CognitiveOS

A cognitive load simulator, not another to-do app. Lay out a day of tasks, meetings, alerts and deadlines; CognitiveOS estimates minute by minute how loaded your mind is, shows what is driving it, and lets you try a different way of working before you live it.

The model adds up six drivers of load, each tied to published research: task complexity, switching and attention residue, notifications, open loops, deadline pressure and fatigue. Because load is a sum, every "why?" answer is exact arithmetic, not an approximation.

Not a diagnostic tool. CognitiveOS produces an estimate under stated assumptions. It does not measure anyone's ability and is not for monitoring other people. Values marked stand-in are chosen values awaiting calibration.

📄 Project documentation (PDF): the science, the model, scenarios, screens, architecture and the 6-week plan, written before any code.

Status

Weeks 1–4 of the 6-week plan are complete, plus the landing page:

• params.ts: every value the model uses, with its citation and a stand-in flag (19 sources in sources.ts)
• Day file format (Zod), with overlap and bounds checks
• Deterministic minute-step engine: seeded alert arrivals, switches with attention residue, alert detours, batching and quiet windows, open loops, deadlines that ease as work gets done, fatigue that drains on breaks
• Exact per-minute explanations: driver shares plus the events still contributing ("Switched from Code to Report at 11:05, still leaving residue")
• Levers as pure functions: batch alerts, write plans, cluster meetings, add breaks, mute alerts, quiet windows, drop or hand over a task
• Scenario library: the sample day plus six people, each with a lever and a checked expectation
• Property tests (fast-check): deterministic, shares add up, bounded, more alerts never help, plans never hurt, breaks never add fatigue, fast enough to re-simulate while dragging
• Scroll-driven 3D landing page (three.js): nine vignettes, including the sample day as a 3D landscape built from the real engine output, morphing into its what-if
• The crew: six original mascots, one per driver (Brainy, Flip, Ping, Loopy, Tick, Yawn), drawn in code with moods from calm to frazzled; they walk across the hero and react to load in the app
• Week 3: Day Builder (/app.html): drag to move, drag edges to resize, drag past a neighbour to swap, click a lane to add, full keyboard control; the engine re-runs in a Web Worker as you drag
• Week 4: Load chart on the same time scale as the timeline, scrubbable cursor, the crew panel and an exact "why?" card, zone minutes and driver shares, alert and open-task controls, the scenario library with a one-click lever
• Week 5: What-If Lab (side-by-side compare, each lever's share of the change)
• Week 6: NASA-TLX check-in, polish, deploy

One honest result already: in the chopped-up maker scenario, clustering meetings cuts morning switches but creates one long hard block, so fatigue rises. Breaks win part of it back. The test suite checks for that trade-off rather than pretending every lever is a free win.

Run it

npm install
npm run dev        # landing page at http://localhost:5173 · simulator at /app.html
npm test           # engine tests (Vitest + fast-check)
npm run build      # typecheck + production build
npm run docs       # rebuild the documentation PDF (needs Chrome)

Layout

src/engine/        the model: pure TypeScript, no UI imports
  params.ts        every number, with its source and stand-in flag (single source of truth)
  sources.ts       the 19 references
  schema.ts        day file format (Zod)
  simulate.ts      minute-step tick loop
  explain.ts       exact "why?" for any minute
  summary.ts       zone minutes, driver shares
  levers.ts        what-if transforms (pure Day → Day)
src/scenarios/     the sample day and the six-person scenario library
src/characters/    the crew (canvas-drawn mascots and cards) and the walking crowd
src/app/           the simulator: Day Builder, load chart, crew and why panels
src/landing/       scroll-driven 3D landing page (three.js scene + chapters)
docs/              project documentation (HTML source + PDF)

The figures in the PDF come from an earlier 90-line prototype (docs/source/model.mjs), so their numbers differ slightly from the engine in src/engine. The landing page uses the real engine.

Credits

• The walking crowd adapts Skiper UI's Skiper 39 · Canvas_Landing_004 CrowdCanvas (skiper-ui.com, by @gurvinder-singh02), itself inspired by zadvorsky's CodePen. Its walkers are the CognitiveOS crew rather than Open Peeps.
• Animation by GSAP (standard no-charge licence); 3D by three.js.
• Visual style shared with my portfolio.

© 2026 Ananya Penumudi.