Design and product review
Reviewed against PLAN.md and against each other. Tools used: check.mjs (layout) and consistency.mjs, copy-lint.mjs, headless Chrome screenshots at 1440 and 390, tsc, eslint and the production build. Most of the kit's automatic FAILs are about its own CSS and motion layer (button-styles, motion-layer, heading-scale on a bespoke system), so I triaged them rather than chasing them all.
Problems in priority order. Status shows what was done.
- Dead links. The local preview links pointed at dev servers that no longer ran, and the landing "operator dashboard" link was hard-coded to a dev port. Anyone opening the deliverables saw a connection error. FIXED: dashboard builds with a relative base and is served as static files next to the other two; one server, one port.
- Dashboard unusable on phones. The upcoming-bookings table clipped Crew and Status off the right edge, so the main operator action (fixing a booking with no barista) was unreachable at 390px. Stat cards stacked into four full-height blocks and the chart labels rendered at about 6px. FIXED: stacked booking cards below
md, two-up stat grid, chart label size, min-width fix so long names truncate instead of pushing the card wider.
- Dashboard flagged problems but offered no way to fix them. "Need a barista: 2, assign before the day" had nothing to click, and "New booking" did nothing. FIXED: Assign button on every short-crewed booking (table and mobile) opens a list of baristas who are actually free (not on leave, no overlapping booking); assigning updates the status, schedule, roster hours and stats. Dead "New booking" button removed rather than left lying.
- Data contradicted the pricing model and the brand copy. Dashboard amounts did not follow the landing quote formula (a 4-hour, 40-guest cart was $560, the formula says $660), and a 140 to 180 guest bar was shown as fully crewed with two baristas while the landing promises a third above 120 guests. FIXED: amounts recomputed from the same formula;
crewNeeded now adds the third barista above 120 guests; landing copy says "two or three baristas".
- Focus ring failed contrast. The keyboard focus ring was steam gold (#e3b25a) on paper, about 1.8:1. Fails WCAG 1.4.11 and was flagged on every link. FIXED: 3px ink ring, steam on dark bands.
- Invented proof and claims on the landing page. "4.9 rating" on a barista and "Serving Sydney and Melbourne" were made up, and the footer read "prices shown are examples", which is a placeholder note on a sales page. FIXED: replaced with factual lines ("Free all morning", "Launch pricing in AUD. The quote you see is the price you pay."). Pricing itself is still a placeholder and is listed below.
- Calls to action went nowhere useful. "Book this setup" jumped to the next section; the office card had no action at all. FIXED: booking buttons point at the app (
app.doppio.example, a reserved placeholder domain), the quote button carries the chosen setup, guests and hours in its URL, and the office card has "Set up a standing booking".
- Section rhythm. Sections were forced to full-screen height, leaving the "manage", "offices" and closing sections 50 to 60 percent empty at 1440, and the right-hand columns started at three different x positions. FIXED: content-driven section height with fluid padding, and one shared 1.1/0.9 split so card edges line up down the page.
- Hero preview semantics. The four dots used
role=tab with no tabpanels or arrow keys, and the stage was aria-live, so a screen reader would read out a new panel every 3 seconds. FIXED: plain buttons with aria-current="step", live region removed (it is decorative), and the reduced-motion preference now reacts if it changes while the page is open.
- Style guide did not deliver what PLAN.md promised. No imagery or iconography guidance, no accessibility rules, no mapping of the tokens into the dashboard's shadcn variables, a type table whose "Aa" samples were all the same size, and a wrong contrast figure (white on copper is 5.3:1, not 5.6:1). FIXED: added Imagery and icons, Accessibility and a shadcn mapping table, corrected the samples and the figure.
- Touch targets. Chips and small buttons were 32 to 36px tall. FIXED: 44px on touch and narrow screens (landing and guide nav).
- Disabled setup chips were unexplained. Bar and Tasting went grey with no reason. FIXED: a visible hint under the chips gives the guest ranges.
Not fixed, and why
- Pricing, ratings and names are placeholders. The numbers are consistent across landing and dashboard, but the business has not set them. The mock names (Ines, Teo, Hana) still trip copy-lint's "default cast" warning; they need replacing with the client's real people.
- No photography. The landing page leads with product UI and line drawings. Real photographs are a client supply item.
- The booking flow itself is not built. The landing CTAs target a placeholder app URL.
- Mobile landing navigation shows only the brand and the main button; the section links are hidden under 900px. Acceptable for a one-page site, worth a menu if it grows.
- Three bespoke button styles (solid, ghost, quiet) plus chips. The checker's "button-styles" FAIL is about this; it is a deliberate system and documented in the guide.
- Dashboard state is in memory. Assignments reset on reload; there is no backend.