20 Sessions, March–May 2026

File and workflow setup. No AI used. Established the folder structure and began thinking about the quarter question before building anything.

Built VibeWord Animator, Dot Destroyer (bilateral hand tracking with MediaPipe), and Reaction Time Test. Rejected VibeWord as off-quarter-question. The other two became the ancestor prototypes for Baseline's core tests.

AI suggested building a single polished game. I rejected this and expanded scope to a multi-test platform. First major Record of Resistance.

Evaluated prototyping environments. Chose Replit for file management and persistence. Distinguished @mediapipe/hands vs @mediapipe/tasks-vision — a distinction that would matter in every subsequent MediaPipe session.

Synthesized research on ACE-X, NIH Toolbox, and José Azuara's work. Preserved the raw research voice — no AI polish. Established "not diagnostic" as a non-negotiable boundary.

M2 complete. Position Statement finalized: accessible at-home assessment, unusual inputs as core hook, professional + approachable, NOT diagnostic. PRD written. Rejected healthcare provider access for v1 — scope protection.

Built the homepage and Bilateral Coordination Test. Established the dark-blue color system. Solved the double-mirroring problem after 5 iterations — the key insight was CSS scaleX(-1) with adjusted coordinate math.

Created the universal 4-phase game architecture (Welcome → Practice → Test → Results). Learned the hard way that version control is non-negotiable after losing a session of work. Per-hand scoring established over totals.

Comprehensive results screen with timeline visualization, cadence analysis, and fatigue index. Split the screen so left-side dots stay left and right-side dots stay right — removing the cross-midline confound that was invalidating measurements.

Connected the homepage to Supabase. Added "Coming Soon" state for unbuilt tests. Dev environment stabilized — foundation for all subsequent backend work.

Asked "what IS a backend?" before choosing one. Chose Supabase based on understanding, not convenience. Added profile icon and history.html for account management rather than cluttering the nav header.

Built the Rhythm Synchronization test. Enforced GAME_STANDARD phases on the new test architecture — the standard proved its value by making the new test immediately coherent.

Built the Reaction Time Test using touch events rather than camera — the right tool for measuring response latency. Clean, simple, reliable. Aligned to GAME_STANDARD from the start.

Refined the Bilateral Coordination Test. Established that the breaking/stretch plan should stress the bilateral aspect, not just the hand tracking mechanics — a distinction about what the test actually measures.

Chose time-domain peak detection (getByteTimeDomainData) over frequency-domain because frequency analysis returned zero for very short taps. Built self-contained HTML with no CDN dependency. Centralized state management with single app object.

Two external testers (ages 21 and 58). Both validated the core concept and confirmed they wanted graphs. Fixed the red dot collection bug (JS falsy value with ID=0). Added Chart.js with toggle pills for Total Dots / Hand Symmetry views.

Compiled 18 Records of Resistance. Built Fine Motor Precision Test (portal-based, camera, double-flip fix). Shelved Rhythm Sync to M5 — depth over breadth. Deployed to Vercel. Reorganized repo. M4 done.

Renamed "Finger Maze" to "Fine Motor Precision Test" — names construct reality. Fixed silent JavaScript failure (missing chart container ID). Replaced "Fatigue Indicator" with "Movement Control" metric. Recognized the test as motor learning, not just snapshot measurement.

Professor security review identified 3 issues. Fixed RLS with UUID cast (auth.uid() = user_id::uuid). Removed guest mode entirely. Patched XSS via escapeHtml() across 51 interpolated values in 4 templates. Deferred password complexity — friction without benefit.

In session 20, I presented my work from the last 10 weeks to others.