Four pixels is the whole cushion,and it is 72px narrower than the 216 it replaces. */ --rail:144px;/* What the rail is worth once the labels are gone,and it is arithmetic rather than taste. The ul keeps its 12px of side padding and the button keeps its 12px,so the chip's 28px needs 2*12 + 2*12 + 28 = 76;the 1px right border is drawn inside the rail's own width (border-box),so the floor is 77,and 78 is that with a pixel to spare - the chip is never on the rail's edge. Narrower than this the rail clips the icons it exists to show;wider is a column of dead space down a nav whose content is 28px wide. */ --rail-icons:78px;} body{padding-bottom:0;padding-left:var(--rail-icons);transition:padding-left 160ms ease;} /* The rail is fixed,so opening it drew over whatever was underneath. That was survivable while every page sat in a centred 1120px column with slack beside it,and it is not on a report:a report takes the window,so the open rail covered the left of the table - the Client column and part of Email. THE PAGES THAT MOVE ARE NAMED,NOT ALL OF THEM. Pushing is the expensive answer - it re-lays-out what is being read,which is the thing the note above this block refused for the general case - so it is spent only where the overlay actually costs a column. That is the CE reports - cepending is six columns including a full URL - the pull preview,whose leftmost column is the task number the whole screen is read for,and the unused newsletters list,whose leftmost is the account,which is the only thing saying whose newsletter is about to be deleted. All three are full-bleed because they wrap in .report-view,and that is what leaves the rail no room to open into. Being on this list is not the same as being a report. Hours is here because it failed the same measurement rather than because it looks like one:at 1100 and below - where the cap that makes `.crm-table` a real table stops applying and its box goes back to the page's own edge - the open rail reaches 144 while the first cell starts at 3,so 141px of it is under the rail. At 1440 that same cell starts at 155 and the rail clears it by 11. The entry is kept for the widths below 1101,and these numbers were re-measured on 2026-10-14:the pair that put Hours on the list (a rail of 216 reaching a cell at 97) was taken before the rail was narrowed to 144. It wraps in .report-view like the others,which is what takes the page off the reading measure and leaves the rail nowhere to open into. Add a route here when a report is wide enough to be covered by the open rail and wide enough that being covered loses something. `data-screen` is the route the report renders as,written by app.ml on the wrapper it owns from the screen's own name - it was `Lib.ReportTable.view`'s `~screen` until 2026-10-13,and it moved because the renderer copies attributes without clearing the ones it is not given:the node that names a screen cannot be one that outlives it. Measured on the pull preview,1440px,2026-10-14,with the rail's own 160ms transition settled:at rest the rail ends at 78 and the table starts at 97,and open it ends at 144 - the rail was 216 when this block was written,and these numbers came down with it - while,with the rule defeated,the table still starts at 3:141px of it under the rail,which is the Task and Account columns. Reading that at the moment of the change shows the value the padding is leaving and reports no overlap at all,because the rule below moves the body on a transition too. Same 160ms and easing as the rail's own width transition,so the two edges travel together rather than one arriving before the other. */ /* THE WHOLE REPORT SET IS HERE,NOT A SAMPLE OF IT,and that is the change a sweep forced:the list had three reports and a measurement found fifteen more with the same defect - every one of them full-bleed through .report-view,first column at 97 while the open rail reaches 144,so 31px of the Account header (113 to 144) sat under the rail. Two of them,plus the tip report,could not be named at all until the components were given a screen:stale,prospects,profitability and tip called ReportTable.view without ~screen and rendered data-screen="" - a selector that matches nothing,which is the failure this list's own comment warns about. Measured after,at 1440 with the rail's transition settled:every screen named here moves its body from padding 78 to 144 when the rail opens,and the pages that are NOT named never needed to - they sit on the reading measure,where the first column starts at 218 and the rail stops at 144. */ body:has( [data-screen="cepending"],[data-screen="ceaccounts"],[data-screen="ceblasts"],[data-screen="pull-preview"],[data-screen="unused-newsletters"],[data-screen="hours"],[data-screen="done-tasks"],[data-screen="am"],[data-screen="amurls"],[data-screen="calllist"],[data-screen="clearfly"],[data-screen="freshprospects"],[data-screen="pending"],[data-screen="profitability"],[data-screen="projects"],[data-screen="prospects"],[data-screen="prospectsall"],[data-screen="referral"],[data-screen="seo"],[data-screen="shortprospects"],[data-screen="stale"],[data-screen="tagcm"],[data-screen="tip"] ):has(.tab-bar:is(:hover,:focus-within)){padding-left:var(--rail);} /* The completed-tasks table while the next window's rows are in flight. The table itself is never taken out of the tree - see the note in components/doneTasks.ml - so this dim is the whole of the feedback,and it has to be on a wrapper this app owns rather than on the library's node. */ .done-tasks__table[data-busy="true"]{opacity:0.55;} .done-tasks__table{transition:opacity 120ms ease;} @media (prefers-reduced-motion:reduce){.done-tasks__table{transition:none;}}} /* THESE ARE NOT DESKTOP RULES,and they were nested inside the query above. The block that begins at `min-width:768px` runs to the rail's labels near the end of this file,so anything written between the two landed in it - silently,because a rule inside a query the window does not match is a rule the browser never sees,and the elements simply take their defaults. The timer's task list and the edit-a-task screen were both written there,so on the 475px popup the timer's list drew as a bulleted list with full-size buttons,and neither screen has ever been styled on a phone. Nothing here is conditional:they are the popup's own layout at every width. The desktop query is closed above and reopened below,and what is inside each half is what was always meant to be. */ /* Open a task by its number. One row:the box,then the button,with the app's own gap between them rather than a margin on either. */ .timer__group{list-style:none;margin:var(--space-3) 0 0;padding:0;display:flex;flex-direction:column;gap:var(--space-1);} .timer__group-row{display:flex;align-items:center;gap:var(--space-3);} .timer__group-id{font-variant-numeric:tabular-nums;color:var(--text-muted);} /* ONE LINE PER ROW,and the name is the part that gives. It takes whatever room the id,the figure and the remove control leave,and what does not fit ends in an ellipsis - a wrapped name made a three-task list taller than the window and a nine-task one unreadable. The id is muted because the name is what a person reads;the figure and the control keep their own width. */ .timer__group-name{flex:1 1 auto;min-width:0;overflow:hidden;text-overflow:ellipsis;white-space:nowrap;} .timer__group-share{color:var(--text-muted);white-space:nowrap;} .timer__group-add{display:flex;align-items:flex-end;gap:var(--space-3);margin-block-start:var(--space-3);} /* WHAT A SEARCH FOUND,UNDER THE BOX THAT FOUND IT - one block for both screens that pick a task,because they are meant to be the same control:the multi-timer's add box and the way into the task editor. One line per row,the name taking the room the number and the account leave,and a bordered row rather than a list of words,since each one opens something. */ .task-search__list{list-style:none;margin:var(--space-2) 0 0;padding:0;display:flex;flex-direction:column;gap:var(--space-1);} .task-search__row{display:flex;align-items:center;gap:var(--space-3);width:100%;justify-content:flex-start;text-align:start;} .task-search__id{font-variant-numeric:tabular-nums;color:var(--text-muted);} .task-search__name{flex:1 1 auto;min-width:0;overflow:hidden;text-overflow:ellipsis;white-space:nowrap;} .task-search__account{color:var(--text-muted);white-space:nowrap;} /* A row that is already in the run says so,where its account would be. Not disabled - a disabled control is a dead one - but its click is inert,because a task cannot be in a run twice. The name is left alone:the marker is the signal,and a muted name would only make the row harder to read. Two classes,because `.task-search__account` sets the muted colour at one. */ .task-search__row--added .task-search__account{color:var(--success-text);} .task-search__add-all{display:flex;justify-content:flex-end;} /* A NUMBER THE SEARCH COULD NOT FIND,offered as itself:the task may be finished,and a finished task is not among the rows any search over the open ones can reach. Dashed,because it is a promise to open something nobody has shown yet rather than a task that was found. */ .task-search__row--plain{border-style:dashed;} .task-search__title{font-size:var(--text-lg);font-weight:600;margin-block-end:var(--space-2);} .task-search__note{color:var(--text-muted);margin-block:var(--space-2);} .task-search__none{color:var(--text-muted);margin-block:var(--space-2);} .task-search__box{display:flex;align-items:flex-end;gap:var(--space-3);} /* === Card === Level 1,the panel - and the same three layers every other panel in the app now takes. The four properties are one material and are meaningless apart:the hairline edge alone is the flat box this replaces,the rim alone has no boundary at the top,and the ambient alone has no face to sit under. --surface is INVISIBLE here,whatever it is set to. Measured:a #ffffff,#fcfbf9 and #f8f6f2 base all come out identical once the rim is on them,because the gradient covers the face - so this rule's background-color is a fallback that never paints,and --surface only has to be right for the things that do show it (the rail,a field,a control's own face). It is set one step above --bg for the rules that read it directly,and it is not what makes a card bright:the rim below is. That property is also why the light pass could not be done by moving --surface alone:the panel is the big bright thing on screen and its whole face is this gradient. --surface moved for the controls,the page moved under it,and the rim moved with them - but only its top and middle stops,for the reason recorded against --panel-rim. */ .card{background-color:var(--surface);background-image:var(--grain-panel),var(--panel-rim);background-blend-mode:multiply,normal;background-size:160px 160px,100% 100%;border:1px solid var(--border);border-radius:var(--radius-xl);box-shadow:var(--panel-lift),var(--panel-bevel);padding:var(--space-4);margin-bottom:var(--space-4);} .card__title{font-size:var(--text-xs);font-weight:600;color:var(--text-muted);text-transform:uppercase;letter-spacing:0.05em;margin-bottom:var(--space-3);} .card__head{display:flex;align-items:center;justify-content:space-between;gap:var(--space-2);flex-wrap:wrap;margin-bottom:var(--space-3);} .card__head .card__title{margin-bottom:0;} .card__head .table-search{margin-bottom:0;} /* A card whose search matches nothing,so the result is stated rather than left as a bare header row. */ .tip-table__empty{padding:var(--space-4) 0;color:var(--text-muted);font-size:var(--text-sm);} .tip-table__note{flex:2;color:var(--text-muted);font-size:var(--text-sm);overflow:hidden;text-overflow:ellipsis;white-space:nowrap;} .tip-table__when{flex:0 0 130px;text-align:right;color:var(--text-muted);font-size:var(--text-xs);font-variant-numeric:tabular-nums;} .tip-table__pager{display:flex;align-items:center;justify-content:space-between;gap:var(--space-2);flex-wrap:wrap;margin-top:var(--space-3);} .tip-table__count{color:var(--text-muted);font-size:var(--text-xs);font-variant-numeric:tabular-nums;} .tip-table__actions{display:flex;gap:var(--space-3);} /* A card's own actions,on a line under the title. The full basis is what puts them there:it forces the wrap itself,so the line is the card's and not whatever the title and the search happened to leave over. */ .card__actions{flex-basis:100%;display:flex;gap:var(--space-3);} /* === The control material === One material,and everything in this app that acts like a button takes it. What that means precisely:- ONE MATERIAL,NOT ONE LOOK. The material is the surface,the edge,the depth,the four interaction states and the hit area. Colour is not part of it - colour is the function signal,and a family keeps its own (blue Read,green Write,red Delete,and the neutral base for a control with no kind). Shape is not part of it either:a sub-tab is still a tab,a menu row is still full width,an icon button is still square,the FAB is still a circle. What the families share is how the surface is BUILT,which is why they can go on looking nothing like each other from across the room. - BOXED AND BARE ARE THE TWO SHAPES OF IT,and they are a rule rather than a list. A control with a face of its own - a kind fill,or --surface - is BOXED:edge,bevel,ambient,sheen. A control with no face - a bare glyph,a text control,a tab,a menu row,a sort header - is BARE:it has the hit area,the type metric,the four states and the press,and no edge,no bevel and no ambient,because all three of those describe a surface and it has none. The line is not decorative:painting a bevel on a transparent box is what made the queue's bare pencil a raised white key while the identical pencil in the activity feed stayed flat,and what drew two lit lines across the dashboard's sort headers and the More menu's divider rows. A family changes sides by acquiring a face,not by being added to a list. - THE BOXED SET is .btn in every kind and in the neutral base,.chip (the anchors as well as the buttons),.detail-header__action-btn,.timer__copy,.ce-cred__remove,the table pager's buttons,and Shared.Fab - which has no call sites and is included anyway,because a shell nothing calls is where a divergent treatment survives unnoticed until something does call it. THE BARE SET is .sub-tab,.menu-group__item,.back-button (and every ghost,which is the same thing) and .tip-table__header button. `.task-row__btn` is on neither list,and that is a correction rather than a gap:it was kind-filled - solid Read and solid Delete - and both lists went on naming it as bare. The activity feed's row edit carried `.activity-feed__entry-action` and was in the bare set for real,which is why it read as a grey pencil rather than as a control;it takes `.task-row__btn` now and the class is gone. `.icon-btn` is on neither list,and that is the point rather than an omission:it is a hit area and a radius,not a face. The queue's edit key and the activity feed's pencil are bare because their background is none;the stickies row's delete carries the same class and is a red key,because Delete gives it a fill. A control's side is decided by what is under it. - THE FORM OF IT LIVES IN TOKENS,so a family cannot drift:--shade for the depth,--btn-bevel and --btn-bevel-hover for the edge,--sheen-* for the curvature,--border-strong for the boundary,--radius-md for the box,--hit / --hit-touch for the target. No family rule writes any of them out;a family rule says what shape it is and nothing else. What follows is the filled button's own declaration of it. Every other boxed family re-states the same two box-shadow tokens because it sits later in the file at equal specificity - which is the one thing this catalogue's flatness costs,and the reason the tokens exist rather than each rule repeating them. A button's colour states what it does to the database. Read changes nothing,write creates or updates,delete removes. The rule only holds if it is applied without exception,so every button in the app is built by Lib.Button.view,which is where the class comes from. The surface is matte and physical rather than flat and glowing:a noise grain over the fill,a