What a Credit Score Is Actually Made Of

The file and the score are different things

Two separate layers exist, and almost every misunderstanding comes from collapsing them into one.

The credit file is a record held by a credit reference agency. It's a factual history: which accounts you hold, when you opened them, what you owe, whether you paid on time, who has searched you. It contains no opinion and no number.

A score is what happens when someone runs a statistical model over that file. The model was trained on how people with similar files behaved, and it outputs a probability expressed as a number on whatever scale the model's author chose.

Because the model is separable from the file, the same file can produce many different scores at the same moment. That isn't an error. It's the design.

This is worth sitting with, because it dissolves a very common frustration. Someone is rejected with a score their app calls "excellent". Nothing is broken. The lender ran a different model, weighted for their own product and their own appetite for risk, and it included information the app never had — such as your income, your existing relationship with them, and their current lending targets.

What's actually in the file

Broadly the same categories everywhere, though the details and retention periods vary by country:

• Identifying information. Name, dates of birth, current and previous addresses, and in some countries whether you appear on the electoral roll.
• Accounts. Every credit account — cards, loans, mortgages, overdrafts, often mobile contracts and utilities — with the date opened, the limit or original amount, and the current balance.
• Payment history, month by month. A running record of whether each account was paid on time, and if not, by how many months it was behind. This is typically the single largest component of a score.
• Searches. A log of who looked at your file and when, split into the two kinds described below.
• Public and court records where applicable, such as judgments, insolvency and bankruptcy.
• Financial connections. In some countries, joint accounts create a link to another person, and their file can then be considered alongside yours.

Two properties of this record matter a lot in practice. First, lenders report it themselves, typically monthly, so the file lags reality by up to a month. Second, negative entries expire after a set number of years, which differs by country and by entry type. They do not stay forever, and they cannot be legitimately removed early by anyone charging a fee.

What the models weigh

Different models differ in the details, but the broad hierarchy is consistent because it reflects what actually predicts repayment.

A few notes on the ones people misread:

• Payment history is about lateness, not amount. A missed payment of a trivial sum is recorded the same way as a missed payment of a large one. The severity comes from how many months behind you went, and how recently.
• Utilisation is a ratio, not a balance. Owing £900 on a £1,000 limit and owing £900 on a £10,000 limit look completely different to a model.
• Length of history rewards old accounts. This is why closing the oldest card you own can move a score in the direction you didn't want.
• Applications are counted, not approvals. The model sees that you asked, whatever the answer was.

Why you have several scores

Three independent reasons stack up:

1. Different agencies hold different files. Not every lender reports to every agency. A card that appears on one file may be missing from another, so the inputs genuinely differ.
2. Different models use different scales. One runs to around 1,000, another to a few hundred, another to nearly a thousand on a different curve. Comparing the raw numbers between them is meaningless; only the band label is roughly comparable.
3. Different models are built for different questions. A model predicting card default and one predicting mortgage default are not the same model, even from the same vendor.

The practical consequence: the free score is a directional indicator, not a value. It's genuinely useful for spotting that something changed, and for reading the file underneath it. It is not the number anyone lends against.

Soft searches and hard searches

Every look at your file is logged, in one of two categories, and the difference is one of the most useful things to know in this subject.

A soft search is visible only to you. Checking your own score is a soft search. So are eligibility checks, identity verifications, and most quotes. Soft searches do not affect scoring models at all — which is the direct answer to the very common worry that checking your own score damages it. It does not.

A hard search is recorded for other lenders to see, and it happens when you formally apply for credit. Models treat a cluster of hard searches in a short window as a signal, on the reasoning that someone applying to many lenders at once may be under pressure. One hard search is a small effect that fades; several in a few weeks is a larger one.

There's an important nuance here. Many jurisdictions and many scoring models apply rate-shopping windows: multiple applications for the same kind of product within a short period are treated as a single search, because shopping for one mortgage is not the same behaviour as opening five cards. Whether this applies, and for how long, depends on the model and the country.

The timing trick nobody explains

Here's a mechanism that surprises almost everyone. Your card issuer reports your balance to the credit agency on a specific date each month — usually your statement date, not your due date, and not the day you pay.

So a person who spends heavily and clears the card in full every month, never paying a penny of interest, can still show high utilisation on their file — because the snapshot was taken at the statement date, when the balance was at its peak. The file has no way to see that it was paid off a fortnight later.

This isn't a loophole or a trick to exploit; it's just how the reporting cadence works. But it explains a genuinely puzzling situation: careful, debt-free card use showing up on a file as heavy borrowing. The date that matters is the statement date, and it's printed on your statement.

What is not in your file

The list of absent things is longer than most people expect, and it corrects several persistent myths:

• Your salary or savings. Credit files do not record income or bank balances. Lenders ask you for income separately, on the application.
• Your debit card spending. Ordinary current-account transactions aren't reported.
• Race, religion, nationality or, in most systems, gender. Using these is generally prohibited outright.
• Your partner's file — unless you hold a joint financial product, which creates an explicit link.
• Whether you rent, in most systems, unless you or your landlord opted into a scheme that reports it.
• Parking tickets, most fines, and unpaid bills, until and unless they escalate into a formal default or a court judgment.

Common misconceptions

• "Checking my score lowers it." No. That's a soft search and it is invisible to models.
• "I have no debt, so my score should be perfect." Models predict how you handle credit. Someone with no history is not a proven low risk; they're an unknown, and unknowns score cautiously. This is why a thin file and a bad file can produce similar outcomes for very different reasons.
• "Closing old cards tidies things up." It shortens your average account age and reduces your total available credit, which raises your utilisation ratio on the same balance. Both effects run against you.
• "One score exists and someone is hiding it from me." Covered above. There genuinely isn't one.
• "A company can remove accurate negative entries." Accurate records stand until they expire. What you can do free of charge is dispute entries that are wrong, and every agency is legally required to investigate.
• "Being rejected damages my score." The application was recorded; the rejection itself is not reported. But applying again immediately adds another hard search, which is why understanding the reason first matters.