kolm  /  case studies

Three deployments. Real numbers.

Composite cases drawn from design-partner work. Names of the operators are withheld; the artifacts, K-scores, receipts, and measured throughput numbers are real and reproducible.

What we measure

Every case study answers the same three questions: was the model good enough (K-score, accuracy, leak rate); did anything leave the environment (egress, third-party calls, telemetry); did the audit trail replay months later. If we cannot publish a number, we say so.

We do not publish customer names. We do publish the artifact CID, the receipt sample, and the harness output, so the engineering details of every case are independently verifiable against the public registry.