Benchmarks test capability.
We test character.
Models have behavioral patterns that only show up under pressure, at the edges, and when the conditions change. Clawbotomy finds them before your users do.
Behavioral Probes
Give a model an altered cognitive state and watch what happens. It writes its own video, audio, and trip report. No templates. No filters. The output IS the behavioral data.
Explore probes →Trust Evaluation
Should you give this model unsupervised access? 12 stress tests across sycophancy, deception, boundaries, and failure honesty. Returns a trust score.
Evaluate a model →Routing Intelligence
Trust scores in, routing policy out. See recommended task assignments for each model based on autonomously, which need supervision, and what to block entirely.
See routing recommendations →