A credit score is not a record held about a person. It is a number produced on demand by running a model over the contents of a credit report, and different models produce different results from the same data. Practice varies by country.
The score is a prediction, not a judgement
Scoring models are built by observing which past borrowers went on to miss payments and identifying the patterns in their reports that preceded it.
The output is an estimate of the likelihood of a particular outcome within a defined period, expressed on a scale that lenders interpret against their own criteria.
Because it is a statistical prediction, it describes a group of people with similar reports rather than making any statement about an individual's intentions.
Payment history carries the most weight
Whether accounts have been paid as agreed is consistently the largest input, since past repayment behaviour is the strongest available predictor of future repayment.
Severity, recency and frequency all matter separately. A recent missed payment weighs more than an older one, and the effect fades as the record ages.
Items such as accounts sent to collection or legal judgements sit in the same category and remain on a report for a period fixed by law.
Balances are read relative to limits
On revolving accounts the model looks at balances as a proportion of the available limit rather than at the absolute amount owed.
The proportion is generally taken from the balance reported on the statement date, which means someone who pays in full each month can still show a high figure.
Installment loans are treated differently, because a mortgage or car loan is expected to reduce on a schedule and does not carry the same signal.
Age and mix contribute smaller amounts
The length of time accounts have been open contributes, which is why closing an old account can affect a score through the average age of the file.
Applications for new credit generate inquiries, and a cluster of them within a short period is generally treated as a single episode of shopping for one loan.
Having several types of credit contributes modestly, though it is far less influential than repayment history and balances.
There is no single score
Multiple scoring companies exist, each with several model versions, and lenders may use industry-specific variants tuned to a particular product.
Reports also differ between bureaus, because not every lender reports to all of them, so the same model can return different numbers from different sources.
A score obtained directly by a consumer is therefore indicative rather than identical to the one a lender sees, which is a routine source of confusion.