A score you can
take apart.

Editorial metascores by provider.

Alphabetical order. A difference of a point or two is not proof that a service is better or safer for your task. The source range is disagreement among published opinions, not predicted performance.

Scroll horizontally for coverage, source ranges and context. Open a provider to inspect its calculation.

Alphabetical provider order; all values are provisional editorial consensus.

Provider	Editorial metascore	Coverage	Source range	Keep the context
CyberGhost	80 / 100Strong reviewsProvisional editorial consensus	3 ratings · 3 familiesBroader coverage	78–83Included publisher scores on the 0–100 scale	Inspect the source caveats. Privacy record
ExpressVPN	90 / 100Excellent reviewsProvisional editorial consensus	3 ratings · 2 familiesLimited coverage	82–94Included publisher scores on the 0–100 scale	Inspect the source caveats. Privacy record
hide.me	81 / 100Strong reviewsProvisional editorial consensus	3 ratings · 3 familiesBroader coverage	80–83Included publisher scores on the 0–100 scale	Inspect the source caveats. Privacy record
NordVPN	91 / 100Excellent reviewsProvisional editorial consensus	3 ratings · 2 familiesLimited coverage	89–92Included publisher scores on the 0–100 scale	Inspect the source caveats. Privacy record
PrivadoVPN	74 / 100Positive reviewsProvisional editorial consensus	2 ratings · 2 familiesLimited coverage	73–74Included publisher scores on the 0–100 scale	Inspect the source caveats. Privacy record
Private Internet Access	84 / 100Strong reviewsProvisional editorial consensus	3 ratings · 2 familiesLimited coverage	79–87Included publisher scores on the 0–100 scale	Inspect the source caveats. Privacy record
PrivateVPN	72 / 100Positive reviewsProvisional editorial consensus	2 ratings · 2 familiesLimited coverage	70–74Included publisher scores on the 0–100 scale	Inspect the source caveats. Privacy record
Proton VPN	88 / 100Strong reviewsProvisional editorial consensus	3 ratings · 2 familiesLimited coverage	84–91Included publisher scores on the 0–100 scale	Inspect the source caveats. Privacy record
PureVPN	72 / 100Positive reviewsProvisional editorial consensus	3 ratings · 3 familiesBroader coverage	65–80Included publisher scores on the 0–100 scale	Sources differ by at least 15 points. Privacy record
Surfshark	89 / 100Strong reviewsProvisional editorial consensus	3 ratings · 2 familiesLimited coverage	85–90Included publisher scores on the 0–100 scale	Inspect the source caveats. Privacy record

METHOD editorial-1

How the editorial metascore is calculated.

1. Choose the question.

   Summarize a provider’s published overall editorial verdicts. A reviewed plan, operating system or actual test window may be unknown; the result is a service-level opinion summary, not a score for today’s cheapest plan or comparable test results.

2. Normalize the declared scale.

   100 × (value − original worst) / (original best − original worst). A 4/5 rating on a declared 0–5 scale becomes 80/100. This linear conversion is an editorial convention; it does not make the publishers’ criteria identical.

3. Cap a family’s influence.

   Keep one eligible edition per publication; average normalized publication ratings within each known publisher family, then give every contributing family one equal share. Merge families if an explicit shared origin is recorded. TechRadar and Tom’s Guide share the Future family; their two ratings together have the same total weight as one contributing family.

4. Require enough eligible coverage.

   At least 2 contributing provisional families. Use one eligible edition per publication. A missing or invalid rating never becomes zero. With fewer families, publish “Insufficient editorial evidence,” not a number.

5. Use a declared edition window.

   The captured article publication/update date must be within 36 months of the fixed 2026-11-07 calculation snapshot. The cutoff is 2023-11-07. This is edition eligibility, not proof of recent testing; there is no age multiplier or recency bonus. Source captures after that snapshot are excluded until the policy date is deliberately advanced. Builds do not silently refresh the snapshot.

6. Round once; show disagreement.

   Round only the final mean to the nearest whole point, with half points rounded up. Retain unrounded family means and normalized inputs in the export. The displayed source range shows the lowest and highest included normalized publisher rating. A spread of 15 points or more is flagged. It is not a statistical confidence interval. Two contributing families get “Limited coverage”; three or more get “Broader coverage.” Neither label means verified safety or high statistical confidence.

What stays out?

• Category-only scores, including RTINGS Private Browsing
• Customer or app-store ratings
• Explicitly mixed free/premium verdicts and free-only verdicts
• Missing, invalid or out-of-bounds overall scales
• Undated, future-dated or older-than-36-month review editions
• Duplicate editions from one publication
• Held speed snapshots, policy statements, audits, country groups and court records
• Source captures dated after the fixed calculation snapshot, until the policy date is deliberately advanced

What the number cannot settle.

• Publishers use different criteria; scale conversion does not harmonize their methods.
• Exact tested plan, platform and test window can be unknown for included provider-level opinions. Those gaps still prevent treating them as comparable test evidence.
• Family grouping is provisional and does not certify independent measurements or exclude all undisclosed data reuse.
• Most included publishers use affiliate links or advertising. Top10VPN's terms also disclose possible commercial influence on comparison ordering and scores.
• Observed source range describes disagreement among included normalized publisher ratings, not a confidence interval, reliability probability or expected outcome for a visitor.
• Commission amounts, current VPN prices, ownership preferences and missing records do not change the weights or contribute points.
• Publisher overall verdicts can already reflect their own speed, privacy and value judgments. We do not add separate points for our speed snapshots, case files or current prices.

The stricter test-readiness checklist remains unchanged. Zero test-ready inputs means we cannot publish a matched-test performance/security synthesis; it does not invalidate a clearly labelled aggregation of overall editorial opinions.

Download all scores, original inputs, weights and exclusions · Download the versioned policy · Coverage and funding.