What the score measures
Every rating on this site is a number between 1.0 and 5.0 measuring one thing: disclosed cost. It is computed from figures the issuer has actually stated — in its own published rates-and-fees disclosure, or in the cardholder agreement it filed with the Consumer Financial Protection Bureau — and from where each of those figures sits among every other card on this site.
Nobody assigns it. There is no panel, no editor's discretion and no adjustment. The same data produces the same score every time the site is built, and no issuer can buy, influence or appeal one.
What the score deliberately ignores
This is the part to read before you use the number.
The rating says nothing about rewards, sign-up bonuses, cashback rates, travel credits, purchase protection, lounge access, customer service, the quality of the app, how hard the card is to get, or whether you personally would be approved. Not because those do not matter — for many readers they matter more than the pricing — but because none of them is in any document we hold. Filed agreements cover pricing and obligations. Issuer rates pages cover pricing. Neither describes a rewards programme in terms we could score, and a programme can be changed by the issuer at will in any case.
So a card can score 4.6 here and still be the wrong card for you, and a card can score 2.1 and be worth every penny for what it pays back. The score is a cost ranking, not a verdict. Anyone presenting a pricing-derived number as an overall judgement of a credit card is overstating what the data can carry, and we would rather say so here than have you find out from a page that does not.
The components and their weights
Six components. Weights are out of 100 and apply before the renormalisation described below.
| Component | Weight | Status | How it is read |
|---|---|---|---|
| Purchase APR | 34% | In use — 165 cards disclose it | Lower is better. The top of the disclosed range, so a card is not credited with a best-case rate it only offers to some applicants. Cards priced as a margin over Prime are never scored here. |
| Annual fee | 20% | In use — 284 cards disclose it | Lower is better. The recurring fee as disclosed. |
| Foreign transaction fee | 14% | In use — 86 cards disclose it | Lower is better. Applies to anything processed outside the US, including online purchases from foreign merchants. |
| Grace period | 12% | Not in use — only 7 cards disclose it | Longer is better. The window in which purchases do not accrue interest if the statement balance is paid in full. |
| Introductory purchase offer | 10% | In use — 99 cards disclose it | Longer is better, measured as the length of a disclosed 0% introductory purchase rate. A disclosed introductory rate above 0% scores as no 0% period; a 0% offer with no stated length is not scored. |
| Penalty APR | 10% | Not in use — only 8 cards disclose it | Lower is better. A filing stating that no penalty APR applies is scored as zero, because that is a disclosure and a favourable one. |
Two of those need explaining, and the table is where most sites would hide it. 2 components are marked not in use: grace period and penalty apr. Fewer than 40 cards on this site disclose either of them, which is below the floor at which we are willing to call a percentile meaningful, so they are dropped for every card rather than used for the handful that have them. No card's score currently includes those components. If coverage improves, they come back automatically and this table will say so.
How a component becomes a number
Each component is scored as the card's position in the distribution of that same figure across every card on the site, not against a scale we invented. A card whose annual fee is lower than four-fifths of the fees on the site scores four-fifths on that component.
Ties take the middle of the tie group rather than the top of it. That matters more than it sounds: several hundred cards here disclose a $0 annual fee, and letting each of them claim to have beaten all the others would hand every no-fee card a component score close to perfect. On the mid-rank they all score at the middle of the group they share, which is what is true.
A component whose distribution holds fewer than 40 values across the whole site is not used for any card. A percentile drawn from nine cards sounds measured and is not.
The renormalisation rule
Only components a card actually discloses are weighted, and the weights of those that remain are rescaled to sum to the whole. A missing figure is never scored as zero, and it is never silently filled with the corpus average either. If a card discloses a purchase APR and an annual fee and nothing else, its score is built from the purchase APR and the annual fee, with their weights of 34% and 20% rescaled to 63% and 37%.
The consequence is worth stating plainly: a score built from two components is a thinner claim than a score built from six, even where the two scores are identical. Each card page names how many components its own score used.
The minimum-disclosure rule
A card that discloses fewer than 2 of the six components gets no score at all. Its page shows no stars, and says why. One figure with a star drawn beside it is not a rating of a card; it is that figure, dressed up.
Of the 783 cards on this site today, 187 carry a score and 596 do not. The second number is published here for the same reason as the first.
Cards priced over Prime are not ranked on APR
Many agreements, particularly from credit unions, price a card as "the Prime Rate plus 9.99%". That is a margin, not an APR, and the resulting rate depends on an index that moves. Those cards are never placed in the APR distribution — doing so would make some of the most expensive cards in the database look like the cheapest. They are scored on whatever else they disclose, and if that is fewer than 2 components they are not scored. Why a margin is not an APR.
What the distribution looks like
Scores currently run from 1.1 to 4.3, with a median of 2.9. Because every component is a position within the corpus, the median sits near the middle by construction: a score of 3.0 is an ordinary card on cost, not a bad one.
| Rating | Cards |
|---|---|
| 1 star (rounded) | 4 |
| 2 stars (rounded) | 48 |
| 3 stars (rounded) | 72 |
| 4 stars (rounded) | 63 |
| 5 stars (rounded) | 0 |
In the page markup
Where a card has a score, its page carries Review structured data with that rating and with CreditCardB — the organisation — as the author. It does not carry AggregateRating, on any page. An aggregate rating represents an average of ratings held by many parties, usually readers or customers. We have one rating, we computed it, and nobody has voted on anything. Publishing it as an aggregate would put a consensus that does not exist into a search result.
What would change a score
Two things, and neither is editorial. A figure changing in the issuer's disclosure or in a newer filing, and the corpus moving around the card as other cards are added or re-read. A card's score can therefore drift without the card changing, because the score is a position and the field moves. If a figure behind a score does not match the document linked on the card's page, tell us — that is a data error, and it is the one kind of rating complaint we can act on.