lit gradient under it,an inset bevel for an edge,an ambient that seats the key in the page rather than lifting it off,and a press that moves. No glow,no pill radius. The two background layers are one job each. The gradient is the curvature - a flat fill at any amount of texture still reads as a rectangle - and the grain is the material. The blend list pairs with the image list position by position:the grain blends `overlay`,and the gradient composites `normal` over the fill so its stops keep the values they were solved at. Those stops are bounded by AA,not by taste. The top stop is 5.5% white and no more:an earlier prototype at 14% white measured 4.03:1 on the read face and 3.88:1 on the write face,against a 4.5 floor,while looking fine in a screenshot. At 5.5% the gradient *alone* measures 4.54:1 at the write top and still gives a ~15% top-to-bottom delta,which is what reads as curvature. The gradient alone is not the face a reader gets,though:the grain composites over it,and which way that moves the rendered number depends on the blend. Under `multiply` - what the fills use - the face comes out darker than the gradient alone,so the sheen's top stop is no longer the binding number. The rendered bands,the two layers measured one at a time and what the other blend does are recorded against the write fill token. */ .btn{display:inline-flex;align-items:center;justify-content:center;gap:var(--space-2);/* A .btn is a button first and an anchor second,and the underline is the one thing that came with the second and not the first. Every control that navigates is an <a> now - href on Lib.Button.view is what makes a tab middle-clickable - and an <a> arrives underlined while a <button> does not,so the same control was underlined in the nav and clean in the rail's foot beside it. Measured,not assumed:the five tabs and the status link both read text-decoration-line:underline,and the theme toggle beside them reads none. */ text-decoration:none;/* 9/18 rather than the 8/16 space tokens:the extra pixel of height puts the label on the optical centre of a bigger key,and the two of width keep the proportions of the 4px-to-6px radius change. */ padding:9px 18px;border:1px solid var(--border-strong);border-radius:var(--radius-md);background-color:var(--surface);color:var(--text);/* The neutral base keeps `overlay` for the grain,and that is the correct choice rather than an oversight. `overlay` against a white backdrop is the identity function,so a neutral key stays effectively clean - 7 distinct colours against the flat control's 1,where the fills measure 44-63 - and its sheen does the visible work instead. `multiply` here would paint the same white surface 23 distinct,which is not more texture:it is the stain the grain token's rule warns about. The three kinds override to `multiply` below,where the surface is the fill and the grain is material. */ background-image:var(--grain),linear-gradient(180deg,var(--sheen-top),var(--sheen-bottom));background-blend-mode:overlay,normal;background-size:160px 160px,100% 100%;/* The type metric is the material's,not each element's. A <button> takes the UA's `line-height:normal` and an <a> inherits the body's 1.4,so a family built on both - the More menu,the account header - renders two boxes of different heights for the same class. Measured before this line:a More menu button row was 40px and the link row directly beneath it 44px. 1.15 is what a button already rendered at,so this pins the button and pulls the anchor to it rather than the other way round. Same reason `min-height` is here:it is the control's floor,and it is what makes the 17px sort header a target. */ font-family:inherit;font-size:var(--text-sm);line-height:var(--leading-tight);min-height:var(--hit);cursor:pointer;white-space:nowrap;box-shadow:var(--shade),var(--btn-bevel);transition:background-color 120ms ease,box-shadow 120ms ease;} .btn:focus-visible{outline:2px solid var(--focus-ring);outline-offset:2px;} /* The neutral base's own hover. Before the kind rules,so a kind overrides it on source order at equal specificity - and until T4-T8 land every btn--primary call site is neutral,so without this the migration window has no hover at all. */ .btn:is(:hover,:focus-visible){background-color:var(--surface-hover);} /* The key travels the whole way down:1px of offset,and it shrinks a hair as well,because a key pressed against a surface does not keep its width. The 2% is the largest scale that measured as a press rather than as a shrunken button at 3x - at 0.95 the label's own letterforms visibly narrowed on the row above it. The transform is not transitioned:the press has to land on the same frame as the mouse,and the release is meant to snap. */ .btn:active{transform:translateY(1px) scale(0.98);box-shadow:var(--btn-press);filter:var(--press-dim);} /* A write in flight must not look pressable again. Two things were wrong with the state as it first shipped. It dimmed the WHOLE key - `opacity:0.55` took the fill down with the label,so a write in flight stopped looking like a write at the one moment the reader most wants to know which write is running. And nothing wired it up,so the state was never seen. Now:the face holds the kind colour and the label steps back (--on-kind-busy),the depth comes off so it does not read as raised,and the cursor says working rather than forbidden. `saturate(0.9)` is the last 10% of the step,and it is a filter rather than a colour so dark needs no value of its own. */ .btn:disabled{opacity:1;cursor:progress;transform:none;box-shadow:none;filter:saturate(0.9);} /* The depth is restated here,not inherited from `.btn`:a kind that declares only the two insets replaces the whole box-shadow list,so the key would sit on the page at rest on the neutral base and float on nothing the moment it carried a kind - which is every button that does something. It is now two tokens rather than five copies of the same two insets,which is what makes it possible to change the control's depth in one place and have every family move together. The blend list carries its second value for the same reason:with one value `overlay` repeats onto the gradient layer,and the top stop then composites at a different alpha than the one it was solved at. multiply,not overlay,for the grain on the fills. `overlay` takes the backdrop's side of its branch,and against these mid-tones its mean effect is to *lighten*,which is the direction that costs contrast:the write face's top band measured 4.53:1 with the pre-wave grain and 4.11:1 with this one. multiply only darkens,so it moves contrast the safe way and the measured top band came back stronger than before the texture existed. The cost is recorded against the fill tokens - the rendered face now sits below its token,which keeps the token the conservative number. */ .btn--read{background-color:var(--read-fill);color:var(--on-kind);border-color:var(--read-fill);background-blend-mode:multiply,normal;box-shadow:var(--shade),var(--btn-bevel);} .btn--write{background-color:var(--write-fill);color:var(--on-kind);border-color:var(--write-fill);background-blend-mode:multiply,normal;box-shadow:var(--shade),var(--btn-bevel);} .btn--delete{background-color:var(--delete-fill);color:var(--on-kind);border-color:var(--delete-fill);background-blend-mode:multiply,normal;box-shadow:var(--shade),var(--btn-bevel);} .btn--read:is(:hover,:focus-visible){background-color:var(--read-fill-hover);border-color:var(--read-fill-hover);} .btn--write:is(:hover,:focus-visible){background-color:var(--write-fill-hover);border-color:var(--write-fill-hover);} .btn--delete:is(:hover,:focus-visible){background-color:var(--delete-fill-hover);border-color:var(--delete-fill-hover);} .btn--read:is(:hover,:focus-visible),.btn--write:is(:hover,:focus-visible),.btn--delete:is(:hover,:focus-visible){box-shadow:var(--shade),var(--btn-bevel-hover);} /* Three classes,not two,and colour restated alongside the fill. A converted button often carries a class of its own - sub-tab,chip,tab-btn,detail-header__action-btn - and that class declares color at one class while this used to set only background-color,so the fill moved and the label stayed the page colour. Several of those classes also set color or background at (0,2,0) and sit later in the file,so they beat a two-class rule on source order. Three classes beats them on specificity,so order stops mattering. :active is folded in because press is its own cascade and patching hover alone moved this failure one state over twice. */ .btn.btn--read:is(:hover,:focus-visible,:active){background-color:var(--read-fill-hover);color:var(--on-kind);border-color:var(--read-fill-hover);} .btn.btn--write:is(:hover,:focus-visible,:active){background-color:var(--write-fill-hover);color:var(--on-kind);border-color:var(--write-fill-hover);} .btn.btn--delete:is(:hover,:focus-visible,:active){background-color:var(--delete-fill-hover);color:var(--on-kind);border-color:var(--delete-fill-hover);} /* The whole of how a sub-tab is marked is its underline,and the rule above repaints border-color on every side. `.sub-tab.active` is (0,2,0) and loses,so pressing the selected tab recoloured its underline to the hover fill's own value and the underline disappeared into the fill. Restated here at four classes - one more than the kind rule,so it wins on specificity rather than on having been written last - and it restates only the underline:the kind rule keeps the fill and the label,which is what it exists for. --on-kind rather than --accent while the fill is up. The underline is drawn on --read-fill-hover once the tab is hovered,and --accent on that measures 1.30:1 in light and 1.82:1 in dark - present,and invisible,which is the bug again. The fills do not flip between themes,so --on-kind is #ffffff in both and measures 6.70:1 on that fill,the same as the label it sits under. */ .btn.sub-tab.active:is(:hover,:focus-visible,:active){border-bottom-color:var(--on-kind);} /* The kind press,at three classes for the reason the hover above is:a converted button carries a class of its own that may set a box-shadow,and the press is the state where losing that fight is invisible - the sweep in the panel's own note found exactly that,a press that went nowhere. Same token the neutral base presses with. */ .btn.btn--read:active,.btn.btn--write:active,.btn.btn--delete:active{transform:translateY(1px) scale(0.98);box-shadow:var(--btn-press);filter:var(--press-dim);} /* THE TWO FAMILIES. Every key in the app is a utility key:6px radius,a matte face,the edge in --btn-bevel. One key per screen may be the PROMINENT one - the thing that screen is for - and it says so with a rounder box,the sheen at three times the depth,and its own fill cast as an ambient under it. Nothing about the colour changes:the kind rule above still owns the fill,so a prominent read is blue and a prominent write is green. The ambient is the one part of this that the app's visual language did not already have - "explicitly not:outer glow" is written in crm-app/CLAUDE.md,and this is a live amendment to it rather than an exception carved quietly. What makes it the app's own vocabulary and not a landing-page glow:it is a negative-spread ambient with the same shape as --shade and the panels',it is tinted with the fill it sits under rather than with a neutral colour,and it is bounded - 16px of reach,6px of offset,so it reads as the key resting in its own light at the size of one control. If it ever reads as a light source,it has been made too wide or too opaque,and those two numbers are the ones to move. The radius is 10px rather than the 6px the surface language states. That is the whole of the shape difference,and it is deliberately small:a pill would read as a different KIND of control - a different thing to press - and these are the same keys,one of which is the one to press. THE FILL IS RESTATED,and that is the second measurement this family needed. A prominent key wears its kind's colour and nothing else can take it away:the account header's keys rest on a bespoke NEUTRAL face at (0,2,0),so the first version of this rule put the veil over grey - a prominent Quick Log that was a shade of dirty beside the two flat keys it sits with,which is the same soiling the grain token's rule records for `--surface`. Restating the fill at three classes settles it in both directions:a key whose bespoke class paints its own face gets the kind's fill back,and the ambient it casts has a colour to be cast from. */ .btn--prominent{border-radius:var(--radius-xl);} .btn.btn--read.btn--prominent{background-color:var(--read-fill);color:var(--on-kind);border-color:var(--read-fill);background-image:var(--grain),var(--prominent-veil);box-shadow:var(--shade),var(--btn-bevel),var(--prominent-cast) var(--halo-read);} .btn.btn--write.btn--prominent{background-color:var(--write-fill);color:var(--on-kind);border-color:var(--write-fill);background-image:var(--grain),var(--prominent-veil);box-shadow:var(--shade),var(--btn-bevel),var(--prominent-cast) var(--halo-write);} .btn.btn--delete.btn--prominent{background-color:var(--delete-fill);color:var(--on-kind);border-color:var(--delete-fill);background-image:var(--grain),var(--prominent-veil);box-shadow:var(--shade),var(--btn-bevel),var(--prominent-cast) var(--halo-delete);} /* The prominent hover keeps the ambient - it is the key's,not a state's - and takes the same lift every other key takes. */ .btn.btn--read.btn--prominent:is(:hover,:focus-visible){box-shadow:var(--shade),var(--btn-bevel-hover),var(--prominent-cast) var(--halo-read);} .btn.btn--write.btn--prominent:is(:hover,:focus-visible){box-shadow:var(--shade),var(--btn-bevel-hover),var(--prominent-cast) var(--halo-write);} .btn.btn--delete.btn--prominent:is(:hover,:focus-visible){box-shadow:var(--shade),var(--btn-bevel-hover),var(--prominent-cast) var(--halo-delete);} .btn.btn--read.btn--prominent:active,.btn.btn--write.btn--prominent:active,.btn.btn--delete.btn--prominent:active{box-shadow:var(--btn-press);} /* A key whose write is in flight. Three classes,for the reason above:the label of a kind is set at one class and a bespoke class later in the file would otherwise keep it. The fill is untouched - it is the kind saying which write is running - and the depth is already gone at `.btn:disabled`. */ .btn.btn--read:disabled,.btn.btn--write:disabled,.btn.btn--delete:disabled{color:var(--on-kind-busy);} /* ...and a prominent key in flight puts its ambient down as well,which is a rule of its own only because the halo is declared at four classes and would otherwise outlive the depth it belongs to. */ .btn.btn--read.btn--prominent:disabled,.btn.btn--write.btn--prominent:disabled,.btn.btn--delete.btn--prominent:disabled{box-shadow:none;} /* THE BARE SET HAS NO DEPTH. This is the whole of what "bare" means and it is one rule rather than a habit,because the failure it prevents is silent:the bevel and the ambient describe a SURFACE,and on a box with `background:none` they draw the surface anyway,in mid-air. That is not a subtlety - it is what the queue's edit pencil was,a raised white key with a hard shadow under it beside an identical pencil in the activity feed that was flat,and what the dashboard's sort headers and the More menu's divider rows were,two lit lines painted across text with nothing behind them. Written at two classes,after every rule above that sets box-shadow on a kind,so the reset holds in EVERY state:rest,hover,and the press,which is `.btn--read:active`'s inset. None of the six has a face of its own - the two icon controls and the sort header have `background:none`,the tab and the back control and the menu row are transparent over their own container - so none of them can carry a bevel honestly. What they do carry is the rest of the material:--hit for the target,the type metric,the four states and the press translation,which is the whole of their affordance. `.icon-btn` is deliberately NOT in this list even though it reads like the family name for two of them. It is a hit area and a radius,not a face:the stickies row's delete carries it too,and that one is a red key because Delete gives it a fill. A control changes sides by acquiring a face,and nothing acquires one by its class being named here. Five of the six selectors here are drawn on BOTH a <button> and an <a> or on a bare glyph;the sixth,`.tip-table__header button`,is a button but matches only through its parent. That last one is why this selector list needs no `.btn` qualifier and the sibling rule above does — see the note on it. */ /* `.tip-table__header .btn` and not `.tip-table__header button`:the second reads like the same rule,and it is the WRONG SHAPE - the header is the button's PARENT,so `header button` matches a button inside a header while `header .btn` is the one that outranks the kind. Written the first way the reset silently did nothing and the sort header kept the bevel and the ambient,which is the exact defect it exists to prevent. */ /* .task-row__btn is deliberately NOT in this list any more. It is a kind-filled key like any other now - solid Read and solid Delete - and the depth is part of what a kind fill carries;stripping it here left the app's reddest control as its flattest. The rest of the list is still bare and still gets the reset. */ .btn.sub-tab,.btn.menu-group__item,.btn.back-button,.tip-table__header .btn{box-shadow:none;} /* The menu row's fill is the whole row,and the row is square. A key's radius rounds that fill into a pill sitting inside a square band - visible only on the two rows that are buttons (the theme toggle and Logout),because the anchors beside them have no radius to begin with. The rounding the list wants is the list's own:.menu-group__list is --radius-xl with `overflow:hidden`,which clips its first and last rows to the corner. */ .btn.menu-group__item{border-radius:0;} /* Ghost:the kind still decides the colour,but the button is outline-only. For a screen with several reads on it - back,tabs,filters - fills would read as a wall of blue and the one write on the screen would stop standing out. The hover and active rules are written `.btn--ghost.btn--read:is(...)