A second, independent reading from WiseAI Council
One reader can't catch its own mistakes. WiseCHECK IQ is connectable to WiseAI Council, so a model from a different lineage reads the same check and the check's own arithmetic settles any disagreement.
Last updated
WiseCHECK IQ is being connected to WiseAI Council, WiseTREND's multi-model AI service, as an independent second reader. The roles are split: the Council reads, reporting which model read the check and whether the readers agree; WiseCHECK IQ decides what the agreed values mean and whether the item is safe to post. The engine live today reads each check once.
The WiseCHECK IQ engine running behind FlexiCapture today reads each check once. The WiseAI Council second reading described on this page is the product direction and is in development. It is not generally available yet.
Why does a check need a second reader?
A reader can't catch its own mistakes. If it drops a digit from an account number, nothing it produced later will disagree with it. The routing number protects itself with a check digit, and the amount is written twice, but the account number has neither. A second reader is the one independent witness it can get.
The second reading only counts if it is independent. Two readings by the same model tend to make the same mistake, agree, and prove nothing. So the second reader has to come from a different model lineage than the first.
Who does what: the Council or WiseCHECK IQ?
The roles are split, and neither side does the other's job.
| WiseAI Council reads | WiseCHECK IQ decides | |
|---|---|---|
| Job | Seats a model to read the check and reports whether the readers agree | Decides what the agreed values mean and whether the item is safe to post |
| Output | A reading of the fields | Findings, a verdict and a trace |
| Free text | Comments are advisory | No rule ever reads free text |
| Approval | Cannot approve anything | Never auto-approves an item because a second reader agreed |
The Council is asked to seat a model from a different lineage than WiseCHECK IQ's own Azure OpenAI reader, so the two readings are genuinely independent.
How will two readings be merged into one record?
Field by field, in this order, and by nothing else.
- Agreement. Both readers give the same value after normalizing it. It stands.
- Constraint. They differ, and a test on the document tells them apart: the ABA check digit or the printed fraction for a routing number, the MICR line for an account or serial, the parsed words for an amount. The value the test accepts stands, and the record says which test decided.
- One-sided. Only one reader gave a value. For an ordinary field it is kept and marked as coverage, not agreement. For a field that decides the payment it is not taken on one reader's word.
- Unresolved. Both gave a value, they differ, and nothing on the document can decide. The field is marked unresolved and reported. There is no tie-break by preference or by confidence.
When the unresolved field is the account number, the item stops, because that is a STRICT finding and it blocks like any other.
What did two readers show in evaluation?
On the 81-page evaluation corpus, the merge settled field disagreements like this: 744 field values agreed, 11 were settled by a test on the document, and 120 were left unresolved. Most disagreements can't be settled from the document, and saying so is the point.
The two readers agreed on 91.6% of the field values both populated, over 24 scored fields (84.5% if a field one reader left blank counts as a miss). That is agreement between readers, not accuracy. In that evaluation one reader was run interactively rather than as an automated service. The full figures and their limits are on Results.
How will a Council reading be requested?
WiseCHECK IQ will submit the check to WiseAI Council as a confidential bank-check task, then re-run its arbitration with both readings. A reading will be requested per item, on demand, and in the first phase only WiseTREND staff will be able to request one. A second reading can't move an item to AUTO by itself: it adds a witness, and the rules decide.
Learn more about WiseAI Council, WiseTREND's multi-model AI service.
Questions about WiseCHECK IQ
Answers written for buyers, operations teams, search engines and AI assistants.
Is the WiseAI Council second reading available today?
Not generally. The engine running behind FlexiCapture today reads each check once. The Council second reading is in development and is the direction the product is going. Talk to us if you want it for your checks.
Can the Council approve a check?
No. The Council supplies a reading of the fields, not a decision, and its free-text comments are advisory: no rule uses them. WiseCHECK IQ's rules decide, and a second reading never auto-approves an item.
Why does the second reader have to be a different model?
Two readings by the same model tend to make the same mistake, so they agree and prove nothing. The Council seats a reader from a different model lineage than WiseCHECK IQ's own, which is what makes an agreement worth something.
Send us the checks that reach your verification queue
The handwritten ones, the bad scans, the stock you have no layout for. We'll run them through WiseCHECK IQ and show you, item by item, what it read, what it checked and why it decided what it did.