01Telling readings apart

Telling readings apart.

Your OCR turned a smudged “4S.00” into text, and every model after it reads the same text. Asking again gets the same wrong number with more confidence. A person looks at the image itself and says which reading it is.

Looks at the originalThe image travels with the question, not the string your extraction made of it.
One of your readingsYou send the candidates. The answer is one of them, never a third guess.
Priced by the mistakeSend what a wrong figure would cost, and the certainty bought matches it.
Says when it cannot tellIf people disagree, the print is unreadable. You are told so, and refunded.

Your agent sends

The printed line reads “TOTAL 4S.00”. Is the total 45.00 or 4.50?

Its reading, 0.41 confident

4.50

The other reading

45.00

Quorum’s analysis

What happened

Priced from the cost of errorA $40.50 mistake buys an answer at 0.96 confidence.

Amara looked at the receiptAnswered 45.00. One answer reaches 0.926.

Joel was asked independentlyAlso 45.00. Together, 0.990.

45.00, at 0.990.

Two people who had not seen each other’s answer read the same total. About five seconds.

How it works

Step by step.

The question is the one your extraction could not settle, with the evidence it was looking at. Nothing else from your system leaves it.

01SendThe image and both readings.

02AskSomeone proven at reading documents.

03WeighA second look if one is not enough.

04AnswerOne reading, with its confidence.

Real-world example

An expense recorded at a tenth of its value.

A finance agent reads a thermal receipt at 0.41 confidence and records $4.50 against a $45.00 purchase. No error is raised, because nothing knows it is wrong. With the question sent to a person, the right figure is back before the entry is posted.

Agent’s reading: 4.50Not used

Answer: 45.00, 0.990Posted

Document extractionReceipts and invoicesLabels and forms

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.

Payments agentA smudged total on a scanned invoiceThe agent reads $20,500 at 62% confidence. Or is it $2,500? Every reread, and every bigger model, sees the same smudge. A person sees a comma under the ink before $18,000 goes out that can’t be taken back.

Treasury agentA figure in a screenshotA counterparty sends a screenshot of the amount owed. Before paying, a person reads which of the agent’s two readings is on the screen.

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.