` at three classes on purpose. The filled hover rule above is `.btn--read:is(:hover)` - and `:is()` takes the specificity of its most specific argument,so that is 0,2,0,which beats a plain `.btn--ghost` at 0,1,0. Written that way the fill lands on the ghost while its label stays the base hue,and the button becomes its own colour on its own colour:measured at 1.30:1 read,1.42:1 write,1.28:1 delete,which is an invisible label. Three classes outrank two and the ghost keeps its own hover. :active is written at three classes for the same reason:at two it tied with the fill rule and source order decided,which gave outline,then solid hue,then a flat grey press. */ .btn--ghost{background-color:transparent;background-image:none;box-shadow:none;} .btn--ghost.btn--read{color:var(--btn-hue-read);border-color:var(--btn-hue-read);} .btn--ghost.btn--write{color:var(--btn-hue-write);border-color:var(--btn-hue-write);} .btn--ghost.btn--delete{color:var(--btn-hue-delete);border-color:var(--btn-hue-delete);} /* Ghosts,three classes,after everything above - this is what makes the outline-only declaration true in every state rather than only at rest,and it mirrors each hover's colour so :active cannot leave a base-hue label on a subtle well. */ .btn--ghost.btn--read:is(:hover,:focus-visible){background-color:var(--read-subtle);color:var(--on-read-subtle);border-color:var(--read-fill);box-shadow:none;} .btn--ghost.btn--write:is(:hover,:focus-visible){background-color:var(--write-subtle);color:var(--on-write-subtle);border-color:var(--write-fill);box-shadow:none;} .btn--ghost.btn--delete:is(:hover,:focus-visible){background-color:var(--delete-subtle);color:var(--on-delete-subtle);border-color:var(--delete-fill);box-shadow:none;} .btn--ghost.btn--read:active{background-color:var(--read-subtle);color:var(--on-read-subtle);box-shadow:none;} .btn--ghost.btn--write:active{background-color:var(--write-subtle);color:var(--on-write-subtle);box-shadow:none;} .btn--ghost.btn--delete:active{background-color:var(--delete-subtle);color:var(--on-delete-subtle);box-shadow:none;} .btn--small{padding:var(--space-1) var(--space-2);font-size:var(--text-xs);} .input{width:100%;padding:var(--space-2) var(--space-3);border:1px solid var(--border-strong);border-radius:var(--radius-sm);background:var(--surface);color:var(--text);font-family:inherit;font-size:var(--text-base);} .input::placeholder{color:var(--text-muted);} .input:focus{border-color:var(--accent);} /* === Stat readout === These were four coloured numbers floating on --bg,and three separate things were wrong with that. The hue was carrying nothing. A stat value is text,and the app's colour vocabulary already means something specific - green writes,red deletes - so a green "10" beside a green Save button spends a signal on decoration. Worse,"Active Prospects" was painted in --warn-text,which says something is wrong when nothing is. The hue now sits on .stat__mark,a 3px rule:a mark can be any colour without claiming a meaning,and a solid 28x3 bar is a better carrier than the thin strokes of a glyph. The numeral is --text,which is 12.71:1 on the plate against 4.92:1 for the worst of the three it replaced. The value went to 40px and dropped tabular figures. Equal-width digits are for columns that align vertically - and the app still uses them there,in .tip-table__col and the hours tables. A display-size value is not a column:tabular-nums gives every digit the width of a zero,so "121" reads loose at 40px. Every mark colour clears 4:1 against the plate in both themes,over the 3:1 a non-text component needs. The row is one strip rather than three floating tiles so it lines up with the cards under it,which are full content width and were not lining up with a 560px centred block. Inside a card it is not a panel inside a panel - see the .card rule below - for the reason recorded against .tip-table. */ .stat{padding:var(--space-5) var(--space-4);text-align:center;min-width:0;} .stat__value{display:block;font-size:var(--text-3xl);font-weight:600;line-height:1;letter-spacing:-0.02em;color:var(--text);font-variant-numeric:proportional-nums;} .stat__mark{display:block;width:28px;height:3px;border-radius:2px;margin:var(--space-3) auto var(--space-2);} .stat__label{display:block;font-size:var(--text-xs);font-weight:600;color:var(--text-muted);text-transform:uppercase;letter-spacing:0.05em;} .stat--accent .stat__mark{background-color:var(--accent);} .stat--success .stat__mark{background-color:var(--success-text);} .stat--warn .stat__mark{background-color:var(--warn-text);} .stat-group{display:grid;grid-auto-flow:column;grid-auto-columns:1fr;margin-bottom:var(--space-4);background-color:var(--surface);background-image:var(--panel-rim);border:1px solid var(--border);border-radius:var(--radius-xl);box-shadow:var(--panel-bevel);overflow:hidden;} .stat + .stat{border-left:1px solid var(--border);} /* One cell,already on a panel:the plate comes off,or the card holds a card. Reusing .stat-group for both is what keeps the number,the mark and the label identical in the two places,which is the part that has to match. */ .card .stat-group{background:none;border:0;box-shadow:none;border-radius:0;margin-bottom:0;} /* Three labels will not fit across a phone. "ACTIVE CUSTOMERS" measures about 125px at this size and weight,and a third of 375 leaves 82px inside a cell once the padding is off - so the row wrapped to two lines and the cells came back ragged,137px beside a 120px neighbour with nothing to say why. Read sideways instead:one line per stat,number then its mark then its label. It is also the shorter of the two,3 x 48px against 3 x 120px. flex-wrap is for the one in the Referral partners card,whose label is long enough that it still drops to its own line at 320 - right-aligned there by the margin,rather than clipped. */ @media (max-width:560px){.stat-group{grid-auto-flow:row;} .stat + .stat{border-left:0;border-top:1px solid var(--border);} .stat{display:flex;flex-wrap:wrap;align-items:center;gap:var(--space-3);text-align:left;padding:var(--space-3) var(--space-4);} .stat__value{font-size:var(--text-2xl);} /* The auto margin goes on the mark,not the label,so the mark travels with the name it colour-codes instead of stranding itself beside the number. */ .stat__mark{margin:0 0 0 auto;} /* The card's stat keeps the stacked,centred treatment. Its label is the long one,so on a phone it wrapped to a line of its own while its mark stayed up beside the number - the colour code and the name it codes,on two lines,one left and one right. Stacked puts them back in one column. */ .card .stat{display:block;text-align:center;padding:var(--space-4) 0;} .card .stat__mark{margin:var(--space-3) auto var(--space-2);}} .skeleton{background:linear-gradient(90deg,var(--skeleton-base) 25%,var(--skeleton-shine) 50%,var(--skeleton-base) 75%);background-size:200% 100%;animation:shimmer 1.5s infinite;border-radius:var(--radius-sm);} @keyframes shimmer{0%{background-position:200% 0;} 100%{background-position:-200% 0;}} .skeleton--text{height:14px;width:60%;margin-bottom:16px;} .skeleton--title{height:24px;width:40%;margin-bottom:8px;} .skeleton--stat{height:44px;width:100%;} .skeleton--row{height:16px;width:100%;margin-bottom:8px;} .skeleton--row:nth-child(2){width:85%;} .skeleton--row:nth-child(3){width:70%;} .skeleton--h1{height:1rem;} .skeleton--h2{height:2rem;} .skeleton--h25{height:2.5rem;} .skeleton--h3{height:3rem;} .skeleton--h6{height:6rem;} .skeleton--w60{width:60%;} .skeleton--w90{width:90%;} .skeleton--w100{width:100%;} .skeleton-row{padding:var(--space-2) 0;border-bottom:1px solid var(--border-light);} .skeleton-row:nth-child(1){animation-delay:0ms;} .skeleton-row:nth-child(2){animation-delay:50ms;} .skeleton-row:nth-child(3){animation-delay:100ms;} .skeleton-row:nth-child(4){animation-delay:150ms;} .skeleton-row:nth-child(5){animation-delay:200ms;} .skeleton-row:nth-child(6){animation-delay:250ms;} .skeleton-row:nth-child(7){animation-delay:300ms;} .skeleton-row:nth-child(8){animation-delay:350ms;} .skeleton-row:nth-child(9){animation-delay:400ms;} .skeleton-row:nth-child(10){animation-delay:450ms;} .skeleton-list{padding:var(--space-4);} .modal-backdrop{position:fixed;inset:0;background:var(--backdrop);z-index:150;} .modal{position:fixed;left:50%;top:50%;transform:translate(-50%,-50%);z-index:151;width:calc(100% - 2 * var(--space-6));max-width:420px;} /* Level 3. The old shadow here was --shadow-md,a tight 0 1px 3px with a short offset and no spread,and in the render that shape is exactly the thing that reads as HOVERING - a hard drop under a box that is otherwise flat. Replaced with the panel's own material plus a wider,lower,negative-spread lift. The grain is the same as the card's,which is what makes the modal and the real .btn buttons inside it read as the same material. */ .modal__content{background-color:var(--surface);background-image:var(--grain-panel),var(--panel-rim);background-blend-mode:multiply,normal;background-size:160px 160px,100% 100%;border:1px solid var(--border);border-radius:var(--radius-xl);padding:var(--space-5);display:flex;flex-direction:column;gap:var(--space-3);box-shadow:var(--overlay-lift),var(--panel-bevel);} .modal__title{font-size:var(--text-lg);font-weight:600;margin-bottom:var(--space-1);} .modal__actions{display:flex;gap:var(--space-3);flex-wrap:wrap;} .edit-form{display:flex;flex-direction:column;gap:var(--space-4);} .edit-form__card{background-color:var(--surface);background-image:var(--panel-rim);border:1px solid var(--border);border-radius:var(--radius-xl);box-shadow:var(--panel-lift),var(--panel-bevel);padding:var(--space-4);} .edit-form__card-title{font-size:var(--text-xs);font-weight:600;color:var(--text-muted);text-transform:uppercase;letter-spacing:0.05em;margin-bottom:var(--space-3);} .edit-form__checkbox-row{display:flex;align-items:center;gap:var(--space-2);font-size:var(--text-sm);padding:var(--space-1) 0;} .edit-field{margin-bottom:var(--space-3);} .edit-field__label{font-size:var(--text-xs);color:var(--text-muted);display:block;margin-bottom:var(--space-1);} /* The required marker - the `*` beside a label whose input carries `required`. The attribute is the half assistive tech reads;this is the half the eye reads,and after the old `"Name *"` placeholder went away the field had nothing on screen saying it was required until the save was refused. It takes --danger-text rather than --danger,which is the whole reason that token exists:--danger measures 4.44:1 as text,under the floor. The asterisk is the thing carrying the meaning,so it is text and answers to the 4.5 floor,not the 3:1 one a shape would. Measured off rendered pixels at 3x - the core of the glyph against the modal colour of its own crop,not off the tokens,because the page's grain is multiplied into --bg and the add row does not sit on the token:light,on the page #9f1818 on #cbc9c5 4.85:1 light,on --surface #9f1818 on #d5d3d0 5.37:1 dark,on the page #f87171 on #0c0a09 7.14:1 dark,on --surface #f87171 on #1c1917 6.32:1 The marker only ever renders in the add row,which sits on the page;the --surface row is the same glyph with that row's own background painted --surface in the probe,so the card face is covered too. Antialiased edge pixels of the asterisk fall to 3.69:1 in light,as the edges of any 12px glyph do - the core,which is the colour,clears the floor in both themes. The margin is the label's own spacing step. The glyph is not read out:the label is the input's accessible name and `required` is its state,so the marker carries `aria-hidden`. */ .edit-field__required{color:var(--danger-text);margin-left:var(--space-1);} .edit-field__value{font-size:var(--text-sm);} .empty-state,.error-state{text-align:center;padding:var(--space-8) var(--space-4);color:var(--text-muted);} .empty-state__icon{margin-bottom:var(--space-3);} .empty-state__title{font-size:var(--text-lg);font-weight:600;margin-bottom:var(--space-2);} .empty-state__description{font-size:var(--text-sm);margin-bottom:var(--space-4);} .login-page{display:flex;align-items:center;justify-content:center;min-height:calc(100dvh - var(--tab-height));max-width:360px;margin:0 auto;} .login-form{width:100%;padding:var(--space-6);background-color:var(--surface);background-image:var(--grain-panel),var(--panel-rim);background-blend-mode:multiply,normal;background-size:160px 160px,100% 100%;border-radius:var(--radius-xl);border:1px solid var(--border);box-shadow:var(--panel-lift),var(--panel-bevel);} .login-heading{font-size:var(--text-xl);font-weight:600;line-height:var(--leading-tight);margin-bottom:var(--space-6);} .login-field{margin-bottom:var(--space-3);} .login-label{display:block;font-size:var(--text-xs);color:var(--text-muted);margin-bottom:var(--space-1);} .login-input{width:100%;padding:var(--space-3) var(--space-4);font-size:var(--text-base);border:1px solid var(--border);border-radius:var(--radius-sm);background:var(--surface);color:var(--text);font-family:inherit;} .login-input:focus{border-color:var(--accent);} .login-input::placeholder{color:var(--text-muted);} /* Layout only - the fill,the label,the edge and the press all come from .btn--read. Those were written out here as --accent-fill/--on-accent,which hold the same values as --read-fill/--on-kind today,so the button painted the read blue by coincidence and would have stopped following the rule the moment either token moved. There is no :active rule for the same reason:the press is .btn--read:active's,and a background here would flatten it. */ .login-btn{width:100%;margin-top:var(--space-4);padding:var(--space-3) var(--space-4);font-size:var(--text-base);font-weight:600;font-family:inherit;cursor:pointer;} .login-error{color:var(--danger-text);font-size:var(--text-sm);margin-top:var(--space-2);min-height:1.4em;} .table-scroll{overflow-x:auto;} /* Phone-first:on narrow screens the accounts list drops the Website column so Name and the tappable Phone number both fit without horizontal scrolling. Scoped to this table by id suffix;other a8b-tables have different columns. */ @media (max-width:560px){[id$="_accts"] th:nth-child(2),[id$="_accts"] td:nth-child(2){display:none;} /* === Touch targets === --hit is 24px,which is WCAG 2.2 AA's target size (2.5.8) and what every control is built to. That is the right floor for a mouse and too small for a finger,so on a phone the floor is --hit-touch instead. THE RULE THAT WAS HERE DID NOT WORK. It read `.icon-btn{min-height:var(--hit-touch)}`,one class,and the rule it had to beat is `.btn.icon-btn,.icon-btn` at line 688 - two classes. A media query adds no specificity,so (0,1,0) lost to (0,2,0) at every width and the touch floor had never applied to a single real icon button,because Lib.Button.view always emits `btn` alongside `icon-btn`. Measured at 390px:the app's own icon buttons were 24x26. The same markup without `.btn` measured 44x44,which is what the rule was written for and never reached. So each selector below is written at the specificity of the rule it has to beat,not at the specificity of the element it describes - the same trap this stylesheet's button catalogue documents and has now hit nine times. Measure any change here;none of them is visible in a screenshot. WHAT EACH ONE IS:.btn.icon-btn the queue's edit key,the feed's pencil,stickies' delete .btn.chip the filter chips on the queue and the reports .btn every other control,36px tall before this .link-button a name or id that acts as a link inside a cell or a card:the account on a log row,a person on the Hours screen. It sets min-height:0 on purpose - its line box is the table row's - so it has to be raised explicitly. .task-row__more-summary the queue row's `[⋯]`. It carries no `btn` and no kind - it is a disclosure,not a control that writes - so none of the four above reach it,and it was 28px on a phone. THE SECOND SELECTOR IS THE ONE THAT MATTERS:the compact density sizes this same box at (0,4,0),so a rule written at the summary's own (0,1,0) loses to it at every width and the touch floor never applies to a compact queue - which is the identical mistake the note above describes. */ .btn.icon-btn,.icon-btn{min-width:var(--hit-touch);min-height:var(--hit-touch);} .btn.chip,.chip{min-height:var(--hit-touch);padding:10px 16px;} .btn{min-height:var(--hit-touch);} .btn.link-button{min-height:var(--hit-touch);} .task-row__more-summary,:root[data-density="compact"] .task-list .task-row__more-summary{width:var(--hit-touch);height:var(--hit-touch);} /* The contacts search is sized in the last block of this file,not here - see the note there. */ /* A small button is 24px tall on 12px type - the Delete on every table row,and the smallest thing in the app. At the touch floor its type comes up to the body size with it:12px is under the 14px legibility floor this stylesheet's own button rules set,and a 44px box around 12px type is a big target with a small word in it. */ .btn--small{font-size:var(--text-sm);padding:var(--space-2) var(--space-3);}} /* One table,wherever it appears. The dashboard's cards already looked like this - a bordered rounded box,an uppercase muted header,light row rules - and the rest of the app's tables did not,so the same data read as two different components depending on the page. These are the dashboard's values,which live on .tip-table for the cards that are built from flex rows rather than a <table>. */ /* Level 1,the panel,with a lighter grain than a card:a table is more face and less margin,and at the card's 0.18 the row rules stop reading as rules against the tooth. The row rules move with the box edge in the same edit - .crm-table td's rule and --border-light are one decision,and the table reads as a grid drawn on paper if only one of them moves. */ /* === The panel === The box every section of this page is drawn in:the activity feed,the stickies list,the AMES list,the sub-tab tables. A screen whose content is NOT in one reads as unfinished rather than as a different shape,which is what the Email tab looked like next to the five tabs beside it - loose fields on the page background while everything else sat in a box. It is its own selector group rather than a third copy of the same declarations. The chrome was already written out twice,in .crm-table and again on the AMES list,and a third copy for a form is the one that should not happen. Each of the three keeps its own geometry below - the table its width and text size,the list its hug,the form its padding and gaps - and takes only the surface from here. */ .crm-panel,.crm-table,.ames-reports__list,.email-tab{border:1px solid var(--border);border-radius:var(--radius-xl);overflow:hidden;background-color:var(--surface);background-image:var(--grain-table),var(--panel-rim);background-blend-mode:multiply,normal;background-size:160px 160px,100% 100%;box-shadow:var(--panel-lift),var(--panel-bevel);} .crm-table{width:100%;border-collapse:collapse;font-size:var(--text-sm);/* `overflow:hidden` clipped a table wider than this box - silently,and one level below .table-scroll,so the wrapper that exists for sideways scrolling reported clientWidth == scrollWidth and drew no scrollbar to find the lost columns with. Measured on cepending at 1898 wide:the Size header sat at x 1569..1637 with this box's right edge at 1532,105px outside a clip nothing could reach. What holds it now is the two caps on the cells that can grow without bound rather than a rule on the table:`.cell-link` stops a ninety-character url at 30ch and `.account-name` wraps,so the table fits the box it is in instead of being cut to it. The hidden clip itself is on `.crm-panel` above,with the rest of the surface. This used to point at "the fixed-layout rule on .report-view .crm-table table below". THERE IS NO SUCH RULE and there never was - no `table-layout` declaration exists anywhere in this file. It was a forward reference to a fix that shipped a different way,and it read as an explanation for two years' worth of edits to this block. */} /* Row padding is the scale's next step up from where it was (8 -> 12),and the figures are tabular:a column of numbers that changes width as it changes value cannot be scanned down. This is the same choice the timer ids and the tip table's columns already make;it is the table's turn. */ /* A CELL THAT CANNOT BREAK PUSHES THE WHOLE TABLE OUT OF ITS BOX. Measured on the account page's Activity tab,2026-10-15,at 1440:a log note reading "Landing page updated:<a ninety-character url>" made the table 944px wide inside a 748px box,so 197px of it - the Edit key included - was drawn outside the frame the reader sees,with no scrollbar to reach it. The note is free text somebody typed,and a url has no break opportunity in it,so the column's min-content width was the width of the url. `anywhere` and NOT `break-word`,and that is the whole of why this works:`overflow-wrap:break-word` does not affect a box's min-content size,and min-content is what a `display:table` box measures itself by - it would have wrapped the text and left the table exactly as wide as before. `anywhere` counts the breaks when the intrinsic size is worked out. On the report views this changes nothing:their cells are `white-space:nowrap` on purpose (one line per cell is what keeps a twelve-row report from being twelve double-height rows),and nowrap has no breaks to offer. TWO RULES,because `anywhere` on every cell fixed the table and cost the columns beside it. It counts breaks when the intrinsic size is worked out,so with it on all four cells every column's minimum was one character:the labels collapsed to "Us er" and "2026-" over three lines of value,visible at 3x on the same screen. `break-word` keeps the word-sized minimum. So the fallback is `break-word` everywhere - it still stops a long token spilling out of a cell that is already as narrow as it can be - and `anywhere` is spent only on the cells that can hold a token with no break in it at all. AND THOSE CELLS ARE NAMED,NOT "ANY CELL WITH A LINK". The first version keyed `anywhere` on `:has(a)`,on the reasoning that a url is a link - and a url in a cell is a link,but so is an account name. Reported 2026-10-19:"Atomic8Ball" in the queue's Customer column drew as "Atomic8Bal" with the last letter on the next line,because `anywhere` had dropped that column's minimum width to one character and auto layout took it at its word. A name is a label and sizes its column;only free text that can contain a url needs to shrink past one. Four renderers linkify their text and exactly two of them are table cells:the queue's Description and the activity feed's Note. */ .crm-table th,.crm-table td{padding:var(--space-3) var(--space-4);border-bottom:1px solid var(--border-light);text-align:left;overflow-wrap:break-word;} .crm-table td[data-label="Note"],.crm-table td[data-label="Description"]{overflow-wrap:anywhere;} /* A TASK ID NEVER BREAKS. "109656" split over two lines reads as a different number,and an id is the thing people quote to each other. Measured on the queue,2026-10-16:at 900 and at 700 the ID cell - which holds the timer link,so the rule above hands it `anywhere` - had been squeezed to 60px and the number drew on TWO lines. The report tables never showed it:they are `white-space:nowrap` already. `nowrap` is the fix rather than a narrower `overflow-wrap` because it does both halves:it stops the break,and it puts the whole number back into the column's minimum width,so the column cannot be squeezed below its id again. The label is the hook because it is the one the app already has for this - the stacked phone layout reads the same attribute - and it is the same scope as that block,so a table that labels a column "ID" is covered wherever it lives. Three do:the queue's library cells,the done-tasks report,and the hand-rolled table in the add form. Every one of them holds a task's number and nothing else. */ :is(.crm-table,.report-table,.task-table,.task-add__tasks) td[data-label="ID"]{white-space:nowrap;} .crm-table{font-variant-numeric:tabular-nums;} .crm-table tbody tr:last-child td{border-bottom:none;} /* Cells hold short fixed-shape values - a timestamp,a count,a dash - and a date and time breaking across two lines took every row with it,so a twelve-row report was twelve double-height rows. One line per cell measures 56px rows down to 37. Report views only,and that is not a style split - the pager,the search box and the scrolling are shared by every table. This is about content:report cells hold a date,a count or a dash,and one line each is right for them. The task queue's Description is prose,and a cell that cannot wrap cannot shrink,so forcing it to one line ran the table to 1687px and pushed the right-hand columns off the edge again. Prose wraps;values do not. */ .report-view .crm-table td{white-space:nowrap;} /* FREE TEXT IS NOT A FIGURE. The rule above is for dates,counts and dashes - one line per cell is what keeps a report from being double-height rows. A done task's name and description are neither:measured 2026-10-19 at 1440,an 87-character name made the table 1790px wide inside a 1328px box,and `.crm-table` CLIPS (overflow hidden),so the rest of the row was unreachable - the cepending failure,still open for these two columns. Wrapping them lets the table size to its box again. */ .report-view .crm-table td[data-label="Task"],.report-view .crm-table td[data-label="Description"]{white-space:normal;overflow-wrap:anywhere;} /* A report's URL column is the widest thing on any of these screens:one campaign URL is ninety-odd characters,and unbroken it measured 859px - more than half the content column on its own,which pushed the table past the page and carried Scheduled and Size off the right edge behind a horizontal scroll. Cut with an ellipsis rather than shortened in the markup,so the anchor's text in the DOM is the whole URL - what a screen reader reads,what a copy takes,and what the href points at. The title carries it for a pointer. 32ch is the width at which the six cepending columns fit the content column together;see the note on the token. */ .report-view .crm-table td a.cell-link{display:inline-block;max-width:var(--link-max);overflow:hidden;text-overflow:ellipsis;white-space:nowrap;vertical-align:bottom;} /* The one genuinely variable-length field. A long company name may still wrap rather than stretching its column and dragging the table sideways. */ .report-view .crm-table td .account-name{white-space:normal;} /* The pull preview's Description,which is the first genuinely unbounded prose on a report table - a task description is a paragraph,and the queue's own description is the case the rule above already carves out for .task-row. Capped as well as wrapped,which .account-name is not:names are short enough that wrapping alone bounds them,and a description is not. Without the cap the cell wraps only if something else constrains the table,and above 1101px `.crm-table table` is `display:table` - shrink to fit - so a long one lays out on a single line and takes the five columns beside it off the page. */ .report-view .crm-table td .pull-preview__desc{display:inline-block;max-width:var(--prose-max);white-space:normal;} /* THE ROW A SEARCH LEAVES BEHIND. When a filter empties a table the library draws this one cell,spanning the columns,saying why - headers over a blank body read as "there is nothing here" rather than "nothing matches". Muted,because it is not data,and not uppercased,because it is a sentence rather than a column heading. */ .crm-table .table-nomatch td{color:var(--text-muted);padding-block:var(--space-4);} .crm-table th{font-weight:600;color:var(--text-muted);font-size:var(--text-xs);text-transform:uppercase;letter-spacing:0.05em;border-bottom:1px solid var(--border-light);background:var(--th-band);cursor:pointer;user-select:none;} .crm-table th:hover{color:var(--text);} /* The sort label presses too. It is the app's most-pressed control after the pager and it had no press at all - a header that changes the order of forty rows under the finger and does not move is the one that reads as not having taken the click. It travels and nothing more:there is no face to press into (the rule below strips this button back to bare type on purpose) and no dim,because dimming the label is dimming text - `--text-muted` already sits at 4.99:1 on the header band and a filter would spend part of that margin for a cue the 1px already gives. */ .crm-table th button:active{transform:translateY(1px);} .crm-table th[aria-sort]:not([aria-sort="none"]){color:var(--text);} /* The header label is a <button> so the sort has a keyboard path. It is a placement,not a control with a face:the <th> above it is what carries the type metric,the padding and the clickable area,so this strips the UA's button chrome back to nothing and hands the type back to the cell. Anything left on it - the UA's `font:400 13.333px Arial`,its buttonface fill and its 2px outset - would make every header in the app a different size from the cells under it. `color:inherit` is what lets `th[aria-sort]` above brighten the sorted column's label:the rule lands on the <th>,and the label is one level in. */ /* Inline-FLEX,and that is the fix for a real defect:as an inline-block with normal wrapping,a label that filled its column pushed the sort triangle onto a second line,so the header row's height changed from table to table and the arrow stopped pointing at the word it sorted - measured on the queue,where Assigned,Created and Status all wrapped. A flex row cannot wrap its items,so the label wraps inside its own box and the triangle keeps its place beside it. The label itself still wraps:`white-space:inherit` is untouched. */ .crm-table th button{font:inherit;color:inherit;letter-spacing:inherit;background:none;border:0;padding:0;cursor:pointer;white-space:inherit;display:inline-flex;align-items:center;} /* A key's ring,the same one .detail-header__action-btn draws. The UA's is a different ring in every browser and does not follow the theme. */ .crm-table th button:focus-visible{outline:2px solid var(--focus-ring);outline-offset:2px;} /* The indicator,and its whole reason for being a pseudo-element is that it cannot reach th.textContent - sorTable