NeuroMetric

Personal cognitive self-assessment instrument. Within-subject, on-device, honest.

A multi-mode personal-science suite with a timing-precise cognitive-task battery at its core. Built as a real tool, not a product — no marketing, no percentiles, no "clinical-grade" claims, no pretending mobile hardware can do things it can't.

What this is

• A PWA that measures inhibitory control, interference control, short-term / working memory, and working-memory updating on your own device, over time, against your own rolling baseline.
• An honest methodology pre-committed in tasks/methodology.md: task assignment is algorithmic (not user-picked), context capture is split between objective pre-task and arousal-sensitive post-task, device fingerprint is split into timing-relevant vs forensic-only, and there are five pre-registered hypotheses tested at α = 0.01.
• A multi-mode scaffold (Sampling Tracker, Reflection Library, Big Five IPIP-NEO-120, Joint Notice, Reflecting Mirror) behind a Supabase auth layer with Row-Level Security and opt-in-per-item pair sharing. Schema and policies are in supabase/migrations/001_init.sql. UIs for these modes are not implemented yet — the cognitive tracker is fully implemented.

What this is not

• Not a medical device. Does not diagnose anything.
• Not a benchmark against other people. Does not compute or display percentiles.
• Not a brain-training app. The published evidence for cognitive-training transfer is weak; the app does not claim otherwise.
• Not maintained for external use. Public so it's readable; not a product.

Running

npm install
npm run dev

HTTP on http://localhost:5173. For PWA install testing on a phone:

npm run dev:https

See tasks/deployment.md for deployment options.

Stack

• Vite + TypeScript, strict
• IndexedDB local-first for cognitive sessions (authoritative for timing precision)
• Supabase (Postgres + Auth + RLS) for the multi-mode cloud layer; optional
• No analytics SDKs, no cloud LLM calls, no third-party behavioral event streams
• PWA manifest + service worker for home-screen install

Multi-mode setup (optional)

The Cognitive Self-Tracking mode runs fully local-first without any backend. The other modes require a Supabase project. See tasks/accounts-deployment.md for:

• Creating the Supabase project
• Running supabase/migrations/001_init.sql
• Configuring .env.local (from .env.example)
• Verifying Row-Level Security with two test accounts before real data

Design system

"The Observational Monolith" — sharp edges, monospace data, lavender primary, amber uncertainty indicator, cool-tone palette on #131313. Spec in stitch_cognitive_baseline_instrument/observatory_precision/DESIGN.md.

Methodology

Read tasks/methodology.md before running a single session. Every choice is pre-committed; changing any of them mid-collection invalidates the baseline. Highlights:

• Task assignment is algorithmic, not user-picked (selection bias would contaminate correlations).
• Stimulus onset timestamps use double-requestAnimationFrame so the value reflects painted frame, not DOM mutation.
• Response capture uses pointerdown, not click, to avoid the 50–300ms synthetic-click delay on mobile.
• Device-adjusted RT subtracts the session baseline-tap median. Does not claim to remove the full hardware floor.
• ICC checks gated at N ≥ 30 per task. Hypothesis tests gated at N ≥ 50.
• Stop conditions are pre-committed: hard stop at 60 assigned sessions if no plateau; soft stop at 6 months if compliance < 40%.

Status

Cognitive Self-Tracking mode: implemented (four tasks — Go/No-Go, Stroop, Digit Span, N-Back 2-back). Sampling Tracker, Reflection Library, Big Five, Joint Notice, Reflecting Mirror: data layer + RLS shipped; UIs to be built.