Design review
What changed across four passes: the finish, the product, then where it lives and its colour.
v5: back to the idea
Researching the market showed that anonymous polls, meeting recaps and decision logs are all built in or owned already. It also showed that nobody lets everyone answer out loud at once and turns that into the meeting. v5 returns to that idea, which was Hivemind’s from the start.
| Area | v4 | v5, and why |
|---|---|---|
| The question | Two-sided debate questions (ship or hold), plus a crux and a guess at the room. | One open question, answered out loud. Speaking is about three times faster than typing, so ten people give ten real answers in a minute. |
| The readout | A stance map and hidden dissent: where the room stood on two sides. | The room’s own thinking: themes with an anonymous quote each, where it pulls apart, ideas only one person raised, unanswered questions. |
| What it does next | A decision and a revisit date. | It becomes the agenda for the rest of the meeting, and Hivemind keeps listening to bring back what the room said about each item. |
| In the call | An add-on on the main stage. | An add-on for the host, plus Hivemind as a participant: it speaks the summary and posts the readout and agenda in the chat. |
| Who it is for | Product teams. | Any team meeting: planning, retros, crits, all-hands. |
| Dashboard | A decision log, which Cloverpop and Teams Facilitator already own. | Team insights over time: who gets heard, what keeps coming up, ideas that went somewhere. |
What the room said
Onboarding is where people get stuck6 of 10
Half our trials never finish setup.
Fix what’s broken before building more4 of 10
We keep shipping features on top of bugs.
We don’t talk to customers enough3 of 10
None of us has been on a customer call this month.
v4: inside the call
v3 decided what Hivemind sells. v4 decided where it lives and how it feels to be in the room: a pause inside the call people are already in, a mute everyone can see, and a result they hear together.
| Area | v3 | v4, and why |
|---|---|---|
| Where it lives | A meeting simulator that looked like its own meeting, with every surface on screen at once. | An add-on inside the call the team is already in (Meet, Zoom). One pause, start to finish: the brief lands, the meeting carries on. |
| The pause | The round started; nothing in the call changed. | Start pause mutes the room with the host’s own control and a banner says so, in words, for the whole pause. The call’s tiles and mic show it too. |
| The result | Read off the shared screen, at whatever speed each person reads. | Read aloud by Hivemind with captions, in step with the stage, then the room is unmuted and the brief lands on every laptop. |
| The question | Ship export to beta, or hold until large files are reliable. | Ship the new onboarding this quarter, or hold it. Any reader gets it without product context. |
| Colour | One purple (#7c3aff) that read pale on graphite; status colours as loose hex; a dull brown warning. | Five OKLCH ramps on one lightness ladder. Every semantic colour aliases a ramp step per theme. Pulse is electric ultramarine-500 on paper and 400 on graphite; warning is a bright amber mark with dark amber text. |
| Meet flow | A generic call window that started the pause itself and muted the room. | A faithful Google Meet in Chrome, and only what an add-on can really do: Activities, a side panel, the main stage, an invitation to join. Microphones and chat stay with the host, guided step by step. |
| Who sees the result | The whole room, the moment the round closed. | The host first, in the side panel. The room sees “The round is in” until the host shares it, so a sensitive result is never broadcast by surprise. |
| Dark neutrals | The light ramp’s violet-grey, darkened. | Ink: a cool blue-black ramp of its own, with wider steps between ground, surface and raised. Charts carry their hero series in Pulse with a soft glow. |
| Type | Geist Mono for the timer, counts, the room code and code. | Geist only. Numbers that change take tabular figures, so they never jitter without a monospace face. |
| Record button | A circle with a soft halo: a generic record control. | The comb from the mark. The centre cell is the button; while it listens, a wave of Pulse runs round the six voice cells. |
| Dashboard | Light only. | Light and dark, switched in the sidebar and remembered per viewer. Every colour was already a token, so dark came from the theme alone. |
| Case study | Every surface in the first exhibit, and an eight-tab prototype. | Four plain steps first, then one artifact per section in the order a person meets it. The prototype has four tabs. |
BEFORE · v3
AFTER · v4
v3: the product
v2 fixed the finish. v3 changed what the product is for. An anonymous poll is a feature other tools already have; a room’s honest position on a decision, and how far that is from what everyone assumed, is not. Every change below follows from that.
