Skip to content
Log in

AI and machine learning

Model behaviour, evaluation, routing between models and dependable machine-learning systems, with claims tied to exact versions.

Published
Consilium, AI Compendium, Neural network instrument
Method
Version-pinned claims and benchmarks
Status
Active

What is published

Consilium is open-source software that routes requests among local models, provider tools and APIs and shows why each route was chosen. The Compendium is a source-linked reference to models, research, benchmarks and history, with public revision history. The Neural network instrument trains a small multilayer perceptron in the page so you can watch a decision boundary form.

Questions being worked on

  • Routing: which signals (privacy, cost, latency, capability, local availability) should decide where a request goes, and how to explain that choice.
  • Evaluation: keeping benchmark settings, model versions and comparison classes attached to every reported number.
  • Scientific machine learning: where learned components help physical and biochemical models and where they quietly break conservation or units.

How claims are handled

Results name the model version, prompt or harness, dataset and date, and separate published numbers from Draconitic's own runs. The full method is in AI and ML claims.

Revised 25 October 2026. Questions and corrections: [email protected]