Project
Principal Data Scientist Framework
A data-first operating system for ML and analytics work with AI coding agents, with evidence labels, leakage and point-in-time discipline, baseline-first modelling, adversarial review, and self-testing regression evals.
- Status
- active
- Started
- Source
- Private repository
AI coding agents make it easy to produce analyses quickly and easy to produce confident nonsense just as quickly. The Principal Data Scientist Framework encodes the discipline of a senior data scientist as an operating contract that an agent must follow: label the evidence behind every claim, check for leakage before celebrating a score, beat a baseline before adding complexity, and submit the result to adversarial review.
Honest scope
The framework enforces a principal’s discipline. It does not possess a principal’s judgment or accountability. It applies rigorous method consistently and can still be confidently wrong. The human supplies domain judgment and owns every decision, which is why the system is built to be challenged and overruled.
How it is used
The framework is distributed as a plugin for Claude Code and is in use on LAISER’s own applied projects. Its roadmap is held to its own rules: the building phase is complete, and the current phase is to prove the loop on real projects and let observed friction, not speculation, write the next version.