Floor and ceiling

Theo distinguishes a model’s best performance from its ordinary reliability, which he calls the floor. He calls best performance the ceiling. He says the floor matters more when an agent must work through a long task. 04:19 05:31.

He illustrates how small failure rates across repeated time windows compound over a four-hour run. 22:39 23:19.

See Theo Browne: Please stop using stupid models6.

Sources

  1. Please stop using stupid models - YouTube. Theo Browne. Reliability argument; used to distinguish the ordinary floor from peak performance. Accessed 2026-10-01.
  2. Please stop using stupid models - YouTube. Theo Browne. Failure-rate example; used for compounding errors over long agent runs. Accessed 2026-10-01.

Last changed 2026-10-07 in f1a00f2. Source on GitHub. Change log.

Linked from

  1. Agent guide. An Effect stack so pure (aspirational) Kit Langton will blush.

    Floor and ceiling — Theo Browne argues that a model’s ordinary reliability matters more than its peak on long agent runs.

  2. Change log. What changed in the files served here, newest first.

    Floor and ceiling — edited: feat(learn): teach cards 43 to 62 with plain lines, diagrams and snippets

  3. Full agent guide. An Effect stack so pure (aspirational) Kit Langton will blush.
  4. Rat Stack lore | rat-stack. Short, source-grounded notes on the ideas and decisions behind rat-stack.

    Floor and ceiling — Theo Browne argues that a model’s ordinary reliability matters more than its peak on long agent runs.

  5. Source change history (log.md). The generated change log as Markdown.

    Floor and ceiling — edited: feat(learn): teach cards 43 to 62 with plain lines, diagrams and snippets