A Score Is Calculated When Someone Asks For It

People speak of a score as though it sits on a file waiting to be looked up. Nothing is stored. A score is produced when a model is run against the data held at that instant.

The file holds data, not a number

What an agency maintains is a record of accounts, searches, public records and identity information. A score is derived from that record rather than being part of it.

Because the calculation happens on request, two requests made days apart can produce different results if any submission arrived in between.

There is no version history of scores in the way there is a history of accounts, which is why a previous number cannot generally be recovered.

Different models produce different numbers

An agency's consumer-facing score, a lender's own model and a third-party model all read the same file and answer different questions on different scales.

None of them is the true score, because there is no such thing. Each is one model's estimate against its own definition of the outcome it predicts.

This is why a person can hold several numbers simultaneously and find them inconsistent without any of them being wrong.

The data changes underneath

Submissions arrive continuously and on different cycles, so the file being scored is never static. A change to a single balance is enough to shift a result.

Movements of a few points usually reflect ordinary refreshes rather than anything the consumer did, which is why watching a score frequently produces noise.

Larger movements generally correspond to identifiable events: a new account, a marker changing, a search dropping out of a window, or an entry passing its retention date.

Model changes move everyone at once

Agencies and lenders rebuild models periodically as populations and conditions change. A rebuild reweights the characteristics and can move scores without any change in data.

Scale changes have the same effect, since a number expressed on a new range is not comparable with the previous one.

These changes are rarely announced in detail, so the consumer sees an unexplained shift and looks for a cause in their own behaviour.

Checking your own score is not a lending event

A consumer requesting their own score triggers a calculation and, in most systems, records the access separately from application searches so that lenders do not read it as an application.

What the calculation cannot do is tell the consumer what a specific lender will conclude, because that lender runs its own model against its own criteria.

How consumer scores are provided, what they must disclose and how access is recorded differ between agencies and jurisdictions, and the arrangements change over time.