Checking a credit score in two places often returns two different numbers on the same day. The discrepancy is expected behavior rather than an error in either figure.
There is no single score
A credit score is the output of a statistical model applied to a credit file. Multiple companies build such models, and each publishes several versions.
Models are revised over time as lending data changes, and older versions stay in use because lenders update their systems on their own schedules.
So a file can be scored by a current model in one place and by an earlier one somewhere else, producing different numbers from identical data.
Three bureaus hold three files
The major US credit bureaus maintain separate files. Creditors choose which bureaus to report to, and not all of them report to all three.
A card that reports to only two bureaus is simply absent from the third file. The same model run against those files returns different results.
Timing adds more variation, since bureaus receive updates on different days and a balance reported on the fifth appears later than one reported on the first.
Models are built for specific decisions
Auto lenders and card issuers care about different outcomes, so industry-specific scores weight the file differently and often use a wider numeric range.
A score built to predict auto loan performance emphasizes prior auto accounts more heavily than a general-purpose score does.
That is why a borrower can be strong for one product and merely average for another without any inconsistency in the underlying behavior.
Free scores are usually educational
Scores supplied by cards and banking apps are real scores, but they may not be the model the lender uses when the application is actually decided.
They remain useful for tracking direction. A rising educational score generally reflects a file improving in ways other models will also register.
Using one as a precise threshold is where it breaks down, since the score pulled at underwriting can differ by a meaningful margin.
The file matters more than the number
Every model reads the same underlying file, so the behaviors that move one generally move the others in the same direction.
Payment history, balances relative to limits, age of accounts and recent applications drive the outcome regardless of which model is running.
Reviewing the file itself, rather than chasing a particular number, also surfaces reporting errors, which are the one thing a score alone will never explain.