Methodology
Methodology v2.1, October 2026
This page is the whole method. If a number on the site cannot be traced back to a rule here, that is a bug and we would like to hear about it.
What the research is
Desk research. For each brand the desk reads the brand's own site (pricing, shipping, returns, warranty, support and privacy pages), then the public record of customer reviews on Trustpilot, the app stores and Reddit, and any regulator or press coverage that bears on the policies. Nothing is purchased, nothing is tested, nobody from the brand is interviewed. The scorecard reflects what the brand promises in writing and how customers say those promises hold up.
Review summaries avoid invented precision. When we have not counted, we say "a large minority" or "a recurring complaint" rather than quoting a percentage.
The five criteria
| Product quality | Does the product do what the page says, hold up over time, and match its price tier? Judged from review themes and the brand's own specifications. |
| Value | Price against the nearest credible rivals, including the real cost after promotions end, add-ons, shipping and renewal pricing. |
| Customer service | Channels offered, hours, whether a human is reachable, and how reviewers describe response times and outcomes. |
| Returns & policies | Return window, refund rules, restocking fees, warranty length and exclusions, cancellation friction. Scored on the written policy, then adjusted for how customers report it working. |
| Transparency | Are prices, fees, renewal terms and data practices stated plainly before purchase? Are reviews, audits and ownership easy to find? |
Each criterion is scored from 1.0 to 9.9 in steps of 0.1. A sub-score of 10.0 is not available; no brand is perfect and the scale should not pretend otherwise.
Weights
| Criterion | Weight |
|---|---|
| Product quality | 0.30 |
| Value | 0.25 |
| Customer service | 0.15 |
| Returns & policies | 0.15 |
| Transparency | 0.15 |
Weights sum to 1.00. They apply to every category. We considered category-specific weights and rejected them: a reader comparing a chair brand to a VPN should be looking at the same yardstick.
How Overall is derived
Overall is computed by the site generator, not typed in by a researcher. Three steps:
- Weighted mean. Multiply each sub-score by its weight and add them up.
- Offset. Add a fixed offset between −0.2 and +0.2 derived from a hash of the brand's URL slug. The offset is the same every time the site is built, so the figure is stable. Its purpose is to stop near-identical brands lining up in a neat 8.3, 8.2, 8.1 staircase that would imply a precision the method does not have. It never moves a brand by more than one grade band and is disclosed on every scorecard.
- Round and cap. Round to one decimal. Cap at 9.8; the site never displays 10.0 or 9.9.
Worked example. Sub-scores 8.6, 7.8, 7.1, 6.9, 8.4 give a weighted mean of 7.90. If the slug offset is +0.1 the displayed Overall is 8.0 and the grade is B+. If the offset is −0.1 it is 7.8 and the grade is B.
Grade bands
| Grade | Overall |
|---|---|
| A | 9.0 and above |
| A− | 8.5 to 8.9 |
| B+ | 8.0 to 8.4 |
| B | 7.5 to 7.9 |
| B− | 7.0 to 7.4 |
| C+ | 6.5 to 6.9 |
| C | 5.5 to 6.4 |
| D | below 5.5 |
The verdict chip
BUY, CONSIDER or SKIP is a judgement call by the desk, not a formula. It answers one question: would we send a reader to this brand's checkout today? The chip is constrained by the Overall so that the judgement and the arithmetic cannot contradict each other:
| Chip | Allowed Overall range |
|---|---|
| BUY | 7.5 to 9.8 |
| CONSIDER | 6.0 to 8.4 |
| SKIP | 1.0 to 6.4 |
Where the ranges overlap, the chip carries the context the numbers cannot: a B brand with a live data-handling problem gets CONSIDER, a B brand with a boring but reliable record gets BUY. Per-product chips inside an article follow the same logic for that item alone and may disagree with the brand chip.
Updates and dating
Every verdict shows the month it was last re-read. A brand is re-checked when a policy changes, when a reader reports an error, or at most twelve months after the previous read. Changes to grades and chips are listed on the changelog. We do not quietly edit old scores.
What we do not do
- No star ratings and no Review structured data. The grade is an editorial opinion and is marked up as such.
- No paid placements. Position on the leaderboard is sort order, nothing else.
- No fabricated counts of reviews, customers or incidents.
- No claims of hands-on testing.
Version history
- v2.1
Transparency weight raised from 0.10 to 0.15; Value lowered from 0.30 to 0.25. Grade bands unchanged. Verdict-chip consistency rule added.
- v2.0
Moved from a 100-point total to five sub-scores out of 10 with published weights. Introduced the letter grade and the BUY / CONSIDER / SKIP chip.
- v1.0
First public version. Single score out of 100, no sub-scores.