resolves the clicked column by looking the header's text up in a label-to-key map,so a glyph in the markup would turn "Account" into "Account v",miss the map,and no-op every sort in the app. pseudo-element content is not text. Drawn as borders rather than a character for the reason the pager is drawn rather than typed:a border is currentColor so it follows the header in both themes for free,and it cannot be emoji-presented or go missing with a font. */ /* `aria-sort` is on every heading that CAN be sorted,with "none" for the ones that are not - which is what the attribute is for,and what a screen reader wants to hear. The indicator therefore keys on the VALUE:keyed on the attribute's presence it drew an up-triangle on every heading in the table and left the sorted one indistinguishable. */ /* The triangle takes the brand mark rather than currentColor. Which column is sorting is a state of the table,and the label beside it already brightens to --text to say the same thing in type;the mark says it in colour,so the sorted column is findable without reading the header row. A mark on a ground is held to 3:1,and this one measures 4.1:1 on the header band in light and 3.7:1 in dark. */ .crm-table th:not([aria-sort="none"]) button::after{content:"";display:inline-block;margin-left:var(--space-1);border-left:var(--space-1) solid transparent;border-right:var(--space-1) solid transparent;border-bottom:var(--space-1) solid var(--mark);} .crm-table th[aria-sort="descending"] button::after{border-bottom:0;border-top:var(--space-1) solid var(--mark);} .crm-table tbody tr:active{background:var(--surface-hover);} /* The table library renders its own toolbar - a rows-per-page select,a search box,a clear control - and its own pager. Both are styled here rather than in the library so this site's tables read like the rest of the app. */ /* The toolbar renders inside the table's own box,so it needs the padding the box corners would otherwise take from it,and its labels have to line up with the column text rather than sitting out against the border. The first label is inset by the cell padding for that second reason. */ .table-search,.crm-table label{display:inline-flex;align-items:center;gap:var(--space-1);margin:var(--space-3) var(--space-3) var(--space-3) 0;font-size:var(--text-xs);color:var(--text-muted);} .crm-table label:first-child{margin-left:var(--space-4);} /* The library writes its labels as the field names,lowercased. Capitalise the label span only:the same rule on the label would also capitalise what someone types into the search box. */ .table-search>span,.crm-table label>span{text-transform:capitalize;} /* The table library's clear-search control is a bare "x" character,which laid out at 8x20 - far under the 24x24 WCAG 2.2 target-size minimum,and the smallest thing to hit on the whole table. It gets a real box here. */ /* The pager and the clear-search control draw Tabler icons now instead of glyph characters,so they take the same sizing and colour treatment as the app's own icons. Shared by every .crm-table - accounts,the task queue and the reports all get this from here rather than each page styling its own. */ .table-icon{display:inline-block;vertical-align:middle;flex-shrink:0;} .table-search>a.search-clear,.table-search>a,.crm-table label>a.search-clear,.crm-table label>a{margin-left:var(--space-1);color:var(--text-muted);text-decoration:none;display:inline-flex;align-items:center;justify-content:center;min-width:28px;min-height:var(--hit);border-radius:var(--radius-md);/* The clear sits level with the field it clears. At the 24px hit floor it was a small mark floating beside a 32px box;32 makes the pair one object,and the floor is still the floor. */ align-self:stretch;} .table-search>a:hover,.crm-table label>a:hover{color:var(--text);background:var(--surface-hover);} /* The clear control had a hover and no press. It takes the faceless press - the 1px and the dim,on top of the fill its hover already sets - because it has no box of its own to press into:it sits inside the field's own border. */ .table-search>a:active,.crm-table label>a:active{color:var(--text);background:var(--surface-hover);transform:translateY(1px);filter:var(--press-dim);} /* THE FIELDS ARE THE APP'S READING SIZE,NOT THE SMALLEST ONE. They were 12px - the label size - while the cells they search are 14 and every other field in the app is 16. A search box is type a person reads as they type it,so it takes --text-sm:14 on a desk,and 16 on a phone because that is what the token becomes there. The radius and the well arrive from the field-material rule at the end of this file;these two declare their own edge,which is why that rule is written last. AND THEY TAKE A FOCUS STATE. Every other field in the app moves its edge to --accent when it is focused;these took only the browser's ring,so the two controls a person types into most often were the two that looked least alive. --accent on --surface is a graphical mark at 7.24:1. */ .table-search>input,.crm-table label>input{padding:var(--space-1) var(--space-2);font-size:var(--text-sm);border:1px solid var(--border-strong);border-radius:var(--radius-sm);background:var(--surface);color:var(--text);font-family:inherit;min-height:32px;} .table-search>select,.crm-table label>select{padding:var(--space-1) var(--space-2);font-size:var(--text-sm);border:1px solid var(--border-strong);border-radius:var(--radius-sm);background:var(--surface);color:var(--text);font-family:inherit;min-height:32px;} .table-search>input:focus,.table-search>select:focus,.crm-table label>input:focus,.crm-table label>select:focus{border-color:var(--accent);} .crm-table nav.paging{border-top:1px solid var(--border-light);padding-top:var(--space-3);margin-top:var(--space-4);} .crm-table nav.paging ul{display:flex;align-items:center;justify-content:center;gap:var(--space-2);list-style:none;margin:0;padding:0;} .crm-table nav.paging li{display:flex;align-items:center;} /* Four of the six pager glyphs (U+23EE U+25C0 U+25B6 U+23ED) are text by default,but U+23E9 and U+23EA are Emoji_Presentation=Yes - so on a platform with an emoji font those two drew as colour emoji beside four monochrome triangles. Forcing text presentation makes all six agree,and they now take the same colour as the rest of the app's chrome. */ /* A pager button is a key and takes the control material,which it did not before:it was the last boxed control in the app still on --radius-sm and with no depth at all,so a table's page controls were flat 4px rectangles beside every other control's 6px raised key. These are the table library's own elements rather than Lib.Button's,which is why the material is written here instead of arriving with a kind - but nothing about that makes them a different object. The colours stay the pager's own. Its hover and active move the BORDER to the accent rather than the fill,which is a navigation signal and not a database one - paging reads nothing and writes nothing,so it has no kind to take. Background-color is written longhand throughout for the reason the chip's is:the `background` shorthand would drop the grain and the sheen with it. */ .crm-table nav.paging button{min-width:44px;min-height:44px;padding:6px 10px;border:1px solid var(--border-strong);border-radius:var(--radius-md);background-color:var(--surface);background-image:var(--grain),linear-gradient(180deg,var(--sheen-top),var(--sheen-bottom));background-blend-mode:overlay,normal;background-size:160px 160px,100% 100%;color:var(--text-muted);cursor:pointer;font-size:var(--text-sm);font-family:var(--font);font-variant-emoji:text;line-height:1;box-shadow:var(--shade),var(--btn-bevel);} .crm-table nav.paging button:hover{color:var(--text);border-color:var(--accent);background-color:var(--surface-hover);} /* The library's keys take the app's press,not a shallower one of their own. They were written before `--btn-press` existed and kept the old 2px inset with no travel and no dim,so the six controls a reader uses most often on any list were the one place a press went only halfway. The library builds these as plain <button>s,so this rule is the only way they can have one. */ .crm-table nav.paging button:active{background-color:var(--accent-subtle);border-color:var(--accent);transform:translateY(1px) scale(0.98);box-shadow:var(--btn-press);filter:var(--press-dim);} /* NOT the brand face,and the reason is a measurement rather than a taste:the vendored pager (lib/vendor/table.mjs,pageNavigation) renders first / back-ten / back-one / a page-number FIELD / next-one / next-ten / last,and no element of it carries aria-current - so this rule matches nothing on any table in the app,and a brand face here would be a claim about a mark nobody can see. The rule is left as it was for the library version that once marked the page. The page you are on is stated by the number in that field,which is a field and keeps a field's face. */ .crm-table nav.paging button[aria-current]{background-color:var(--accent-fill);background-blend-mode:multiply,normal;color:var(--on-accent);border-color:var(--accent-fill);font-weight:600;} .crm-table nav.paging label{display:flex;align-items:center;gap:var(--space-1);font-size:var(--text-xs);color:var(--text-muted);} .crm-table nav.paging input{width:48px;min-height:44px;padding:6px;font-size:var(--text-sm);border:1px solid var(--border-strong);border-radius:var(--radius-sm);background:var(--surface);color:var(--text);font-family:inherit;text-align:center;} /* Phone:seven 44px paging buttons overflow the screen,so keep prev/next and the page number and drop the first/last and +/-10 jumps. This block sits after the paging rules above on purpose - it matches them at equal specificity and has to win on source order. The 16px input size is the iOS zoom guard:a focused field under 16px makes Safari zoom the viewport. */ @media (max-width:560px){/* `.table-search>input` is in this list and was not before. The zoom guard covers every field on a phone,and the search box is the one people tap first - its parent is a span rather than a label,so the label selector never reached it and it stayed at 12px,which is the size that makes Safari zoom the viewport out from under the tap. Measured at 390px before it was added:the rows-per-page select was 16px and the search box 12px,one line apart in the same toolbar. */ .crm-table label>input,.crm-table label>select,.table-search>input,.crm-table nav.paging input{font-size:var(--text-base);} /* === The toolbar on a phone === It was two ragged rows. Measured at 390px:"Rows" sat at x=33 - its own `margin-left:--space-4` - and the search box beneath it at x=17,so the two left edges did not line up;the search stopped at 270px inside a 356px box,leaving 86px of nothing;and the clear control floated at x=243 with dead space to its right,reading as a third thing rather than as part of the field. They stay two rows,and that is a decision rather than the old behaviour kept. ONE ROW WAS TRIED FIRST and measured:the page size is 97px with its word,so at 390px the search field came down to 163px and at 320px to 93px - narrower than the 218px it had before,on the one control in the toolbar that is actually typed into. Dropping the word "Rows" to buy the space back was the other option and it leaves a bare "10" with nothing saying what it counts. So the search gets the width instead:the page size keeps its own row,and the field spans the whole card beneath it - 280px of input at 390 against 218 before. Both rows are inset by `--space-4`,the padding the table's own cells use,so the toolbar,the cells and the pager start on one line down the card instead of two. Grid and not flex,because the label and the search are siblings of the table and the pager inside the library's own section - the section is the row box whether it wants to be or not,and a flex row there would put the table beside them. Naming the rows keeps the DOM order,so the tab stops still run label-then-search down the screen. `minmax(0,1fr)` on the search row and not `1fr`:a grid track's minimum is auto,which is a content floor,and a long search term would push the row past the card instead of letting the field shrink. */ .crm-table>section.a8b-table{display:grid;grid-template-columns:minmax(0,1fr);align-items:center;} /* THE SECTION HAS TWO LABELS,NOT A LABEL AND A `.table-search`. This block first read `> label` for the page size and `> .table-search` for the search,and that was written from the STYLESHEET rather than from what the library emits. The library builds its search with `field()`,which returns `html('label',{},[span,input])` - a plain classless label - so `.table-search` matched nothing,and `> label` matched BOTH labels and stacked them in the same grid cell at row 1. The two controls drew on top of each other on every table in the app. `.table-search` is real,and it is the DASHBOARD's - `searchControl` puts it on a label of the same shape. So it is a live class with no relation to the library's markup,which is exactly how it came to be trusted here. The search label is therefore identified by what it HOLDS rather than by where it sits or what it is called:it is the one carrying the clear key. That cannot be wrong about which label is which,and the two labels have no class to tell them apart. Verified against the real library before writing this - four earlier measurements of this block used hand-written markup with `class="table-search"` on it,which is markup no table in the app has ever produced. */ .crm-table>section.a8b-table>label:first-of-type{grid-column:1;grid-row:1;margin:var(--space-3) var(--space-4) 0 var(--space-4);} .crm-table>section.a8b-table>label:has(a.search-clear){grid-column:1;grid-row:2;margin:var(--space-2) var(--space-4) var(--space-3) var(--space-4);position:relative;/* A flex item's automatic minimum is its content,so without this the label refuses to shrink below the width of its own word plus the input's intrinsic size - measured on the dashboard's copy at 320px,where it came out 267px inside a 262px row and hung 5px past the card. */ min-width:0;} .crm-table>section.a8b-table>table{grid-column:1;grid-row:3;} .crm-table>section.a8b-table>nav.paging{grid-column:1;grid-row:4;} /* The field takes the row and gives back the width an input has by default - about twenty characters,which would otherwise set the row rather than fill it. NO `width:100%`:a grid item already stretches to its cell,and adding it made the label 100% of the cell AND kept its margins,so it came out 32px too wide - the field's right edge landed past the card and `.crm-table`'s `overflow:hidden` cut the search box in half. */ .crm-table>section.a8b-table>label:has(a.search-clear)>input{flex:1 1 auto;min-width:0;width:auto;padding-right:calc(var(--hit-touch) + var(--space-2));} /* The clear key goes INSIDE the field. As a sibling of the input it sat past the field's right border with a gap between them,reading as a third control standing beside the search rather than as part of it. Every search field a person has used puts it inside,at the right edge,so the label becomes the positioning context,the key leaves the flow,and the input takes right padding so a long term cannot run under it. The row height does not change - the key is out of flow and the 44px input sets it. */ .crm-table>section.a8b-table>label:has(a.search-clear)>a.search-clear{position:absolute;right:var(--space-4);top:50%;transform:translateY(-50%);margin-left:0;} /* === The dashboard's searches === `searchControl` builds the same label-then-input shape the table library does,carrying `.table-search`,and it is used by all three dashboard cards that have one - TIP,referral partners and Recent Logs. On a phone its label is already wrapping under the card's title,because `.card__head` is a wrapping flex row and there is no room beside a heading. What it did not do was take the room it wrapped into:measured at 390px it sat at x=29 in a card whose inner edge is 377 and stopped at 296,leaving 81px of dead space to the right of the field,and its input started at x=78 - indented past the word "search" rather