Field Notes

Essays on Enterprise RL Environments

Analysis on the evolving market for reinforcement learning training data, frontier model capabilities, and the infrastructure required to train models that operate within real enterprise constraints.

Analysis April 13, 2026 · 8 min read

Reading the Claude Mythos System Card: What It Means for RL Environments

Anthropic's newly released Mythos System Card is the most detailed public window into how a frontier model is trained, evaluated, and judged. A closer reading reveals something important about the training data landscape — and why top-down environment construction is becoming the preferred approach.

Read the full post →