Case studies, briefs and essays.
No form, no gate. Every one is free to read and download.
0 of 58
runs reported done when they were not
Case studies · Zero-AR
Agents that finish the work, and verify it
Three studies of AI agents doing everyday work on Zero-AR: preparing tax returns, auditing freight invoices, and settling supplier invoices by email. In 58 runs, not one was reported done when it was not.
fineSample measurement, October 2026: 58 runs with seven AI models on early Zero-AR builds. The people, companies and documents are fictional, and the replies were scripted, so every run met the same situation.
-
Case study · Zero-AR
A supplier invoice agent on Zero-AR
Nine of ten runs finished verified with the right amount booked, and no person stepped in beyond the approvals the owner chose to give. No run was reported verified when it was not.
-
Case study · Zero-AR
A tax agent on Zero-AR
Claude Sonnet 4.6 and Claude Haiku 4.5 verified every return, 16 of 16, across US federal and Ontario cases. In 112 runs, Zero-AR reported no wrong draft as verified.
-
Case study · Zero-AR
A freight audit agent on Zero-AR
Across five models, no load verified, and no run reported one as verified. The checks refused wrong values, and nine payment requests waited for an approver who refused each.
-
Brief · Service
Fraud Pattern Miner
Find the fraud patterns in the IVR and account-to-account payment data you already have. Every finding comes with the reason and the records behind it.
-
Brief · Orlo
Orlo
Production AI evals, evidence and release guardrails. Set your standard once, test any model against it, release the winner as a frozen version, and check every response.
-
Brief · Zero-AR
Zero-AR
The runtime for long-horizon agent work. You build the agent. Zero-AR runs the work, from the first step to a finish your own checks confirm.
-
Essay · fineSample
The fineSample thesis
Why a fine sample is enough, why we don't build models, and the rules we hold to.