than filling the row. `flex-basis:100%` makes the wrapped line its own,so the field spans the card;the input then takes what the label's word leaves. Only under 560px:above it the title and the search share one line and `justify-content:space-between` is what places them. */ .card__head .table-search{flex:1 1 100%;min-width:0;} .table-search>input{flex:1 1 auto;min-width:0;width:auto;} .table-search{position:relative;} .crm-table nav.paging ul{gap:var(--space-1);} .crm-table nav.paging li:nth-child(1),.crm-table nav.paging li:nth-child(2),.crm-table nav.paging li:nth-child(6),.crm-table nav.paging li:nth-child(7){display:none;} /* The rest of the table toolbar,at the touch floor. These belong HERE and not in the touch block further up the file,and that is not tidiness:the rules they have to beat are the toolbar rules immediately above at equal specificity,so a copy placed earlier loses on source order and silently does nothing. Measured with the rules in the wrong block - the rows-per- page select stayed 32px and the clear control stayed 28x24. */ .crm-table label>input,.crm-table label>select,.table-search>input,.table-search>select{min-height:var(--hit-touch);} .table-search>a.search-clear,.crm-table label>a.search-clear{min-width:var(--hit-touch);min-height:var(--hit-touch);}} /* THE TOAST LIVES AT THE BOTTOM NOW,and the sweep is why. It used to be fixed at the top right,which is where the app keeps its own controls:measured on the account page,Edit / Quick Log / Payment sit at y 16-40 with their right edge at 1303 and the container started at y 16 and reached 1424 - over the three buttons for the four seconds a toast is up. Nudging it down one row fixed that and put it over the New Account form's first fields instead,which is the same problem one row later:a 320px panel in a top corner will always be on top of something a person is using. The bottom is free on the desktop - pages are anchored to the top and the rail is on the left - and on a phone the bar is there,so the container sits above it and drops to the floor when there is no bar,the way the body's own padding-bottom already does. Found by sweeping every page for overlapping boxes rather than by looking:a toast is absent from almost every screenshot,and the hits are with controls,not with text. */ .toast-container{position:fixed;right:var(--space-4);bottom:calc(var(--tab-height) + env(safe-area-inset-bottom,0px) + var(--space-4));display:flex;flex-direction:column;gap:var(--space-2);z-index:200;max-width:320px;} body:not(:has(.tab-bar)) .toast-container{bottom:var(--space-4);} .toast{background-color:var(--surface);background-image:var(--panel-rim);border:1px solid var(--border);border-radius:var(--radius-xl);padding:12px 16px;box-shadow:var(--overlay-lift),var(--panel-bevel);animation:slideInRight 300ms ease-out both;font-size:var(--text-sm);cursor:pointer;} .toast--success{border-left:3px solid var(--success);} /* The one toast that copies:a button,so it has to say it is one. Laid out as the row the other toasts are,with the copy glyph at the end where the eye looks for what the click will do. */ .toast--copy{display:flex;align-items:center;gap:var(--space-3);width:100%;justify-content:space-between;text-align:start;font-family:inherit;font-size:var(--text-sm);color:var(--text);} .toast__copy{display:inline-flex;color:var(--text-muted);} .toast--error{border-left:3px solid var(--danger);} .toast--info{border-left:3px solid var(--accent);} @keyframes slideInRight{from{transform:translateX(100%);opacity:0;} to{transform:translateX(0);opacity:1;}} @keyframes slideInRight{from{transform:translateX(100%);opacity:0;} to{transform:translateX(0);opacity:1;}} /* THE NAV MARKS ARE INK,AND THIS IS THE RULE ALWAYS MEANT TO WIN. `.tab-bar .tab-btn{color:var(--nav-text)}` is written at the top of each of the bar's two blocks - the phone bar's and the rail's - and both comments record the same defeat:"the kind's colour still wins on source order (the .btn--ghost.btn--read pair is also two classes,and it is later)". It did. So the rail and the phone bar have been five action-blue marks in a column,the same blue every button that reads or writes wears,with the selected chip the only one of them saying anything else. Three selectors beat the pair,and the scoping does two jobs. THE NAV LIST,not the whole bar. The foot's controls are anchors and buttons with tab-btn on them too,and one of them is the DeepSeek light,whose colour IS its state:an ink rule over the whole bar would repaint an outage grey,and that light is the one mark here that has to stay loud. AND IT IS TOP LEVEL,not inside either block. Written into the rail's block it applied above 768 only - measured at 414 the phone bar's marks were still rgb(26,61,168) - because a media query adds no specificity and simply does not match below its width. One rule at both widths is the point. Measured after,on the rail's own face (#d5d3d0 light,#1c1917 dark):--nav-text is 5.70:1 in light and 11.74:1 in dark,where the blue it replaces measured 4.90 and 4.75. The mark got quieter and more legible at once,which is what a column of destinations wants. */ .tab-bar ul .tab-btn{color:var(--nav-text);} .tab-bar ul .tab.active .tab-btn{color:var(--mark-ink);} .error-state{display:flex;flex-direction:column;align-items:center;gap:var(--space-2);} .error-state__icon{color:var(--danger);} .error-state__message{font-size:var(--text-sm);color:var(--text);} /* === The skip link === First in the document,so it is the first tab stop,and held above the viewport until it takes focus - at which point it lands over the rail's top corner,where the eye already is when the keyboard is being used. z-index 200 clears the rail's 100 and the FAB's 150;it is not a layer,it is a control that happens to be fixed. Built through Lib.Button like every other control,so it is a Read key and wears the read family when it appears. */ .skip-link{position:fixed;top:var(--space-3);left:var(--space-3);z-index:200;transform:translateY(calc(-100% - var(--space-6)));} .skip-link:is(:focus,:focus-visible){transform:none;} /* The account header when its load FAILED,as opposed to still running:the message the load reported,which names the route,and the way to ask again. The skeleton above is the loading state and stays that. */ .detail-header__failed{display:flex;align-items:center;gap:var(--space-3);flex-wrap:wrap;} .detail-header__failed-msg{margin:0;color:var(--danger-text);font-size:var(--text-sm);} /* === Surface and material language:level 2,the field === Every text field,select and textarea in the app,in one rule,and it is last in the file on purpose:these are all (0,1,0) selectors and every one of them already declares its own border,so at equal specificity the later rule is the one that wins. Written here rather than edited into each of the twelve rules above so the family cannot drift apart - a field that keeps a 4px radius or a --border edge while its neighbours move is exactly the half-applied material this pass exists to avoid. Two things change,and the second is not cosmetic:- --radius-md,not --radius-sm. The radius scale puts every control at 6px and reserves 4px for chips and badges. A field at 4px beside a 6px button reads as a different family of object. - --border-strong,not --border. --border is a hairline now and a hairline is NOT a control boundary. --border-strong is the WCAG 1.4.11 non-text boundary for a control - measured 4.80:1 on --surface light and 3.78:1 dark,both over the 3:1 the criterion requires - and a shadow or a sheen cannot be relied on to supply that. Fields that already used it keep it;the ones that used --border move up to it,which also fixes a pre-existing failure:at #d6d3d1 on white the old edge measured under 1.5:1. No grain,deliberately,and it is the same finding as the button's:on a 44px field the fine panel grain reads as speckle behind the label rather than as material. A control does not have the area for texture to establish itself. */ .input,.search-bar__input,.login-input,.report-months,.table-search>input,.table-search>select,.crm-table label>input,.crm-table label>select,.crm-table nav.paging input{border-color:var(--border-strong);border-radius:var(--radius-md);box-shadow:var(--well);/* The face is the field's own token now - see --field-face. Written here,with the rest of the family's material,so the fourteen rules above that still say `background:var(--surface)` cannot drift apart again. */ background-color:var(--field-face);} /* === The control material:the boxed --surface families === Three controls whose face is --surface rather than a kind fill,in one rule and last in the file for the same reason the field family above is:they are all (0,1,0) selectors,each of them already declared its own border,radius and background,and at equal specificity the later rule is the one that wins. Written here rather than edited into the three rules above so the family cannot drift apart - which is what it had done. These are the account header's Edit and Quick Log keys,the timer's task number and the CE credential remove,and they were three different objects:two carried `box-shadow:none` from a reset that also covered two families that genuinely have no face,and all three carried their own border colour and their own radius. The header's keys and the Payment anchor beside them had a --border hairline where every other control in the app has --border-strong. What they were missing is exactly the material,so what they take is exactly the material and nothing else. Their shape stays theirs:the small type,the 4/12 padding,the 34px square,the number's tabular figures. The sheen is here and the GRAIN IS NOT,and those two are different decisions. Grain on --surface is grime - measured,twice - so `background-image` carries the gradient alone and the blend list has one value. The sheen is the curvature,it is what a neutral .btn already had,and leaving it off is what made a header key a flat white rectangle beside a .btn with a face. */ .detail-header__action-btn,.timer__copy,.ce-cred__remove{background-color:var(--surface);background-image:linear-gradient(180deg,var(--sheen-top),var(--sheen-bottom));background-blend-mode:normal;background-size:100% 100%;border:1px solid var(--border-strong);border-radius:var(--radius-md);min-height:var(--hit);box-shadow:var(--shade),var(--btn-bevel);} /* === Stacked tables on a phone === Below 560px every table in the app stops being a table and becomes a list of labelled blocks. The header row leaves the screen and each cell draws its own column name back above its value,out of the data-label the renderer puts on it - the library's cells in vendor/table.mjs,the hand-rolled ones at their own <td> sites. WHY. A table on a phone is a sideways scroll,and a sideways scroll on a table is how a column gets lost:at 390px the six-column cepending table measured 1053px inside a 356px box,so four of its six columns sat off the right edge behind a gesture with nothing on screen announcing it. Stacking puts every column on the screen at once. It costs height,and that is the trade this makes deliberately. The cells already carried their names in the <th>;what they did not carry was any way to say WHICH column they were once the heading row was gone,because a table says that with position and position is what stacking removes. THE HEADER ROW GOES OFF-SCREEN,NOT display:none. Off-screen keeps it in the accessibility tree,so a screen reader still announces the column names in the heading and then each value under its own name. display:none would take the names away from the one reader who cannot see the labels either. SORTING IS THE COST,and it is not paid here. Every .crm-table th holds a <button> that sorts,and with no heading row on screen there is nothing for a finger to tap - a keyboard and a screen reader can still reach it,a thumb cannot. Sorting the list on a phone therefore stops being possible. That is a real loss and it is stated rather than hidden;giving it back needs a control beside the search field,which is a design decision and not this rule. The box stays .crm-table's. Rows become blocks INSIDE it rather than cards of their own,so a stacked table is the same bordered rounded box with the same surface,and only what is inside it changed - one look everywhere,which is the rule the rest of this stylesheet follows too. */ @media (max-width:560px){/* 1. Un-table every family. :is() keeps it one rule;the names are the four the app renders tables under - .crm-table is most of them,.report-table is the field:value reports,.task-table is the project list and .task-add__tasks is the add-task screen's Active Tasks panel. */ :is(.crm-table,.report-table,.task-table,.task-add__tasks),:is(.crm-table,.report-table,.task-table,.task-add__tasks) :is(thead,tbody,tr,th,td){display:block;} /* Two classes on the table itself,so it outranks the group above and has to be said again - this is the Availability table,whose own rule is what holds `display:table` at every width. */ .settings__section .crm-table table{display:block;} :is(.crm-table,.report-table,.task-table,.task-add__tasks) thead{position:absolute;top:-9999px;left:-9999px;} /* 2b. A numeric cell goes left with everything else. Figures are right-aligned on the desktop so a column of them lines up under its heading,and that is exactly what the stack removes:the heading row is off screen,and the column name this block draws in ::before INHERITS the cell's alignment - so a right-aligned cell put its own name against the right edge too,with the left half of the block empty. Measured on the radar traffic table at 414px before this rule:"THIS MONTH / 1,234" hugging the right of a 380px block. Both classes are named rather than matched with [class*=num],because a name that only looks numeric is how a value ends up aligned by accident. */ :is(.crm-table,.report-table,.task-table,.task-add__tasks) :is(td,th):is(.radar__num,.hours__num){text-align:left;} /* 3. One row,one block. A rule between them and no border around them:the box is already .crm-table's,and a card inside a box is two surfaces saying the same thing. */ :is(.crm-table,.report-table,.task-table,.task-add__tasks) tbody tr{border-bottom:1px solid var(--border);padding:var(--space-2) 0;} :is(.crm-table,.report-table,.task-table,.task-add__tasks) tbody tr:last-child{border-bottom:0;} /* 4. One cell. The desktop rules that bound a cell to one line are lifted here,or every value is a single unbroken line and the block is as wide as the longest one - which is the sideways scroll this rule exists to remove. `.cell-link` keeps its own 30ch cap:an ecampaign URL is ninety-odd characters and a capped,ellipsised one still fits the narrowest phone,with the whole address in the title and in the DOM. */ :is(.crm-table,.report-table,.task-table,.task-add__tasks) td{border:0;width:100%;box-sizing:border-box;padding:var(--space-1) var(--space-4);} .report-view .crm-table td{white-space:normal;} .crm-table .stickies__row-action{white-space:normal;} /* And the link inside the cell,which holds its own nowrap and would sit there on one line whatever the cell does. `--link-max` is 30ch and a ch grows with the type scale,so raising the phone's body size to 16px grew this cap with it. Measured at 320px:the cap computed to 305px inside a 254px cell,the anchor's right edge landed 34px past the box's,and `.crm-table`'s `overflow:hidden` cut it - a URL with its tail missing and nothing on screen to say so. Capped to the cell instead of to a character count:one line,ellipsised the same way,and it cannot be wider than the space it is in however the type scale moves. */ .report-view .crm-table td a.cell-link{max-width:100%;} /* 5. The label. `attr(data-label)` is the whole mechanism:the value is in the cell,the name of the column it is under is on the cell,and this draws it. A cell with no data-label draws nothing,which is what the action cells want - they hold a button,not a value. */ :is(.crm-table,.report-table,.task-table,.task-add__tasks) td::before{content:attr(data-label);display:block;font-size:var(--text-xs);font-weight:600;color:var(--text-muted);text-transform:uppercase;letter-spacing:0.05em;} /* 