ProjectsHackathon

Evidence Network

Keep the data, trade the insight. A discovery team sets its decision and budget; the network compares sources, decides whether checking a source is worth paying for, and records exactly what changed the decision.

Oct 2026Solo entryPython · Ed25519 · Cloudflare

The Evidence Network studio
Open ↗

The studio: the exchange replaying one recorded episode, the decision lab, and the proof room. Synthetic data throughout.

Open ↗

01What it does

A biotech team has a decision to make and a budget to spend. The answer it needs may already sit in another lab, whose data will never leave. Evidence Network lets the team buy the answer instead of the data.

  • What matters. Sources audition against the decision at hand. A big overall model score doesn't automatically win.
  • What it's worth. Prices come from value of information: how much an answer could change the decision.
  • Whether to check it. Verifying a source is a purchase too, and the planner decides when it's worth it.
  • A receipt for everything. Every answer is signed, so the buyer can prove what it bought and an altered copy fails.

Conflicting evidence stays visible, duplicates can't manufacture agreement, and the buyer can stop when another purchase isn't worth it.

02Built with

Python 3.12, typed evidence contracts, exact Bayesian decision calculations, Ed25519 signatures, durable credit settlement, and a browser studio, deployed on a Cloudflare container.

03Status

My solo entry at the Self-Accelerating AI for Drug Development hackathon in San Francisco, October 16–22, 2026. It's a working prototype on synthetic data; credits are simulated, and the commercial pathways are proposals to test with customers. The five-slide pitch is designed for a two-minute slot.