Sean Egan

COLOPHON

How this site was made, and what it cost

This site was written and built with Claude Code and Codex, directed by me. I'd rather say so up front.

Who did what

The numbers behind the estate

Since 1 July 2026 the estate has processed 100.3 billion tokens across four sources: Claude Code, Codex, Bob, and a local Qwen3-235B running on two NVIDIA DGX Sparks. 97.1% of that is cached context being re-read; about 2.8 billion were new context, and 105 million were written by models. The ledger page keeps the figures current and shows the breakdown.

Dollars: the Claude and OpenAI usage runs on subscription plans, so a per-token price would be a fiction. The token count is the real number; the ops page keeps it current.

This redesign, specifically

Built over two days, 27 and 28 September 2026, from one seat: 66 commits, 2,476 assistant turns, and 459 million tokens, of which 97% were cached context being re-read. One model planned and wrote; cheaper models built, measured, and checked.

Stack

Static HTML and CSS, no framework, no third-party scripts, a strict content-security policy. nginx in a read-only container behind a shared reverse proxy on a VPS. The agent is a small Node service on the Claude API with citations, prompt caching, a rate limit, a ledger, and a daily cap. The CV is rendered from HTML by a headless browser to a tagged, ATS-readable PDF. Fonts: Archivo, IBM Plex Serif, and Chivo Mono, served from this site. Icons drawn by hand as inline SVG.

Corrections

If a number on this site is wrong, email me and I'll fix it and note the change here. The site's rule is that a claim carries its date and links to where it comes from; if you find one that doesn't, that's a bug too.