6. The first cell is the row's identity,not a labelled field - it is the account,the client,the person,the project. It reads as the block's title and its own label is switched off,because a card headed "ACCOUNT / Heritage Baptist" says account twice. */ :is(.crm-table,.report-table,.task-table,.task-add__tasks) td:first-child{font-size:var(--text-base);font-weight:600;padding-top:var(--space-2);padding-bottom:var(--space-2);} :is(.crm-table,.report-table,.task-table,.task-add__tasks) td:first-child::before{display:none;}} /* === The phone input floor === LAST IN THE FILE,ON PURPOSE,and it is the same reason the paging block gives:this rule has to beat `.settings__hours{min-height:32px}` and every other field that carries its own height,all of which sit above it at one class. A media query adds no specificity,so the only thing that makes this win is being after them - the first attempt put it up with the rest of the touch rules and it silently did nothing,measured at 390px as a 32px settings field beside a 44px toolbar field. A field is where a near miss costs the whole interaction:the tap opens the keyboard and puts the caret somewhere else. `.input` is 32px,which is the right floor for a mouse and short of the touch one. */ @media (max-width:560px){.input{min-height:var(--hit-touch);} /* === The contacts search === `html "input" [class "input contacts-search"]` inside `.contacts-toolbar`,the wrapping flex row it shares with the sort select. HERE AND NOT WITH THE OTHER TOUCH RULES,because the rule it has to beat is `.contacts-search{flex:0 1 18rem}` at the settings block further up at one class - and a media query adds no specificity,so a copy placed earlier loses on source order whatever it says. Measured before this moved,at 390px:the field computed to `flex:0 1 288px` and stopped 44px short of the card's inner edge,while at 320px it happened to fill,because 18rem clamps to the container at that width. A rule that works at one width and not another is the shape this failure takes. `flex:1 1 100%` gives it the row it is already wrapping onto,so the field spans the toolbar and the sort select takes the line below. */ .contacts-search{flex:1 1 100%;width:auto;} /* The dashboard's search control carries `.table-search`,whose base rule ends with a `--space-3` right margin - a 12px gap between the field and the card's inner edge once the control has a line of its own,which is exactly the ragged edge this work was for. */ .card__head .table-search{margin-right:0;}} /* === The finish step === One collapsed row per task,because a run finishes its tasks one at a time and eight tasks' worth of note boxes will not fit a window that cannot be resized once open. Native <details>,the way the hours screen discloses a day. */ .finish-list{list-style:none;margin-block-start:var(--space-3);padding:0;} .finish-row{border:1px solid var(--border);border-radius:var(--radius-md);background-color:var(--surface);margin-block-end:var(--space-2);} .finish-row__head{display:flex;align-items:center;gap:var(--space-3);padding:var(--space-2) var(--space-3);cursor:pointer;font-size:var(--text-sm);} .finish-row__head:focus-visible{outline:2px solid var(--accent);outline-offset:2px;} .finish-row__id{font-variant-numeric:tabular-nums;color:var(--text-muted);} .finish-row__name{flex:1 1 auto;min-width:0;overflow:hidden;text-overflow:ellipsis;white-space:nowrap;} .finish-row__account{color:var(--text-muted);white-space:nowrap;} .finish-row__mark{color:var(--warn-text);white-space:nowrap;} .finish-row__body{padding:0 var(--space-3) var(--space-3);} /* The run's tasks on the finished card,one button each,because a browser opens one window per gesture and these are the gestures. Bare like the search's results,for the same reason:a list you press,not a row of actions. */ .timer__finish-row{display:flex;align-items:center;gap:var(--space-3);width:100%;justify-content:flex-start;text-align:start;} .timer__finish-id{font-variant-numeric:tabular-nums;color:var(--text-muted);} .timer__finish-name{flex:1 1 auto;min-width:0;overflow:hidden;text-overflow:ellipsis;white-space:nowrap;} .timer__note{color:var(--text-muted);margin-block:var(--space-2);} .timer__record{margin-block:var(--space-2);} .task-lookup{padding-block:var(--space-6) var(--space-2);} .task-lookup__note{color:var(--text-muted);margin-block:var(--space-2) var(--space-4);} .task-lookup__form{display:flex;align-items:flex-end;gap:var(--space-3);flex-wrap:wrap;} /* Back to the desktop query,which the rail's labels and the rest of the rail belong to. */ @media (min-width:768px){/* THE RAIL'S LABELS,PINNED OPEN. `data-rail-labels` is written on the document element at boot from localStorage,the same shape data-theme and data-density use,so this and the two blocks below are the whole of what the preference means. The list of screens above stops mattering here,and that is the one thing this preference simplifies. That list exists because the rail normally opens only under the pointer,so it can only cover a page that has nowhere to open into - and which pages those are had to be named one at a time. With the labels pinned the rail is that wide all the time,so EVERY screen has to make room for it and there is no list to keep. */ :root[data-rail-labels="always"] body{padding-left:var(--rail);} .tab-bar{top:0;bottom:0;right:auto;width:var(--rail-icons);border-top:none;border-right:1px solid var(--border-light);padding-bottom:0;/* A column,so the theme toggle can be pushed to the foot of the rail - five tabs at the top,the one control that is not a destination at the bottom. */ display:flex;flex-direction:column;/* A label is wider than the collapsed rail,so the row it sits on paints past it. This is what clips that,and it is also what keeps the label from spilling into the page while the rail is opening. Vertical is auto rather than hidden so that a window too short for the tabs AND the foot scrolls the rail instead of losing the toggle off its bottom edge;it is the short window that is the risk,not the tall one,and 5 tabs plus the toggle need about 330px. */ overflow-x:hidden;overflow-y:auto;transition:width 160ms ease;} .tab-bar:is(:hover,:focus-within){width:var(--rail);} .tab-bar ul{flex-direction:column;gap:var(--space-1);padding:var(--space-4) var(--space-3);} .tab-bar li{flex:none;} /* margin-top:auto is what pins it:everything else in the column is content-sized,so the free space collects above the foot. */ .rail-foot{display:block;margin-top:auto;padding:0 var(--space-3) var(--space-4);/* A rule between the destinations and the controls. The foot holds three things that are not places - the labels toggle,the theme toggle and the DeepSeek light - and with the marks now in ink (below) the column would otherwise read as eight of the same kind. */ border-top:1px solid var(--border-light);} /* At rest the chip IS the nav,so it does not get to shrink:a flex item shrinks toward its content by default,which would leave the chip's width up to how long the label beside it happens to be. */ .tab-bar .tab-icon{flex:none;} .tab-bar .tab-btn{flex-direction:row;justify-content:flex-start;gap:var(--space-3);padding:var(--space-2) var(--space-3);border-radius:var(--radius-md);font-size:var(--text-sm);min-height:44px;} /* The selected row is the whole pill here,so the icon does not also need its own chip behind it. The label takes --on-accent-subtle,the token for anything sitting on --accent-subtle:the plain accent on its own tint measures 4.24:1 light and 3.13:1 dark,and --accent-hover only reached 4.52:1 in dark - less legible than the unselected labels beside it,which take --nav-text and measure 5.70:1 light and 11.74:1 dark on the bar,so the selected tab read as the dimmest item in the nav. Now 5.49:1 and 6.38:1. */ /* The parenthetical that used to sit after "unselected labels beside it" read (6.93). That was --text-muted on --surface in dark;the unselected label has since moved to --nav-text,which is #d6d3d1 there,so the figure no longer described the thing it was attached to. Measured,not inferred:#d6d3d1 on the bar's --surface,#1c1917,is 11.74:1. */ .tab.active .tab-btn{background:var(--select-fill);color:var(--mark-ink);} .tab.active .tab-icon{background:none;} /* Shown by the rail,not by the label changing its own display:opacity keeps the label in the flow - so the open rail is the layout this nav always had - while the rail's own clip keeps it off the page. display could not fade,and both display and visibility would take the button's only text out of the accessibility tree with it. The name does not depend on any of that:it is on the button as an aria-label,so the five tabs are named whether the rail is open or shut. See tabBar.ml. The delay is on the way in only,so the label lands after the rail has opened past it and does not fade up into the closing edge. */ .tab-label{font-size:var(--text-sm);opacity:0;transition:opacity 90ms ease;} .tab-bar:is(:hover,:focus-within) .tab-label{opacity:1;transition-delay:70ms;} /* Pinned open. Both rules are needed:the width or the rail stays 78px while its labels paint past it,and the labels or five words sit invisible in a rail that is already wide enough for them. Written at (0,2,0) against the hover rule's (0,2,0),so source order decides and this is below it - and it drops the delay,because a reveal that waits 70ms for a transient is a flicker when the thing is simply always there. */ :root[data-rail-labels="always"] .tab-bar{width:var(--rail);} :root[data-rail-labels="always"] .tab-label{opacity:1;transition-delay:0ms;} /* THE FOOT KEEPS NO WORDS,and this is the rule that pays for the rail's width. "Switch to dark mode" needs 181px and "DeepSeek is up" 151,against the 115 the widest TAB needs - so two labels about the app's state were setting the size of a rail whose five destinations fit in 115. Hiding them is what lets --rail be 144. Nothing is lost to a screen reader or to a keyboard:both buttons carry their sentence as an aria-label,so the names are exactly what the visible words were,and the rail's five tabs have worked this way since they were built - their names are aria-labels and the span below is only the drawing of them. `display:none` and not opacity,unlike the tabs above. Those fade because the rail opens past them and a fade is what that reveal is;these are never shown at any width,so they are not in the layout at all - and a hidden span that is still laid out would go on setting the rail's content width the moment anyone measured it. */ .rail-foot .tab-label{display:none;} /* The FAB sat above the bottom bar;with no bottom bar it just sits in the corner. Two classes,not one,and it is specificity rather than tidiness that needs them:the base .fab rule is further down this file and sets the mobile offset at the same (0,1,0),so source order handed the desktop corner the 70px bar offset instead of 24px. Written at (0,2,0) this wins wherever the base rule sits. */ .fab.fab{bottom:var(--space-6);right:var(--space-6);} .login-page{min-height:100dvh;}} .form-error{color:var(--danger-text);} .form-message{color:var(--success-text);} .field-note{color:var(--text-muted);font-size:var(--text-sm);margin:0 0 var(--space-3);} .group-list{list-style:none;padding:0;margin:0 0 var(--space-4);columns:3;column-gap:var(--space-4);} .group-list li{margin:0 0 var(--space-1);break-inside:avoid;} .sent-filter{display:flex;align-items:center;gap:var(--space-2);margin-bottom:var(--space-4);} .stat__error{color:var(--danger-text);} /* A hand-built table has no sort behaviour,so its headings must not claim any;and a figure column aligns right on the heading and the cell alike. */ .static-table th{cursor:default;} .static-table th:hover{color:inherit;} .crm-table td.num,.crm-table th.num{text-align:right;} /* The legacy screens sat flush against the rail and the window edge. A page frame gives the content a margin and a readable measure;the rail keeps its own full-height column. */ .app-main{padding:var(--space-5) var(--space-6) var(--space-6);max-width:1200px;margin-inline:auto;} .app-main .stat-group{margin-bottom:var(--space-5);} .app-main .empty-state{text-align:center;padding:var(--space-6) 0;} @media (min-width:561px){#pending-table td:nth-child(1),#sent-table td:nth-child(2){white-space:nowrap;} #pending-table td:nth-child(3){min-width:10rem;} #pending-table td:nth-child(4) a,#sent-table td:nth-child(3) a{overflow-wrap:anywhere;}} @media (max-width:700px){.app-main{padding:var(--space-4);}} /* The rail's foot is 78px wide:its key is the icon,the words live in the same hidden label the tabs use. */ .rail-foot .btn.rail-signout{width:100%;justify-content:flex-start;padding:var(--space-2) var(--space-3);gap:var(--space-3);} /* The rail's reserved width is the whole column,not just its icons:with only the icons reserved the open rail drew over the first column of every page. */ @media (min-width:768px){body{padding-left:var(--rail);} .tab-bar{width:var(--rail);} .tab-bar .tab-label{opacity:1;transition-delay:0ms;}} /* Fixed column measures keep a long newsletter URL from squeezing the name and date columns into two-line stubs. The stacks below 560px are unaffected. */ @media (min-width:1101px){#pending-table,#sent-table,#leads-table{table-layout:fixed;width:100%;} #pending-table th[data-label='When']{width:11rem;} #pending-table th[data-label='Subject']{width:24%;} #pending-table th[data-label='Name']{width:11rem;} #pending-table th[data-label='Status']{width:7rem;} #pending-table th[data-label='Size']{width:6rem;} #pending-table tr th:last-child{width:8rem;} #sent-table th[data-label='name']{width:10rem;} #sent-table th[data-label='when']{width:11rem;} #sent-table th[data-label='size'],#sent-table th[data-label='bounces'],#sent-table th[data-label='opens'],#sent-table th[data-label='leads']{width:6rem;} #leads-table th[data-label='Email']{width:60%;}} /* A URL is a link,not prose:one line,clipped,with the whole address in the title. Breaking it anywhere split hosts and paths mid-word. */ @media (min-width:561px){#pending-table td:nth-child(4),#sent-table td:nth-child(3){overflow:hidden;} #pending-table td:nth-child(4) a,#sent-table td:nth-child(3) a{display:block;white-space:nowrap;overflow:hidden;text-overflow:ellipsis;}} @media (min-width:561px){#pending-table td[data-label='Subject'],#pending-table td[data-label='Name']{white-space:nowrap;overflow:hidden;text-overflow:ellipsis;}} /* Work's table behaviour,copied:the table scrolls sideways inside its own box between the phone stack and the wide layout,and only becomes a real table when there is room for it. */ .crm-table table{display:block;overflow-x:auto;} @media (min-width:1101px){.crm-table table{display:table;}} /* Between the stack and the wide layout the table is auto-laid and scrolls sideways in its own box,the way work's tables do:nowrap cells give the scrollWidth,and the URL is capped at 30ch like work's .cell-link. */ @media (min-width:561px) and (max-width:1100px){#pending-table td[data-label='When'],#sent-table td[data-label='when']{white-space:nowrap;} #pending-table td[data-label='Subject'],#pending-table td[data-label='Name']{white-space:nowrap;} #pending-table td[data-label='URL'] a,#sent-table td[data-label='URL'] a{display:inline-block;max-width:30ch;overflow:hidden;text-overflow:ellipsis;white-space:nowrap;vertical-align:bottom;}} /* A URL and a timestamp are single values,not prose:one line,clipped with the whole value in the title. The URL breaks the way work's .cell-link does (30ch cap,lifted to the cell's own width on a phone);the date takes the same treatment rather than splitting "11/3/2026," from "9:30:00 AM". */ .crm-table a.cell-link{display:inline-block;max-width:30ch;overflow:hidden;text-overflow:ellipsis;white-space:nowrap;vertical-align:bottom;} .crm-table td.cell-date{white-space:nowrap;overflow:hidden;text-overflow:ellipsis;} @media (max-width:560px){.crm-table a.cell-link{max-width:100%;} .crm-table td.cell-date{max-width:100%;}} /* Long values wrap inside their