Evaluation frameworks are part of the product, not paperwork around it
When an LLM or predictive model enters a workflow, the evaluation setup becomes part of the system design. It determines what the team can trust, compare, and improve.
Research notes
Methods · data · models
Registry data is valuable because it exposes process, not just outcomes
Good registry work is rarely glamorous, but it gives you a direct view of definitions, coding behavior, and quality measurement in operational settings.
Baseball forecasting keeps you honest about what a model can know
A useful baseball model has to live with variance, aging, role changes, and incomplete information. That makes it a strong proving ground for calibration and disciplined claims.