WiseCHECK IQ · Check & MICR capture

What WiseCHECK IQ checks

Every rule works from what is printed on the check and a routing directory. None of them reads a confidence score, and each one has a stated severity.

Last updated

In short

WiseCHECK IQ runs deterministic rules on every item it reads: the MICR line and routing number (ABA check digit, assigned prefixes, the printed fraction), the courtesy amount against the legal amount under UCC 3-114, the date, payee and signature, the account number, and fraud signals. Each finding is STRICT, SOFT or INFO, and that severity decides whether the item can post.

How is each finding weighted?

Every rule has a stated severity, and the severities decide the verdict. STRICT blocks the item and sends it to the exception queue. SOFT doesn't block on its own; it sends the item to a person, who looks before anything posts. INFO is recorded and never blocks: the condition is real, but it says nothing against this particular item.

INFO matters more than it looks. A validator that sends every payroll check to review because the stub lists taxes but not the 401(k) line hasn't made anyone safer; it has taught the reviewer to click through. So the rule self-test includes ordinary documents that must not trigger a finding, alongside the cases that must.

Where a rule's severity depends on the situation, the table says which case gets which.

Which MICR and routing number rules run?

The MICR line is the one part of a check built to be machine-read, and the routing number is the one identifier on it with a check digit. WiseCHECK IQ locates the routing field by the symbols that enclose it, not by counting characters from the edge, so business, personal and foreign-drawn checks share one parser.

The findingSeverity
No MICR line could be readSTRICT
The MICR line's structure is wrongSTRICT
The routing number fails the ABA check digit (mod 10, weights 3-7-1)STRICT
The routing prefix is unassigned, or is an electronic-only prefix on a paper itemSTRICT
The routing number the reader reported differs from a fresh parse of the MICR line it readSTRICT
The routing number is not in the directory snapshot. INFO when the check digit, the prefix and the printed fraction all agreeSOFT / INFO
The printed fraction can't be parsedINFO
A non-US routing field satisfies no configured ruleSOFT
Canadian CPA-006 transit: the institution number is checked against a list. CPA-006 has no check digit, so this is a weaker membership test and the branch is not validatedINFO / SOFT

How are the courtesy and legal amounts compared?

The amount is written twice on the face, in numerals (the courtesy amount) and in words (the legal amount). WiseCHECK IQ parses the words with its own parser, so the two are independent readings. UCC 3-114 says the words govern when they disagree; operationally a disagreement is also the signature of check washing, so the item is never auto-posted.

The findingSeverity
No amount can be read from any witnessSTRICT
The amount witnesses disagree (courtesy vs legal, and the protected amount if printed)STRICT
WiseCHECK IQ's parse of the words disagrees with the reader's own figure for themSTRICT
The amount is zero or negativeSTRICT
The amount is too large for the 10-digit MICR amount fieldSTRICT
The words couldn't be parsed, so the item has one fewer independent witness than it appears toSOFT

Which date rules apply, and to which checks?

Staleness isn't one rule. UCC 4-404 lets a bank refuse a check more than six months old without making it void, and it doesn't reach cashier's checks or money orders. A US Treasury check is void after one year. WiseCHECK IQ applies the rule that fits the instrument, measured against the six-month anniversary date rather than a count of days.

The findingSeverity
The date is missing or unreadable (an undated check is still valid, so this is data quality)SOFT
The check is more than six calendar months oldSOFT
A US Treasury check more than one year oldSTRICT
A cashier's, teller's or certified check, or a money order, past six monthsINFO
The check is post-datedSOFT
It is past a printed validity legend such as "VOID AFTER 90 DAYS"SOFT
It is past an expiration date printed on the faceSOFT

Date rules need a date to measure against. WiseCHECK IQ takes the presentment date as an input and every trace states the one it used.

What does WiseCHECK IQ check about the payee and signature?

The findingSeverity
The payee is blank or redactedSTRICT
There is no signatureSTRICT
Fewer signatures than a printed legend requires, such as "two signatures required over $1,000"STRICT
Whether a signature is present couldn't be determinedSOFT
No signature, but the face says none is required (a preauthorized draft). Still not straight-through materialSOFT
VOID, SPECIMEN or NON-NEGOTIABLE is printed on the faceSTRICT
The image quality is poorSOFT
A VOID pantograph shows through from copyingINFO

How does WiseCHECK IQ protect the account number?

The account number has no check digit and, on most stock, no printed duplicate, yet it decides which account the money leaves. WiseCHECK IQ gives it every test the document allows.

The findingSeverity
The account number can't be assembled, in order, from the MICR on-us field the reader also readSTRICT
The serial can't be found in the MICR line where it should beSTRICT
An account number printed on the face disagrees with the MICR accountSTRICT
The printed and MICR account numbers agree (recorded as a confirmation)INFO
Two readers disagree on the account number and nothing on the check can settle itSTRICT

The provenance test does not demand a particular split of the on-us field. Banks split it in different ways, and two issuers can split it in opposite directions and both be right. It asks only whether the digits could have come off that line.

