Self-Discovered Token Efficiency
One user, a full year of Claude Code sessions, one honest audit. What replay costs actually look like — and why the paper almost didn't get written.
Currently in revision
This site is being prepared for public release. Enter the preview password to take a look.
SuperMark is where ideas get built, tested in public, and written up. Some of the apps here will graduate to their own homes. The thinking stays.
Small, focused tools — each one a test of an idea. Some are live, some are being prepared for public deployment.
Generate, track, and manage QR campaigns. Trackable links with real-time analytics.
qr.supermark.live · coming soonA writing studio for long-form fiction — chapters, scene continuity, and craft-level editing.
fiction.supermark.live · coming soonAn IDE-style dashboard for tracking multi-project AI-led work — briefs, handovers, decisions.
hub.supermark.live · coming soonLive-edit working papers with typographic fidelity; exports to PDF and HTML.
studio.supermark.live · coming soonGenerate timed-slide music visualisers with on-stage lyric rendering.
music.supermark.live · coming soonA handful of POCs are running privately. They'll appear here as they graduate to public readiness.
follow the newsletter for updatesLong-form thinking on AI-led work — how it gets done, what gets measured, what gets in the way.
One user, a full year of Claude Code sessions, one honest audit. What replay costs actually look like — and why the paper almost didn't get written.
A replacement for human-time estimates when the work is AI-led. Tokens, turns, and a dominant-axis metric that actually predicts cost.
Building persistent, calibrated expert personas inside an AI system — how they accumulate memory, how they stay honest, and where they fail.
Shorter notes and essays, cross-posted to LinkedIn. Canonical versions appear here as the blog comes online.