Freightpost — Ant Design

Cross-border freight: 140 open consignments, their customs status, and the ones that have stopped.

Direction

Lever	Decision
Archetype	Shell. Ant `Layout.Sider` at 208px, collapsing to a 64px icon rail under 992px; one scroll region for the table.
Data stance	Table-first at maximum density. Fourteen columns, the checkbox and the reference frozen, the horizontal scroll held inside the table. Filters live in the column headers.
Density	Ant `small`: 8px cell padding on 14px text, 43px rows measured, 24px header controls, 32px page controls.
The bold thing	An exception is a sentence, not a colour. The row keeps its own background and gains a second line naming what stopped it. That sentence is sticky: past the frozen columns on a desktop, and at the left edge of the scroll box below 992px, where it wraps to as many lines as it needs.

Generator

ant_palette.py "#0B5FFF"                 # tokens.css (committed, never edited)
ant_palette.py "#0B5FFF" --contrast      # the audit, below
ant_palette.py "#0B5FFF" --dark          # the ramp under the dark algorithm
ant_palette.py --map                     # neutrals, status, dimensions, motion
ant_palette.py --map --dark
ant_palette.py --presets                 # the gray ramp, which is special-cased

tokens.css holds the seeded ramp alone. tokens.support.css holds everything else and is generated the same way: a script reads the four commands above and emits it, so nobody typed a value into either file.

The contrast audit, and what it changed

step  hex         on white  on black   use
5     #3381ff        3.67:1     5.72:1   hover state of the primary
6     #0b5fff        5.13:1     4.10:1   THE primary — colorPrimary
7     #0045d9        7.36:1     2.85:1   active / pressed
Step 8 (text) on white: 9.88:1 passes 4.5:1
White on step 6 (primary button): 5.13:1 passes 4.5:1

White on step 6 passes at 5.13:1, where Ant's own #1677ff fails at 4.10:1. The primary fill carries text unchanged and needs no darker step.
The hover does need one. Ant lightens a primary button to step 5 on hover, which drops white text to 3.67:1. Hover is step 7 (7.36:1) and pressed is step 8 (9.88:1) instead, so the label holds AA through the whole press.
Ant's Tag pairs step 7 on step 1 and three of the five hues used here fail: green 3.37:1, cyan 3.39:1, gold 4.25:1. Every tag reads step 9 on step 1, which measures 5.44:1 (green), 5.83:1 (cyan), 6.53:1 (gold), 7.04:1 (red) and 10.94:1 (geekblue).
colorBorder is 1.41:1 on white and bounds every control. The generator's gray ramp replaces it, picked by rule rather than by eye: the first step clearing 3:1 against that theme's container — gray 3 (#8c8c8c, 3.36:1) in light, gray 7 (#6a6a6a, 3.41:1) in dark.

Target size (SC 2.5.8). Row checkboxes are 16px, which the spacing exemption carries: 43px between vertical neighbours and 40px to the reference link. The empty saved view carries no select-all, since a table with no rows has nothing to choose — and that also removes the only pair of 16px targets whose centres fell inside 24px of each other.

The empty saved view carries no filter and no sort either, on the same argument, and it had to: those four controls shared ids and handlers with the open table above, so filtering from the empty header re-sorted and re-filtered a table 660px off the top of the screen while opening its popover at top: -193px, where nobody could see it.

Measured at 390px: the sentence wraps to six lines and 100% of it is visible, sticky at the left edge of the scroll box. Before the critique round it was one 1264px line inside a 308px box — 9% of it showed.

Deviations from CONTRACT.md

A second generated stylesheet. tokens.css carries only the ten brand steps, and a page needs neutrals, status hues, spacing, type, radius and motion. Those come from ant_palette.py --map, --map --dark and --presets through a script into tokens.support.css, rather than being written into the page as raw values.
Ant ships no typeface, so the page loads no web font and uses the fontFamily and fontFamilyCode seeds verbatim.