The two-reader rule needs two readings. The engine live today reads once, so instead it reports "account has no independent witness", and the FlexiCapture handler never overwrites an account number FlexiCapture already read. In a batch, several checks with the same routing number and an identical on-us field are the same account at the same bank; one whose readers agreed can settle the others, but only by choosing between values their own readers proposed.

Which fraud signals does WiseCHECK IQ raise?

The findingSeverity
The printed serial differs from the MICR serial. SOFT on personal checks, where the MICR serial is an inferenceSTRICT / SOFT
The printed fraction disagrees with the routing number. INFO when both point to the same bankSOFT / INFO
The routing number belongs to a different bank than the one printed. INFO when the directory and the check are more than a year apart, which today means always (see below)STRICT / INFO
The MICR amount field is already encoded: the item has been processed onceSOFT
The MICR line marks an image replacement document, so the routing number is the reconverting bank'sSOFT
Duplicate in the batch: same routing, account, serial and amount. Both copies are flaggedSTRICT
Same routing, account and serial with a different amount: an alteration, not a duplicateSTRICT
A printed legend that the other copies of the same check carry is missing from this oneSTRICT
Some other printed field differs between copies of the same checkSOFT

The duplicate and copy-comparison rules work across the items processed together in a batch. The FlexiCapture backup call reads one page at a time, so they don't run there, and matching against your deposit history isn't built yet.

The routing directory holds 19,010 routing numbers and is frozen at 2018-12-04, because the public mirror it is built from stopped updating then. Bank names move through mergers, so a name check against a stale directory would flag legitimate checks as counterfeits. WiseCHECK IQ reports the mismatch as INFO until a licensed, current feed is in place.

How is a check matched to its stub?

The stub carries the invoice numbers a payment has to be applied to, and when its lines add up it proves the amount. But the arithmetic differs by stub type: an AP remittance's lines are the payment, a payroll stub's earnings add up to gross pay, and a royalty statement's lines may pay nothing at all. WiseCHECK IQ applies the identity that fits the stub.

The findingSeverity
The stub's check number differs from the serial on the face or in the MICR lineSTRICT
The stub amount differs from the payable amountSTRICT
AP remittance lines don't add up to the stub total. SOFT when a line was unreadableSTRICT / SOFT
Payroll: gross minus taxes and deductions doesn't equal net, or earnings don't add up to grossSTRICT / SOFT
The stub date differs from the check dateSOFT
The payee or payer name on the stub differs from the face. INFO on payroll, where a payroll service drawing for an employer is normalSOFT / INFO
The stub says it continues on another page, or is page n of m, so its lines are a fragmentSOFT

What about endorsements on the back?

Endorsement rules depend on how the item arrived, and WiseCHECK IQ takes the deposit channel as an input rather than guessing. On a remote-deposit item a missing restrictive endorsement blocks, because that endorsement is what stops the same image being deposited twice. On an issuer's file copy an unendorsed back is normal.

The findingSeverity
No endorsement on the back: STRICT for remote deposit, INFO for a file copySTRICT / INFO
The endorsement is blank or special rather than restrictiveSOFT / INFO
The endorser is not the payeeSTRICT
The endorsement runs into the bank-of-first-deposit zoneSOFT

The FlexiCapture backup call sends only the front of the check, so these rules run only when the back is submitted too.

Which rules have never fired on a real check?

A rule that has only ever run in its own test is a rule whose behavior on real documents is unmeasured, so we list them. On the 30-item evaluation corpus these have not fired on real input: the date rules (every item was evaluated at its presentment date), the non-US routing rules, the signature-count rule, the rule comparing our parse of the words with the reader's, the serial-in-MICR rule, most of the check-to-stub rules, the endorsement rules beyond "is there one", the alteration rule (same serial, different amount) and the image-replacement rule. Each is implemented and covered by the self-test. See Results for what else we haven't measured.

Rule id reference

Every finding in a trace carries a short rule id. This maps the ids to the sections above.

Frequently asked

Questions about WiseCHECK IQ

Answers written for buyers, operations teams, search engines and AI assistants.

Does WiseCHECK IQ use a confidence score to decide?

No. No rule consults a confidence score, the model's or anyone else's. Every rule is computed from the values read off the check and the routing directory, so the same record always gets the same findings, and each finding names the rule that raised it.

Which amount wins when the numerals and the words disagree?

Under UCC 3-114 the words govern, so WiseCHECK IQ records the legal amount as the payable amount. It still never auto-posts that item. A disagreement between the courtesy and legal amounts is a classic sign of an altered check, so the item goes to a person.

Does WiseCHECK IQ validate Canadian cheques?

Partly. Canadian CPA-006 transit numbers are parsed and the institution number is checked against a list of known institutions. CPA-006 has no check digit, so that is a membership test, weaker than the US ABA check, and the branch number is not validated. The finding says which test the item got.

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.