| Area | v2 | v3, and why |
|---|---|---|
| Value | Anonymous answers, shown back as themes. Polling tools already do that. | The decision brief: where the room really stands, the gap from where it looked, what the split hinges on and what would bring people together. |
| The round | One open question. | One debate question with two named sides, asked as three prompts: position, what would change your mind, and what most of the room will say. |
| Readout | Up to three theme bars, one split, one lone signal. | A stance map, hidden dissent, the crux, conditions and the lone signal. On stage it plays in two beats. |
| Phone | Answer and send. | Three steps with a progress line. The crux is typed or spoken; the prediction is a single tap. |
| Console | Charts of activity: hives run, agreement, themes. | A decision log with each call’s hidden dissent and revisit, and a team profile over time. Up to 1400px wide. |
| Brand | Six dots and a halo; a monoline wordmark; “Hive Mind”. | A seven-cell comb; Geist Semibold outlined to a path; Hivemind, one word with a capital H. |
| Case study | Six exhibits stacked in one stage, each with a sample-data pill. | One artifact per stage, each openable in its own tab. Sample data is said once, in the facts row. |
Why a v2
v1 was correct and unpolished. Set against the Wildeye system it looked like a wireframe: hard black, flat cards, charts that showed data without saying anything. v2 keeps every rule and changes the finish.
- Nothing about the product changed. Anonymity rules, the Pulse budget, counts before rates: all the same.
- Everything visual is a token change first. The surfaces moved because the tokens moved.
What changed in v2
| Area | v1 | v2, and why |
|---|---|---|
| Palette | Pure #000 grounds and pure #fff text. Harsh on the stage and phones, and the neutrals were untinted greys that sat apart from Pulse. | A graphite ramp with a faint violet tint. Dark grounds are #1d1c23, light grounds are paper #f8f8fa, text is graphite #141318. Every text pair re-measured: all pass 4.5:1. |
| Cards | Flat surfaces with a 1px grey border. No depth, so the console read as a wireframe. | shadow-card: a hairline, a contact shadow and a soft lift, low enough to stay calm. Hover lifts one step. Radii up a step (cards 16, sheets 24). |
| Charts | Plots without a point: the reader had to work out the answer. | Wildeye’s grammar: the answer first (latest value, change, verdict), monotone lines with a 10% fill, gutter labels, an end dot, n/a bands instead of zeros, a running period drawn “so far”. |
| New chart | No view of what is happening today. | A state timeline: one lane per team, Pulse while a round is open, ink while the readout is on stage. |
| Phones | A home-made black slab with a fake island. | The same silver IPhoneFrame as the Wildeye and Aeva studies, at the 402×874 screen the Wildeye mobile kit uses. |
| Logo | Dots too small for their ring: at 16–24px the mark read as specks. The wordmark stroke was a hairline next to Geist. | Heavier dots on a tighter ring, plus a halo around the shared mind. The wordmark stroke goes from 5 to 6.5 units. A graphite app tile with a faint Pulse glow. |
| Controls | Flat primary slab; secondary as an outline drawn on the page. | Primary gets a 1px highlight edge. Secondary sits on surface with the card shadow. Inputs gain a focus halo. Tabs select with a lifted pill. |
| Open in a new tab | The console and readout were squeezed into the column with no way out. | Every wide exhibit opens in a new tab, to a full-viewport view in the prototype. Full screen was cut: the exhibits scale to their box and rarely looked right blown up. |
| Coverage | Ten component pages, no state matrices, one phone exhibit. | Button and Text field state matrices, five new pages (Badge, Text field, Tabs, Toast, Room code), and the phone flow as four held moments. |
Palette
The same voice row, timer and button on the v1 and v2 dark grounds. Graphite keeps the contrast and drops the glare.
BEFORE · v1
What’s stopping us?Send
AFTER · v2
What’s stopping us?
Cards
One card grammar for the console, the readout and every specimen.
BEFORE · v1
Hives run48
AFTER · v2
Hives run
48
Charts
The chart now leads with its answer, the way Wildeye’s do. A reader who stops at the first line still leaves with the point.
BEFORE · v1
AgreementShare of voices in each round’s top theme.
AFTER · v2
This week71%+19 pts since 20 Jul
Strong agreementLogo
Same idea, six voices around a shared mind, now literally a hive. v2 made the dots heavier; v3 replaced them with a seven-cell comb that holds together down to 14px, and set the wordmark in Geist Semibold instead of a monoline drawing.
BEFORE · v1
AFTER · v3
Still open
- Test the ultramarine on a projector and a compressed screen share. Saturated blue-violets are the first colours a projector loses.
- Write the ramps and the v4 tokens back to the system’s source of truth (the design-system artifact), so the file and this site agree.
- Find out what the Meet add-on and Zoom app platforms let an add-on do to the room’s microphones; the design assumes only the host’s own mute control.