Reading

What we measured, and what we point people at

Benchmarks and engineering notes from building the assurance layer, published with their limits, plus the outside frameworks a reviewer is likely to be holding already.

Benchmarks and notes

Every figure traces to an artifact we can hand you, and the limits of each run are published with the result.

AnalysisAug 20265 minThe rule a stateless authorization engine cannot holdReading tool arguments is table stakes and getting more common. The harder problem is the rule that depends on what this customer already did, in a different conversation, last week.
BenchmarkJul 20266 minA governed small model beat a model costing three times as muchWe ran the same banking conversations through three arms: a small model with enforcement, a larger model alone, and the small model alone. The model upgrade bought a discount on the failure rate. Enforcement bought the guarantee, at a third of the cost.

Frameworks your reviewers already use

Your reviewers already have a vocabulary. We map our controls onto theirs.

Want the raw runs?

Method, per-conversation data and a sample evidence pack go to anyone evaluating us. Bring the workflow you cannot approve and we will point the same instrumentation at it.

Book 30 min