column;they are never clipped or ellipsised. The URL breaks at its own slashes where it can and anywhere it must,and a date breaks at its space rather than losing its time. */ .crm-table a.cell-link{display:inline;max-width:none;white-space:normal;overflow:visible;text-overflow:clip;overflow-wrap:anywhere;word-break:normal;} .crm-table td.cell-date{max-width:none;white-space:normal;overflow:visible;text-overflow:clip;overflow-wrap:break-word;} @media (min-width:561px){#pending-table td[data-label='Subject'],#pending-table td[data-label='Name'],#pending-table td[data-label='URL'],#pending-table td[data-label='When'],#sent-table td[data-label='URL'],#sent-table td[data-label='when'],#sent-table td[data-label='name']{white-space:normal;overflow:visible;text-overflow:clip;} #pending-table td[data-label='URL'] a,#sent-table td[data-label='URL'] a{display:inline;max-width:none;white-space:normal;overflow:visible;overflow-wrap:anywhere;}} /* Between 561 and 1100 the columns get percentage measures,so the URL has a real column to break inside instead of collapsing to a character a line. `break-word`,not `anywhere`:a URL still breaks when it must,but its min-content width is the whole address,so auto layout cannot squeeze the column to nothing. */ @media (min-width:561px) and (max-width:1100px){#pending-table,#sent-table,#leads-table{table-layout:fixed;width:100%;} #pending-table th[data-label='When']{width:14%;} #pending-table th[data-label='Subject']{width:20%;} #pending-table th[data-label='Name']{width:13%;} #pending-table th[data-label='URL']{width:25%;} #pending-table th[data-label='Status']{width:10%;} #pending-table th[data-label='Size']{width:8%;} #pending-table tr th:last-child{width:10%;} #sent-table th[data-label='name']{width:13%;} #sent-table th[data-label='when']{width:14%;} #sent-table th[data-label='URL']{width:25%;} #sent-table th[data-label='size'],#sent-table th[data-label='bounces'],#sent-table th[data-label='opens'],#sent-table th[data-label='leads']{width:8%;}} .crm-table a.cell-link{overflow-wrap:break-word;} /* A table whose columns are measured has to be a table:`display:block` (the sideways-scroll shape) makes column widths advisory,which is how the URL column collapsed to a few characters. Our hand-built tables fit their box by wrapping,so they stay real tables and the wrapper keeps the scroll option. */ @media (min-width:561px){.crm-table.static-table{overflow-x:auto;} .crm-table.static-table table{display:table;}} /* Mid-size screens get column measures in rem and a table wider than the page,scrolled by its wrapper - the same trade work's wide tables make. The earlier percentage measures were narrow enough to break dates and statuses mid-word,which is not "breaking down",it is mangling. */ @media (min-width:561px) and (max-width:1100px){#pending-table,#sent-table,#leads-table{width:75rem;} #pending-table th[data-label='When']{width:10.5rem;} #pending-table th[data-label='Subject']{width:14rem;} #pending-table th[data-label='Name']{width:9rem;} #pending-table th[data-label='URL']{width:24rem;} #pending-table th[data-label='Status']{width:6rem;} #pending-table th[data-label='Size']{width:5rem;} #pending-table tr th:last-child{width:6.5rem;} #sent-table th[data-label='name']{width:9rem;} #sent-table th[data-label='when']{width:10.5rem;} #sent-table th[data-label='URL']{width:24rem;} #sent-table th[data-label='size'],#sent-table th[data-label='bounces'],#sent-table th[data-label='opens'],#sent-table th[data-label='leads']{width:5rem;}} .group-list{list-style:none;padding:0;margin:0;display:grid;grid-template-columns:repeat(auto-fill,minmax(15rem,1fr));gap:var(--space-1) var(--space-4);} .group-previous{margin-top:var(--space-4);} .group-previous summary{cursor:pointer;color:var(--text-muted);font-size:var(--text-sm);margin-bottom:var(--space-2);} .edit-form .field-note{margin:var(--space-2) 0 0;} .edit-form input[type='file'].input{padding:var(--space-2);} .edit-form__actions{display:flex;justify-content:flex-end;gap:var(--space-3);padding-top:var(--space-1);} .edit-form__row{display:grid;grid-template-columns:1fr 1fr;gap:var(--space-4);} @media (max-width:700px){.edit-form__row{grid-template-columns:1fr;}} .blast-layout{display:grid;grid-template-columns:minmax(0,1fr) 24rem;gap:var(--space-5);align-items:start;} .blast-side{position:sticky;top:var(--space-4);background-color:var(--surface);background-image:var(--panel-rim);border:1px solid var(--border);border-radius:var(--radius-xl);box-shadow:var(--panel-lift),var(--panel-bevel);padding:var(--space-4);} .blast-side__title{font-size:var(--text-xs);font-weight:600;color:var(--text-muted);text-transform:uppercase;letter-spacing:0.05em;margin:0 0 var(--space-3);} .side-list{list-style:none;margin:0;padding:0;} .side-blast{display:grid;grid-template-columns:minmax(0,1fr) auto;gap:var(--space-1) var(--space-3);padding:var(--space-2) 0;border-bottom:1px solid var(--border-light);} .side-blast:last-child{border-bottom:0;} .side-blast__when{grid-column:1 / -1;font-size:var(--text-xs);color:var(--text-muted);} .side-blast__name{font-size:var(--text-sm);color:var(--text);text-decoration:none;} .side-blast__name:hover,.side-blast__name:focus-visible{text-decoration:underline;} .side-blast__size{font-size:var(--text-sm);color:var(--text-muted);font-variant-numeric:tabular-nums;} .side-note{color:var(--text-muted);font-size:var(--text-sm);margin:0;} @media (max-width:1100px){.blast-layout{grid-template-columns:1fr;} .blast-side{position:static;}} .side-blast{display:flex;justify-content:space-between;align-items:baseline;gap:var(--space-3);} .side-blast__client{font-size:var(--text-sm);color:var(--text-muted);text-align:right;} /* The schedule is the thing being scanned in the sidebar:it takes the text size and weight,the client takes the quiet line beneath it. */ .side-blast{flex-direction:column;align-items:flex-start;gap:0;} .side-blast__when{font-size:var(--text-base);font-weight:600;color:var(--text);} .side-blast__client{font-size:var(--text-sm);color:var(--text-muted);text-align:left;} .group-list li.is-hidden{display:none;} .recipients-tools{display:grid;grid-template-columns:minmax(0,1fr) 16rem;gap:var(--space-4);align-items:start;} @media (max-width:700px){.recipients-tools{grid-template-columns:1fr;}} .rail-heading{list-style:none;font-size:var(--text-2xs);font-weight:600;letter-spacing:0.05em;text-transform:uppercase;color:var(--text-muted);padding:var(--space-3) var(--space-3) var(--space-1);} .edit-form__header{display:flex;align-items:center;justify-content:space-between;gap:var(--space-4);flex-wrap:wrap;} .edit-form__header .page__title{margin-bottom:0;} .edit-form__card-head{display:flex;align-items:center;justify-content:space-between;gap:var(--space-3);margin-bottom:var(--space-3);} .edit-form__card-head .edit-form__card-title{margin-bottom:0;} .sub-layout{display:grid;grid-template-columns:minmax(0,1fr) 14rem;gap:var(--space-5);align-items:start;} .sub-side{position:sticky;top:var(--space-4);} .sub-side .btn{width:100%;} @media (max-width:1100px){.sub-layout{grid-template-columns:1fr;} .sub-side{position:static;}} .import-source-row{display:flex;align-items:center;gap:var(--space-3);margin-bottom:var(--space-3);} .import-source-row .edit-field__label{margin:0;} .import-source{border:0;padding:0;margin:0 0 var(--space-3);display:flex;gap:var(--space-4);flex-wrap:wrap;align-items:center;} .import-source legend{padding:0;margin-right:var(--space-2);} .groups-toolbar{display:flex;align-items:center;justify-content:space-between;gap:var(--space-3);flex-wrap:wrap;margin:0 0 var(--space-3);} .image-grid{display:grid;gap:var(--space-3);margin:0 0 var(--space-4);} .image-row{display:flex;align-items:center;gap:var(--space-3);flex-wrap:wrap;padding:var(--space-3);border:1px solid var(--border-light);border-radius:var(--radius-md);background:var(--surface);} .image-row .input{flex:1 1 12rem;min-width:0;} .image-row__preview{max-width:12rem;max-height:6rem;object-fit:contain;border:1px solid var(--border-light);border-radius:var(--radius-sm);background:var(--surface-hover);} .image-row__size{font-size:var(--text-xs);color:var(--text-muted);font-variant-numeric:tabular-nums;} .image-row__size:empty{display:none;} .stat-grid{display:grid;grid-template-columns:repeat(auto-fit,minmax(8rem,1fr));gap:var(--space-3);margin:var(--space-4) 0 0;} .image-row__preview:not([src]){display:none;} #groups-table td[data-label="Active"],#groups-table td[data-label="Bounced"],#groups-table td[data-label="Unsub"]{white-space:nowrap;} .rail-client{font-size:var(--text-xs);font-weight:600;text-transform:uppercase;letter-spacing:0.05em;color:var(--text-muted);padding:var(--space-2) var(--space-4) var(--space-1);} /* Send-as account search:the Add Task account filter's list shape,scoped to the staff blast form. The list scrolls;a match is a full-width button. */ .staff-picker__matches{list-style:none;margin:var(--space-1) 0 0;padding:0;display:flex;flex-direction:column;border:1px solid var(--border-light);border-radius:var(--radius-sm);overflow-y:auto;overflow-x:hidden;max-height:288px;} .btn.staff-picker__match{width:100%;justify-content:flex-start;border-radius:0;text-align:left;padding:var(--space-2) var(--space-3);background:none;box-shadow:none;border:0;border-bottom:1px solid var(--border-light);color:var(--text);white-space:normal;overflow-wrap:anywhere;} .staff-picker__matches li:last-child .staff-picker__match{border-bottom:none;} /* A result has to say more than a company name:nine client records share one name and seven of those also share one email. The email line splits the first group;the account number splits the second,and is drawn only for exact duplicates so ordinary rows stay quiet. */ .btn.staff-picker__match{flex-direction:column;align-items:flex-start;gap:2px;} .staff-picker__name{font-weight:600;color:var(--text);} .staff-picker__email,.staff-picker__id{font-size:var(--text-xs);color:var(--text-muted);} /* The kind's hover/focus fill repaints the button and sets the button colour,but these spans carry their own colours,so the labels stayed dark on the blue fill (measured 2.83:1 and 1.46:1). Restate them for the states that move the fill. */ .btn.staff-picker__match:is(:hover,:focus-visible,:active) .staff-picker__name,.btn.staff-picker__match:is(:hover,:focus-visible,:active) .staff-picker__email,.btn.staff-picker__match:is(:hover,:focus-visible,:active) .staff-picker__id{color:var(--on-kind);} /* The count is a live region that exists before it has text;an empty one should not leave a paragraph gap under the input. */ .staff-picker__count:empty{margin:0;} /* The rail carries eleven destinations and three section headings. On a phone they were laid out as one flex row with every item flex:1,so the headings painted as giant words,the labels wrapped,and the last destinations fell off-screen. The headings are a desktop rail concept;the phone bar is one horizontally scrollable row of tabs,each at a thumb-sized minimum. */ @media (max-width:767px){:root{--tab-height:58px;} /* Body padding reserves room at the document end only. Focusing a field near the fold still lets the browser scroll it just inside the viewport bottom,behind the fixed bar. This moves the scroll box itself. Phones only:the desktop rail sits beside the content and overlays nothing. */ html:has(.tab-bar){scroll-padding-bottom:calc(var(--tab-height) + env(safe-area-inset-bottom,0px));} .tab-bar{overflow-x:auto;overflow-y:hidden;overscroll-behavior-x:contain;-webkit-overflow-scrolling:touch;scrollbar-width:none;} .tab-bar::-webkit-scrollbar{display:none;} .tab-bar ul{flex-wrap:nowrap;padding:4px var(--space-2);gap:2px;} .tab-bar li{flex:0 0 auto;} .tab-bar .rail-heading{display:none;} .tab-bar .rail-client{display:none;} .tab-bar .tab-btn{min-width:64px;padding:7px 8px 5px;} .tab-bar .tab-icon{width:26px;height:26px;} .tab-bar .tab-label{max-width:72px;overflow:hidden;text-overflow:ellipsis;white-space:nowrap;font-size:11px;line-height:1.15;} /* Touch targets:row actions in tables were the smallest controls on the page. 44px is the floor this stylesheet already uses for tabs. */ .crm-table .btn,.crm-table .btn--small{min-height:var(--hit-touch);min-width:var(--hit-touch);padding:var(--space-2) var(--space-3);} /* The widget docks to the viewport bottom by default,which puts its toggle and its open panel over the phone nav. Lift both by the bar's height. */ body #chat-client .status-bar{bottom:calc(var(--tab-height) + env(safe-area-inset-bottom,0px));right:var(--space-3);} body #chat-client .chat-form{bottom:calc(var(--tab-height) + env(safe-area-inset-bottom,0px) + var(--space-2));right:var(--space-3);width:min(380px,calc(100vw - 2 * var(--space-3)));max-height:calc(100dvh - var(--tab-height) - env(safe-area-inset-bottom,0px) - var(--space-4));}} /* A dashboard row is five cells with the same two lines:value,then label. Subgrid makes the label line shared,so a wrapped label no longer lifts one number above its neighbours. Phone layout is the existing flex row and is deliberately excluded. */ @media (min-width:561px){.stat-group:not(.stat-group--error){grid-template-rows:auto auto;} .stat-group:not(.stat-group--error) .stat{display:grid;grid-row:span 2;grid-template-rows:subgrid;align-content:start;row-gap:0;}} /* A form row's fields are a label over a control. Subgrid ties every control in the row to the same second track,so a wrapping label cannot shorten one input and leave its neighbour taller. Browsers without subgrid ignore the template and fall back to the field's own stacking. */ .edit-form__row{grid-template-rows:auto auto;row-gap:var(--space-1);margin-bottom:var(--space-3);} .edit-form__row>.edit-field{display:grid;grid-row:span 2;grid-template-rows:subgrid;margin-bottom:0;} .edit-form__row>.edit-field>.edit-field__label{margin-bottom:0;} @media (max-width:700px){.edit-form__row{grid-template-rows:none;row-gap:var(--space-4);} .edit-form__row>.edit-field{display:block;grid-row:auto;} .edit-form__row>.edit-field>.edit-field__label{margin-bottom:var(--space-1);}} /* Below 560 the one-row scroller becomes a wrapped grid:every destination is visible instead of hidden past the right edge,and each row keeps the same icon-over-label tab. --tab-height is updated at runtime from the bar's real height so the page reserve and the chat dock follow the number of rows. */ @media (max-width:560px){/* Ten tabs cannot fit one row below 560 (they need 594px),so the bar is always at least two here. 58 is the one-row number and would leave the page reserve short until the first measurement lands. 118 is two rows;the runtime sync refines it and takes over as an inline value. */ :root{--tab-height:118px;} .tab-bar{overflow:visible;} .tab-bar ul{display:grid;grid-template-columns:repeat(auto-fit,minmax(56px,1fr));padding:4px var(--space-2);gap:2px;} .tab-bar li{flex:none;min-width:0;} .tab-bar .tab-btn{min-width:0;padding:6px 4px 4px;} .tab-bar .tab-label{max-width:none;}} /* One line above the screen content,only while the staff account is acting for a client. The fill is the same surface the cards use so it reads as part of the page,and the text keeps the body colour for contrast. */ .staff-bar{display:flex;align-items:center;justify-content:space-between;gap:var(--space-3);max-width:var(--page-max);margin:var(--space-4) auto 0;padding:var(--space-2) var(--space-3);background:var(--surface);border:1px solid var(--border-light);border-radius:var(--radius-sm);color:var(--text);font-size:var(--text-sm);} .staff-bar__text{font-weight:600;} .search-group{margin:0 0 var(--space-5);} .search-group__title{font-size:var(--text-base);font-weight:600;margin:0 0 var(--space-2);} .search-hit__label{display:block;font-weight:600;color:var(--text);} .search-hit__sub{display:block;font-size:var(--text-xs);color:var(--text-muted);} .link-check{margin-top:var(--space-2);} .link-check__list{list-style:none;margin:var(--space-2) 0;padding:0;font-size:var(--text-xs);} .link-check__row{padding:2px 0;color:var(--text-muted);overflow-wrap:anywhere;} .link-check__row--bad{color:var(--danger-text);} 