The Standard Score™

Evidence without false precision

Every card review now exposes a structured evidence record under each of the six inputs. Those records are generated from the same typed fee, offer, earn-rate, currency, benefit, eligibility, and fit fields used elsewhere on the review, so a card-data correction updates its evidence too. The evidence makes the judgment auditable; it does not pretend the editorial rating was produced automatically.

Calibration and sensitivity

We test the official weights against 12 nearby models: each of the six weights is lowered and raised by 20% on its own, with the full set renormalized to 100%. Across the current roster, every nearby model retains at least 9 of the official top 10, and at least 99.88% of card pairs with an official ordering are not reversed. A pair that becomes tied is not called a reversal. That is a stability check, not a search for weights that preserve a preferred winner.

We also move one editorial component on one card by ±0.5 — the amount two reasonable reviewers might differ — while holding everything else fixed. The largest resulting change to a published score is currently 0.2 points. Because cards cluster tightly, that can still move a card several positions. Treat differences of roughly 0.2 as near-peer territory, not proof that one card is universally better.

Finally, the personalized engine is rerun against 10 reviewed personas spanning food-heavy and travel-heavy spend, cash-back patterns, no-fee and no-FX constraints, students, newcomers, premium lounge users, and small businesses. Tests fail if the engine breaks a hard constraint, leaves a reviewed set of sensible outcomes, or stops responding when two otherwise similar households spend in materially different places.

The principles behind the number

• Every ranking is explained. If we can’t articulate why a card outranks another in plain language, the scores are wrong.
• Points are valued at published rates. All dollar math uses The Points Standard Index, updated monthly. When valuations move, scores move.
• Perks count only at realistic usage. We never sum brochure values.
• Compensation never touches scoring. Referral relationships (where they exist) are disclosed, and a card’s score is set before any monetization decision. Cards that pay us nothing rank above cards that could.
• Every card record carries a source URL and a last-reviewed date. If a detail is stale, you can see exactly how stale.

When a category needs a different ruler

The Standard Score is the default ordering for broad rewards, travel, network, and issuer rankings. It is not allowed to overrule the defining job of a specialist card. Low-interest cards are ranked by annual fee plus purchase interest on a disclosed $3,000 average balance; secured cards are ranked by published guaranteed-approval terms first, then minimum deposit, fee, and Standard Score. Those pages show their category metric beside the Standard Score so the reason for the order is visible.

General rankings use the one-decimal score readers see. Cards with the same published score are co-ranked, with card name used only to make their display order deterministic. The overall ranking is the actual top ten across every personal card in the database, not a hand-maintained shortlist. Low-interest cards with the same carrying cost are co-ranked too; the Standard Score only stabilizes their order on the page.

What the grades mean

• 8.5+ — Benchmark. Category-defining. The card others are measured against.
• 7.5–8.4 — Excellent. A top recommendation with minor caveats.
• 6.5–7.4 — Strong. Very good for the right person; check the fit notes.
• 5.5–6.4 — Situational. Works in specific setups; most people have better options.
• Below 5.5 — Pass. We’d skip it.

How the first-year rating is anchored

The first-year value component is a judgment call, but it isn’t freehand: every card starts from a fixed anchor ladder that maps its first-year net value (welcome tranches issued before the first anniversary at our published valuations, plus realistic first-year earning, minus the annual fee) to a baseline rating. Anniversary, renewal, and month-13-or-later tranches are excluded even when their qualifying spend occurs during the first twelve months, because collecting them requires carrying the account through renewal. This is the ladder our corrections log refers to when we re-score cards for consistency.

From that baseline we apply a documented adjustment of up to −1.0 where the headline number overstates reality — a bonus paid in a weak or locked currency (TD Rewards at 0.5¢ is the clearest case), or a large tranche gated behind spending most applicants won’t reach. Adjustments are judgment, but they’re recorded: when anchor-ladder re-scores change a published score, the old and new numbers appear in the corrections log.

Data provenance, live

These numbers are computed from our datasets at publish time, not written by hand — if the data ages, this section says so:

• 98 published card reviews, every one verified against the issuer’s own page. Oldest last-reviewed date currently on the site: 2026-07-12.
• Point valuations marked as of 2026-08-01, re-derived monthly from observable award pricing — each mark’s basis is published on the Index.
• 117 welcome-offer records in our archive (tracked since 2026-07-11), so offer claims are checkable against dated history, not memory. Material changes appear in the change log.

Limits and honesty

Component ratings involve editorial judgment — two reasonable people applying the same rubric could differ by half a point. What we guarantee is that the judgment is consistent across cards, the weights are fixed and public, the assumptions are published, and no commercial consideration ever reaches the scoring step. Rankings do not use hidden hundredths to break a tie readers cannot see.