Design system
Hypcheck exists to stress-test a claim against evidence, on a schedule, without asking anyone their opinion. Every choice below tries to dramatize that premise: a hypothesis is a written claim, a PMF score is a verdict stamped onto it, and a revision is a copy-edited correction — not a chat bubble. This page is the system that follows from that idea, kept as a living page rather than a static deck so it can’t drift from what actually ships.
01 — Typography
Three faces doing three different jobs, never interchangeably: Newsreader sets the hypothesis itself — the thing under test — in italics, like a quoted claim. Geist Sans runs the console: labels, nav, buttons. Geist Mono is reserved for anything that's a measurement — scores, dates, token counts — so a number always looks like data, not prose.
02 — Verdict bands
Forest, slate teal, brass, and brick — hand-tuned instead of Tailwind's default emerald/teal/amber/rose, and deliberately not reused anywhere else. This is the one scale that answers exactly one question: is the evidence holding up?
strong
≥ 70
promising
55–69
emerging
40–54
weak
< 40
03 — Analyst’s call
A five-step diverging dial, a completely different hue family from the bands above — because 'how strong is the signal' and 'what should you do about it' are different axes and answering one shouldn't visually imply the other.
04 — The verdict stamp
The hero score treatment — a caliper-ring mark rather than a pill or a gauge, because the product's whole idea is a periodic reading, not a one-time grade. Same component at every size, from a table row to a report header.
05 — Compact reads
Where a stamp is too heavy — table rows, inline mentions — the same band and stance colors resolve down to a pill. Every one of these reads from the exact same token source as the stamp above; nothing is a separate palette.
06 — Editorial redline
Blue pencil for insertions, red pencil for strikethroughs — the mark-up idiom an editor uses on a manuscript, not a code diff. Deliberately separate tokens from the verdict-band reds and greens: correcting a claim's wording isn't a verdict on it.
Proposed hypothesis revision
B2BEnterprise salesAEs teamsrunning complex, multi-stakeholder deals will pay for an AI copilot that assembles meeting-prep briefs.briefs from CRM, email, and call history.
07 — Mobile
Not a hamburger drawer — three destinations fit a thumb-reachable bottom bar without overflow, and nothing about the app's controls is ever a tap away behind an icon. The per-hypcheck switcher stays sidebar-only; the dashboard grid already does that job one tap from here.