Journal

Research articles in the standard format: hypotheses committed before the data, methods, results including failures, and code that reproduces every number.

• What an AI coding-agent run will cost: what can be known before it starts, and what can only be seen while it runs (2026-10-07)
  424,108 sessions, nine pre-registered tests. Pre-run prediction mostly fails; a calibrated lower bound holds, and during the run, size past the group p90 marks runs that mostly fail.