TheoryCraft

THE PROBLEM

TheoryCraft answers the question I kept coming back to while trading: would this strategy actually work?

Most trading ideas look fine on a chart. The hard part is proving the edge on enough data, with costs, out-of-sample checks, and a notebook I can inspect instead of a black box.

THE BUILD: 291 DAYS

I started it on Sep 3, 2025 and launched on Jun 21, 2026. It became my first shipped app after roughly 10 months of build.

The stack is not tiny: Phoenix for the product, Postgres for the app data, Python and Jupyter for research notebooks, gRPC for runtime and data boundaries, Docker and Fly.io for exploration sessions, and MCP so Claude, Codex, Cursor, Windsurf or Gemini CLI can drive the workspace.

I also built too much before the launch. Ship or Die forced the cut: finish the MVP, remove features that were not needed, send the launch emails, then deal with marketing in public.

WHERE IT IS NOW

MRR is still $0. There are 2 users in the product, about 30 pre-registrations from the launch window, and the current work is mostly SEO/GEO: comparison pages, llms.txt, a real blog, public notebooks, Bing/Google indexing and AI citations.

LESSONS

1. A deadline works. 18 days of Ship or Die did more for the launch than months of “almost ready”.
2. The MCP and notebook system is the product’s real wedge: external agents can create notebooks, edit them, run research and keep the output portable.
3. Marketing started too late. I was proud of the build before I was proud of the landing page, and that cost time.
4. The next hard part is not technical. It is proving that people can find it, understand it, and pay for it.