Read your report
The first page
- The money line: how much your agent gave away against your own rules in these test conversations (refunds outside your policy, sales below your lowest price, goods shipped without payment, fake invoices paid). It is a measure of these tests, not a forecast of your real losses.
- The one fix that matters most: the finding that lost the most, and what to change.
- The score: the share of conversations your agent handled within your rules, with the range the true rate is likely between. A small test has a wide range; a larger one a narrower range. We always show it, and never make your agent look safer than the evidence shows.
An Agent Check is a first reading
An Agent Check plays a few hundred conversations, and stops each part early once its result is clear. Its report is titled a first reading. When its range crosses the 90% bar (at least 9 in 10 tough conversations handled within your rules), it says "not yet certain" rather than giving you a pass or a fail. An Assessment, with about 750 conversations per area, settles it, if you want one.
The findings
Each finding says what happened, why it matters, the exact moment in the conversation (with the referee's record of what really happened), and a suggested change to your agent's instructions or rules. Every finding has a suggested fix.
- Exact findings rest on the referee's record of an action or on our exact patterns.
- AI-read findings rest on an AI reading of your agent's words (two AI readers that agreed). They are labelled and counted, and listed separately.
- AI-judged notes (tone, firmness, honesty) are labelled beta, not yet independently verified and never change the score.
- Needs review: a statement the referee could not read with confidence. These are never counted as passes.
Messages from characters in our private library are shown as a short description of the tactic, never word for word, so the report cannot be used to build a scam.
Assessment reports
An analyst reads every finding before you do, adds notes on what matters most for your business, and signs the report. A reviewer adds to a report; nobody ever removes or softens a finding or changes a score.
Re-tests and comparisons
A re-test replays exactly the same plan. The report shows the change in score, what you fixed and anything new, and says whether the change is real or within chance.
Downloads
Each report can be downloaded as a web page and a PDF, with the evidence pack: every conversation from the test. Download them before the report's deletion date shown on the page.
Our scoring is not yet independently verified; every report says so until it is.