- replace default Onyx surfaces with paper/ink, signal orange and Archivo - make /plan answer peak, staff-hours, cost vs flat and coverage above the fold - add the weekly shift board, shared plan helpers and a pulse line - rebuild /live as a phone-first dark board and /stage as a full-bleed projector - label the real 14-20 Nov window; remove 'next week' claims - keep Onyx components, SSR, live scenario updates and accessible states
4.3 KiB
Product
Platform
web
Users
The primary user is the duty manager of an inner-city Heilbronn retailer who has to decide the next staff schedule. The secondary audience is a hackathon judging room: a presenter on a projector, judges and audience on their own phones, and a presenter operating a control page.
Product Purpose
PulseCast turns inner-city footfall into a transparent staffing plan for the coming store week. It exists so a manager can answer one question — who works when, and what does that week cost? — with the forecast, assumptions, uncertainty and trade-offs visible rather than hidden.
Positioning
Measured footfall becomes a modelled forecast, the forecast becomes a shift board, and the same decision is broadcast to the room. The product shows the cost against a flat schedule and names the overstaffed hours it buys. A neighbouring dashboard could copy the chart; it cannot copy the honest chain from sensor to shift to phone.
Operating Context
- Heilbronn inner-city store, Europe/Berlin, store hours 10:00–20:00.
- Eight fixed sensors in a roughly 580 m pedestrian cluster; 15-minute cadence.
- Shift blocks of 2–5 hours; default 4 hours; 25 visitors per staff-hour; ×1.15 safety factor; €16/hour labour assumption.
- No POS, sales, queue, payroll or labour-cost feed. Cost is an assumption, clearly labelled.
- The plan heuristic is decision support, not an optimizer. Coverage is at or above requirement by construction because each block is sized to its peak; overstaffed hours are the visible trade-off.
- The demo uses the live Go API.
/controlbroadcasts a scenario over SSE;/stageand/livefollow it with a 5-second polling fallback. - The live plan window currently returned by the API is 14–20 Nov 2026, not next Monday's week. The UI must label the real window and never say "next week" while the data says otherwise.
Capabilities and Constraints
- Nuxt 4 SSR with Onyx as the component system. Routes:
/,/forecast,/plan,/simulate,/about,/stage,/live,/control. /planreads Go/api/staffingthrough Nitro. Local assumption edits are an explicitly labelled preview; they do not change the API plan./stageand/livefollow the live scenario and must show connection state honestly (Live, Polling, Offline) and a last-updated time.- No new backend behavior, no new dependencies, no fabricated numbers. Read plan values from the API.
- Preserve SSR, keyboard access, contrast, reduced motion, responsive behaviour and the existing build and tests.
- The plan window on 14 Nov 2026 is 14–20 Nov. The paper brief's 5 staff / 210 staff-hours / €3,360 is stale and not reproducible; the reproducible sunny baseline is 5 peak, 172 staff-hours, €2,752 vs €5,600 flat.
Brand Commitments
- Name: PulseCast.
- Voice: direct, quiet confidence; honest about what is measured, modelled and assumed. "Staff the pulse, not the noise."
- The visual world is pinned as a shift board: paper/ink ground, signal orange for the peak, heavy timetable rules, tabular numerals and physical time blocks.
/stageinverts to a dark board. - Onyx remains the component system; default Onyx visuals must not ship.
Evidence on Hand
- Live Go API at
https://api-pulse.noelsabia.devand checked-in mock JSON. - Baseline screenshots and audit in
/tmp/opencode/audit(session artefact). - Tests in
test/including a captured Go staffing fixture. - No customer testimonials, no POS data, no city-wide heatmap claim. Do not invent them.
Product Principles
- Answer first. The peak shift, staff count, cost and coverage are the first reading order on every decision surface.
- Show the chain. Measured, modelled and assumed are visibly different things.
- One plan, every surface. The stage, the phone and the plan are the same numbers, propagated live.
- Keep the trade-off visible. Overstaffed hours and uncertainty are part of the answer, not a footnote.
- Earn the projector and the phone. Ten-metre legibility and one-thumb clarity are design constraints, not afterthoughts.
Accessibility & Inclusion
WCAG 2.2 AA. Keyboard access for all controls and scrollable data; visible focus; 4.5:1 body text and 3:1 large text; no colour-only meaning; reduced-motion alternatives; touch targets at least 44×44 px where practical; stage text legible from 10 m.