Open-source diagnostic prototype

Decision-state diagnostics for applied AI workflows

Decision-PGA is a prototype framework for describing the shape of uncertainty around AI decisions before a workflow acts. This site gathers the article series, the kinematic trajectory companion, a synthetic demo, the agent toolkit, the public code repository, and PDF copies.

Read the article Read the follow-up Read kinematics Try the demo Use the toolkit View code Download PDF

This is a personal technical perspective, not an institutional statement. The article uses no patient data, is not clinical validation, and is not a medical device or clinical decision support product.

What This Site Is For

This site is a compact public entry point for Decision-PGA: the framing article, the Telescoping Decision-PGA companion perspective, the Kinematic Decision-PGA trajectory companion, the synthetic document-triage demo, the agent toolkit, the open-source prototype repository, and PDF versions of the articles.

Article

The article explains why workflow-oriented AI systems need diagnostics for the shape of decision uncertainty.

Telescoping Decision-PGA

The follow-up perspective shows how broad uncertainty clouds can contain smaller local substructures and cross-document evidence bridges.

Kinematic Decision-PGA

The kinematic companion explains how observed decision states can move across retrieval, tool use, and final output, including RAG/tool whiplash and jerk as review signals.

Synthetic demo

A document extraction triage fixture shows how probability clouds can map to workflow actions.

Agent toolkit

A five-minute path shows CLI, Python API, MCP launch, and synthetic agent payloads for developers who want to try the diagnostic contract in their own workflow vocabulary.

Public code

The initial Decision-PGA prototype is available at github.com/zmichels/Decision-PGA.

PDF

A browser-rendered PDF is included for readers who prefer a document-style copy of the article.

Current Status

Article series and PDFs: available for public reading and critique.
Kinematic trajectory page: companion perspective on velocity, drift, and jerk.
Demo: synthetic, non-clinical, and designed to build intuition.
Toolkit: copy-paste agent examples and a local MCP quickstart.
Code: public initial research release at https://github.com/zmichels/Decision-PGA.