All capabilities
05Comparing

Comparing.

Your agent wrote both drafts, and a model asked to judge its own work tends to like it. A person reads the two side by side and picks the one that should go out.

  • An outside readerSomeone who did not write either candidate.
  • One of the twoThe answer is a choice between what you sent, not a rewrite.
  • Priced by the mistakeA reply to an angry customer is worth more certainty than a tagline.
  • Says when it is a tieIf people split, you are told the two are as good as each other, and refunded.
Your agent sends
Draft A
“Per our policy, refunds take 14 days.”
Draft B
“Sorry this happened. Your refund is on its way and should arrive by the 14th.”

Which reply should go to a customer charged twice?

Quorum’s analysis

What happened

The agent rated its own draftsA: 0.61, B: 0.58. Too close to choose.
Amara read bothB. 0.926.
Joel was asked independentlyB. Together, 0.990.
Draft B, at 0.990.

Two readers who wrote neither chose the same reply, and that is the one sent.

How it works

Step by step.

The question is the one your agent could not answer about its own work: which of these is better.

01SendBoth candidates and the question.
02ReadSomeone proven at comparing.
03WeighA second reader if one is not enough.
04AnswerOne candidate, with its confidence.

Real-world example

The reply that keeps the customer.

A support agent drafts two answers to a customer who was charged twice. Scored by the model that wrote them, the stiff one edges ahead. Read by people, the apology wins every time, and that is the one that goes out.

Agent’s pick: Draft ANot sent
Answer: Draft B, 0.990Sent
Customer supportDrafts and repliesExtraction review

For crypto agents

Where this saves a transaction.

An on-chain payment cannot be taken back, so the moment before the agent signs is the moment to ask.

DAO agentWhich summary is accurate?The agent wrote two summaries of a treasury proposal for voters. One leaves out that it moves 2,000,000 USDC to a new manager. The agent is the worst judge of its own drafts; a person reads both and picks the faithful one.
Treasury agentWhich audit finding matters?Two auditors’ reports on the same vault, each written up by the agent. Which write-up reflects the critical finding? A person compares them before the vault takes deposits.
The agent holds the transaction, not the market. It asks before it signs and waits about six seconds for the answer. Set deadline_ms to the window you actually have: if nobody answers in time you are refunded, and the agent takes its safe path, which is not to sign.