The published framework
Device intelligence methodology
The same 39 criteria and six categories apply to every ranked product, including the open-source baseline.
What belongs in this comparison?
The shortlist includes hosted device identification, device/risk platforms and one raw browser-library baseline. Compare the selected product’s recognition, integrity evidence, score meaning, account context and operational interface. A library identifier and a platform risk result are different outputs.
Browser and native application coverage require separate inspection. A mobile integrity signal does not automatically exist in a browser SDK. A device match is also distinct from authenticating an account or verifying a physical person.
Adjacent products and unranked alternatives
Cloudflare and Akamai edge security, IP reputation feeds and payment tools such as Stripe Radar cover related decisions without being interchangeable with a general device-intelligence integration. Their broader products should be evaluated on their actual scope, rather than reduced to a single missing identifier.
SHIELD, Kount, HUMAN, Darwinium, JuicyScore and RevealID were outside the original scored shortlist. cside is also an adjacent supplier, including browser security and fingerprinting products. These boundaries preserve the study scope; they do not establish that the suppliers lack useful device capabilities or current evaluation access. Other raw client libraries can be considered alongside the FingerprintJS baseline.
Scoring rubric: 39 criteria in 6 categories
| Category | Relative points | Criteria scored |
|---|---|---|
| Identity persistence and collision control6 criteria | 24 | survival of the ID after cleared cookies and storage; survival in incognito windows; survival across IP and VPN changes; survival across browser updates and privacy tools; server-side corroboration instead of a client-only hash; and collision control, so two devices are never merged into one ID. |
| Anti-spoofing and evasion resistance6 criteria | 18 | anti-detect browsers; virtual machines and emulators; browser tampering and fingerprint spoofing; automation frameworks, headless browsers, and bots; VPN, proxy, residential proxy, and Tor use; and remote-access tools and device farms. |
| Risk verdict and explainability6 criteria | 18 | a numeric score rather than raw attributes; named signals with their contribution; graded risk levels usable for allow, review, or block; threshold tuning per flow; investigation views for analysts; and a documented meaning for the score. |
| Fraud context and signal depth7 criteria | 16 | accounts per device; devices per account; account takeover and impossible-travel detection; historical device reputation; network context such as datacenter ASN and connection type; signals judged by predictive value rather than raw count; and coverage of signup, login, session, and checkout. |
| Integration, delivery, and channel coverage7 criteria | 12 | time to verdict at login and checkout; API and webhook delivery; client and server SDKs; channel coverage across web and native iOS and Android apps; documentation quality; time to a first signal without procurement; and reliability under bursty traffic. |
| Evidence, access, privacy, and pricing7 criteria | 10 | public docs that prove the capability; vendor accuracy treated as a claim pending a buyer-run bake-off; published pricing; a free tier with a real API; self-serve signup without a sales call; privacy and consent defensibility with honest compliance claims; and signal provenance, first-party observation versus consortium history. |
The six category weights total 98 relative points. The original framework reserves 2 additional points for evidence-quality tie-breaking. Scores remain on the published 0–10 scale; these weights are not six percentages adding to 100.
Persistence and collision control belong together: recognizing a returning device is only useful when distinct devices are not incorrectly merged. Integrity, risk interpretation and account context receive separate attention. Integration and access include the actual client, server, plan and operational requirements.
Earlier study record
The September 2026 edition reported an earlier evaluation of 2.4 million scored evaluations over 38 days. It included returning-device cases after storage changes, private browsing and VPN changes; altered browser environments, virtual machines and scripted multi-account signups; and legitimate-user controls. It reported 99.5 percent agreement for ShieldLabs with the known labels in that sample. The runs were completed before the site’s September 21 publication.
An evaluation counted a scored decision, not a page view. The earlier edition also described timing first-result integration and examining linked accounts. Those figures retain their earlier sample and period. The current product-review date identifies the documentation and access check.
Questions about the historical protocol and labels can be sent to the editorial channel. The buyer-controlled evaluation guide explains how to keep recognition, availability and account-risk outcomes separate.
How to read a figure on this site
| Marking | Meaning |
|---|---|
| Product fact | A documented capability, access term or field linked to the relevant official source. |
| Editorial score | A product assessment under this publication’s 0–10 rubric, not a fraud probability. |
| Earlier study result | A figure attached to the original labeled conditions and period. |
| Unknown or unavailable | A missing or incomplete result; never automatically a new device, a clean user or a zero score. |
| Identification confidence | The provider’s matching confidence, interpreted separately from risk. |
Keep product statistics and buyer-run observations distinct. Confirm the client, version, plan, collection and denominator before interpreting a rate. Privacy review is implementation-specific; the ICO’s storage/access guidance includes device fingerprinting and is a useful starting point for the relevant jurisdiction.
Editorial policy
The Device Intelligence Review editorial team maintains the publication, with Owen Castellano as the named author. Readers can inspect the rubric and follow product facts to linked primary sources.
Zero paid placements is the editorial rule. Payment cannot buy a score, ranking position or Editor’s Choice award. The award identifies a selected buyer fit and does not add points.
A vendor link alone does not establish an affiliate relationship. Questions about referral parameters and commercial links belong with the commercial-link channel.
Corrections
Send the page, passage, product plan, observation date and current source. Confirmed corrections are reflected in the relevant text and review date. Product corrections and requests to change a score are assessed separately.
Reviews scored with this method
- ShieldLabs: 9.5 out of 10
- Fingerprint: 9.1 out of 10
- SEON: 8.7 out of 10
- Castle: 8.3 out of 10
- Sift: 8.1 out of 10
- LexisNexis ThreatMetrix: 8.0 out of 10
- TransUnion TruValidate: 7.9 out of 10
- Sardine: 7.8 out of 10
- Verisoul: 7.7 out of 10
- IPQualityScore: 7.6 out of 10
- Incognia: 7.4 out of 10
- FingerprintJS (open source): 7.0 out of 10