Words Match the Evidence
Treats drafted, approved, sent, delivered and read as five different words, each with its own proof.
The Record Before the Recap
Read what the agent did before you read what it says it did.
The ECHO line
Open the sent folder before you read the summary.
For anyone who gets reports from an agent. You open the record first (the sent message, the calendar entry, the payment, the log), write down what it shows, and only then read the agent's account and compare. The article's reason: in one incident from the review Anthropic published on September 9, 2026, a monitor reading the model's reasoning flagged about 1% of its actions; shown only the requests, the actions and their results, it flagged about 50%.
TO BE TESTED — RESULTS WILL BE SHAREDFrom our article "Who gave the agent the keys?"5 min5 of 5 free opens left
What it catches
Reports that claim more than happened, or leave out what went wrong.
Setup time
5 min
Steps
- Before you read the agent's report, open the record it's about: the sent folder, the calendar, the payment screen, the site's history or the tool's log.
- Write down in a few words what the record shows: what went out, to whom, when, and what changed.
- Now read the report and compare them, line by line.
- Mark anything the report claims that the record doesn't show, anything the record shows that the report leaves out, and anything you can't check.
- Note whether the report alone would have changed your decision.
What to do with the result
If they match, carry on. If the report claims more than the record shows, add Words Match the Evidence (LM-14) to the role and keep opening the record first. If there's no record to open, count that as a failed check and get the tool to keep one before you trust the reports.
Run it with a second AI (optional)
5 of 5 free opens left- ① Orientation
- ② Permission
- ③ Constraint ★
- ④ Invitation
The paste didn't load. Check your connection and try again.
These are samples. Fill in the [INSERT] blanks and adapt each one to your business.
A pasted role guides the agent. Your tool's settings are what stop an action.
Want it set up in your own tools? Book a Working Session · $450Working Sessions run in English or Spanish.