People treat a credit score as a measure of financial character. It is a statistical model estimating one specific probability, and understanding what it is actually estimating removes most of the anxiety around it.

What it predicts

The typical consumer credit score estimates the likelihood that a borrower will become seriously delinquent on some obligation within a defined future window.

That is the whole target. Not wealth, not income, not reliability in any general sense.

Which is why a person with substantial assets and no borrowing history can score poorly. The model has nothing to work with, and no data is not the same as good data.

What goes into it

The inputs are the contents of a credit file, which is narrower than people expect.

Payment history on credit accounts, carrying the most weight, because past payment behaviour predicts future payment behaviour better than anything else available.

Amounts owed relative to available limits, which is the utilisation measure.

Length of credit history, including the age of the oldest account and the average age.

Recent applications, since a burst of applications correlates with distress.

The mix of account types, which carries the least weight and receives disproportionate attention.

Income is not in there. Savings are not in there. Employment is not in there.

There is more than one score

This is where most confusion originates.

Multiple scoring models exist from different developers, each with multiple versions in active use.

Different lenders use different models, and some use industry-specific variants tuned for particular products.

Separately, the underlying files at different credit bureaus contain different data, because not every lender reports to every bureau.

So the same person, on the same day, has many different scores, and the number a free consumer app shows may not be one any lender ever sees.

Which means chasing a specific number is chasing something that does not have a single value.

Utilisation is measured at a moment

A practical point with real effect.

Card issuers report a balance to the bureaus once a month, generally at the statement date rather than at the payment date.

Which means someone who charges heavily and pays in full every month can show high utilisation, because the snapshot is taken before the payment.

Paying down before the statement date, rather than before the due date, changes the reported figure. The interest cost is identical either way.

This is a genuine quirk of the reporting mechanism rather than a trick, and it is the single most common reason people with no debt problem see a utilisation drag.

Hard and soft enquiries

An enquiry that occurs because you applied for credit appears in the file and affects the score modestly.

An enquiry for pre-approval screening, or your own check of your file, does not.

Multiple applications for the same type of loan within a short window are generally treated as one enquiry by the scoring models, on the reasoning that shopping for a mortgage is not the same as applying for six cards.

The exact window varies by model, which is another consequence of there being several.

Closing an old account

Frequently advised against, for two reasons that are worth separating.

It removes the available limit, which raises utilisation across the remaining accounts.

And it may eventually affect the average age of accounts, though closed accounts in good standing generally remain in the file for years.

Whether that matters depends on the rest of the file. For someone with many accounts, closing one is close to irrelevant.

What actually moves it

Paying on time, consistently, over a long period. That is most of it.

Everything else is refinement at the margin, and the amount of content devoted to those margins is out of proportion to their effect.

The error question

Credit files contain errors at a rate high enough that checking yours is worth the time.

You are entitled to obtain your file from each bureau, and the dispute process is defined in law with timescales the bureau must meet.

That is the highest-value action available to most people, and it is unglamorous enough that very few take it.

None of this is advice about borrowing, which depends entirely on individual circumstances. It is a description of how the measurement works.

Thin files and the alternative data question

Where the model's limits are most consequential.

A substantial population has too little credit history for conventional scoring to produce a reliable result, which restricts access to exactly the people who might benefit most from it.

Newer models incorporate rent payments, utility payments and bank transaction data, on the reasoning that these demonstrate payment behaviour just as credit accounts do.

Which expands the assessable population and raises questions about consent and about what else transaction data reveals.

Some schemes are opt-in, allowing individuals to add rent or utility history to their file deliberately, and take-up has been low mainly because few people know they exist.

How long things stay

Negative information generally remains for a defined number of years, set by law, after which it must be removed.

The clock typically runs from the date of the default rather than from the date it was settled, which means settling a debt does not restart it — a detail that changes what the optimal action is.