The 180

Close

How the 180 works

You get one prompt and three minutes. You talk out loud the whole time. The page turns your speech into text, splits it into phrases at your pauses, and scores each phrase. Everything below is exactly what the code does.

1 · Hearing you

• Browser speech-to-text (Chrome, Edge, Safari, Android): fast and live. The audio is processed by the browser maker (Google or Apple). It always uses your system's default mic.
• On-device speech-to-text (any modern browser, any mic): the page records the mic itself, cuts it at pauses of ~0.65 s, and transcribes each phrase on your device with OpenAI's Whisper model (tiny.en, ~40 MB, downloaded once). Nothing leaves your device. It runs a few seconds behind you.
• Auto uses the browser engine on phones and on-device transcription on computers. Desktop browser engines can fail silently: they get the audio and return nothing. If the browser engine reports an error, the page switches to on-device automatically. If it hears nothing for 12 seconds, it offers to switch.

2 · Scoring each phrase

Each phrase is turned into a 384-number "meaning vector" with a sentence model (all-MiniLM-L6-v2) that runs in the page. Two phrases about the same thing have vectors pointing the same way; their cosine similarity is near 1. Unrelated phrases score near 0. All thresholds below were calibrated on sample brainstorms with this model.

What	Rule
Idea	On topic, and similarity to everything you've already said is below 0.40.
Build	Similarity to an earlier idea is 0.40–0.62: you're adding to it. It's listed under that idea.
Repeat	Similarity 0.62 or more: you're restating it. Not counted.
Leap	A new idea (after your first three) with similarity under 0.20 to every earlier idea: a jump to distant territory.
Stray	Similarity to the prompt below the stray cut-off (0.17 on Normal) and not building on an idea. Kept in a separate list.
Ideas per minute	Ideas ÷ minutes elapsed. This is fluency in creativity research.
Domains	How many of the nine thinking modes your ideas touched. This is flexibility: the number of different categories.
Focus %	Seconds spent green, violet or cyan ÷ seconds spent talking (grey seconds don't count against you).
Longest flow	Longest unbroken run of on-topic seconds.
Stalls	Times you went silent for more than 5 seconds.

Short fragments with fewer than two meaningful words ("um, and so") are joined onto the next phrase rather than scored alone.

3 · The colours

Focus level: 60% the current phrase and 40% the last 15 seconds, both compared with your prompt, placed on a 0–1 scale between "unrelated" (0.08) and "clearly on it" (0.26) on Normal, then smoothed. Loose uses 0.04–0.20, Tight 0.12–0.32. Building on one of your ideas, or saying one of your on-topic words, counts as on topic.

Colour	When
Grey · Stalled	No speech for 5 seconds.
Red · Off track	Focus level below 0.30.
Amber · Drifting	Focus level 0.30–0.55.
Violet · Leap	For 2.5 seconds after a leap.
Cyan · Deepening	On topic but no new idea for 25 seconds: time to jump.
Green · Flowing	On topic with new ideas arriving.

Measured example, prompt "How might we get more people downtown on weekends?": "a farmers market on the river walk" scores 0.25 against the prompt, near the top of the Normal band → green. "Did you see the game last night" scores 0.09, under the 0.17 stray cut-off → a stray, and the screen slides to amber. Follow it with "the fourth quarter was unbelievable" (0.02) → red.

4 · The thought map

This is not a brain scan. A microphone can't measure your brain. The map shows which kinds of thinking your words express, placed on the brain systems neuroscience most commonly links to that kind of thinking. It's an illustration of your content. Real brains use many regions for every task.

A phrase activates a domain from two signals: words from that domain's word list (in the spirit of the LIWC word categories, written for this page; words from your prompt are ignored) at 0.22 each, up to three, plus similarity to a one-sentence description of the domain above 0.22 × 1.6. A total of 0.30 or more lights it. It stays lit for about six seconds. When two or more are lit, the main brain turns to each in turn and each one branches off into its own brain. Broca's area glows faintly whenever you're speaking.

Domain	Brain system	Source

5 · Why this format

• Defer judgement, go for quantity, build on ideas, welcome wild ones: Osborn's original brainstorming rules. Osborn (1953), Applied Imagination.
• Fluency (how many), flexibility (how many categories) and originality (how far out) are the standard ways divergent thinking is scored. Guilford (1967), The Nature of Human Intelligence; Torrance (1966), Torrance Tests of Creative Thinking.
• Scoring originality as distance between meaning vectors is an established research method. Beaty & Johnson (2021), "Automating creativity assessment with SemDis," Behavior Research Methods 53; Olson et al. (2021), "Naming unrelated words predicts creativity," PNAS 118.
• The models: Reimers & Gurevych (2019), Sentence-BERT; Wang et al. (2020), MiniLM; Radford et al. (2022), "Robust Speech Recognition via Large-Scale Weak Supervision" (Whisper).

6 · Limits, honestly

• Speech-to-text mishears; a misheard phrase can score as a stray.
• Everyday talk that happens to be related in meaning can pass as on topic. On the prompt above, "I need to grab groceries and call my mom" scores 0.26 because errands and "downtown" are related. Tight strictness helps a little.
• Similarity scores are good at "same topic or not" and rough at fine distinctions. Two genuinely different ideas phrased alike can count as a build.
• Nothing here has been validated as a way to improve focus or creativity. It measures and reflects; the rules and the clock do the work.
• History and personal bests are stored only in this browser.