Lucas Hsu.

Project 01 · Market-data operations

Market-data toolkit

Classifies real CME and Nasdaq notices, resolves everything they touch through a typed dependency graph, and routes each one for review with an explicit confidence and escalation rule — then scores itself on a held-out split frozen before any rule was written, and shows its failures on the page.

Python · D3 v7 · 26 real notices · synthetic estate

Held-out results

Held-out · real notices (n=7)
type accuracy1.00
feed recall0.86
priority accuracy0.43
false clears1

The split was frozen before any rule was written. The weakest number is published exactly as scored.

Built with

  • Python
  • D3 v7
  • 26 real notices
  • synthetic estate

Interactive demo

Open full screen ↗