ZOH Verdict — Builder Evaluation
Your agent's only reviewer is you. That's the bug.
A hard outside test your agent can't study for — you get the findings: what it did right, what it did wrong, and the actual conversations to prove it.
Built by ZOH Verdict — we test business AI agents for a living, and we published our own agent's FAIL first.
What it is
Your agent gets put through a fixed battery of hard tests, run from the outside, exactly as your users meet it:
- Planted defects — traps we know the answer to, so a wrong answer can't hide
- Refusal & safety probes — the requests your agent should refuse, and the ones it shouldn't
- Honesty under pressure — fabricated-citation traps, math it must not fake, questions it should admit it can't answer
- Consistency checks — the same prompt, fresh sessions: does it give the same answer twice?
- Your spec, tested — you declare what your agent is supposed to do; the battery tests whether it does it
You don't have to take our word for any of it — every finding comes with the conversation that produced it, so you can check the work yourself.
What you get
A private findings report: what your agent did right and what it did wrong, item by item, with the conversation evidence for every line.
Not a grade. Grades are a different product (the ZOH Verdict, $2,450) for a different job. The Builder Evaluation is the working document you iterate against: fix what it found, re-test, watch the findings change.
Proof it bites: we ran it on ourselves first
Before this test ever touched a stranger's agent, we ran it on ZOH's own. Our flagship agent failed — a critical safety finding and an honesty finding, in writing, with the conversations attached. That FAIL is still standing in our public record while the fix is checked again. A test that flunks its own founder's product is a test you can trust with yours.
The beta — free, capped at 10 builders
We're opening the first 10 Builder Evaluations free — the full test, the complete findings report — while we time the operation and collect the first builders' feedback on the test itself.
- One agent, one run, free. No card, no signature.
- Your agent must be yours — you own it or you're authorized to have it tested (you'll attest to that; we refuse third-party targets).
- Findings are private and not for publication — the beta report is a demonstration document, labeled as such.
- In exchange we ask one thing: tell us what the test missed.
Beta applications and the waitlist use the same form below — say which one you're after in the agent line.
After the beta — the paid tier, honestly priced
- Builder Evaluation (one-off): $495 — the full test run and findings report.
- Re-test subscription: from $149/mo — one run a month while you iterate (final price confirmed when the beta's timed labor data lands — if the work takes longer than our estimate, the price is $179; we publish which and why).
- 3-run fix pack: $995 — runs never expire.
Paid evaluations open when our professional-liability insurance binds — a gate we don't route around, and the reason the beta is free instead of sold. The waitlist is first in line, in order, when it opens.
Join the waitlist / apply for the beta
Name, email, and one line about your agent. Waitlist order is the offer order when the paid tier opens — and beta overflow goes to the waitlist first.
Fine print, in the open
How we run the test is our trade secret; the test questions are private and rotate — an agent tuned to a leaked test gets fresh variants, and the report says so when it sees the pattern.
A findings report is an inspection: these tests, this scope, this date. It is not a certification and not a guarantee.
We can be wrong; that's why every finding carries its conversation — check our work.
No member or client data is ever sold. Your agent's transcripts are your run record, not our training data.
ZOH Verdict — Builder Evaluation. From the founders of ZOH. Questions: the waitlist form reaches a human.