A consistent, public, evidence-based way of scoring platforms across twelve safety dimensions. Designed so that two independent reviewers reach the same score for the same evidence.
Every platform assessment follows the same five-step method. Intermediate outputs are published so the public scoring can be challenged and refined.
Gather documented behaviour: policies, terms of service, transparency reports, victim testimony, regulator notices.
Cross-check each item against at least one second source. Reject single-source claims.
Place each verified item under one or more of the criteria for the relevant risk area.
Apply the published rubric. Two reviewers score independently; differences reconciled by a third.
Publish the score with the underlying evidence list. Notify the platform and publish their reply in full.
Scoring is on a five-band scale per risk area. Bands are defined behaviourally, not numerically.
Practice is ahead of regulator expectations and ahead of peers. Independent audit confirms.
The platform consistently meets the published criteria; gaps are minor and time-bound.
Materially short of criteria in some areas, with credible plans and visible progress.
Fails criteria in ways that are documented and have not been addressed despite reasonable notice.
Behaviour repeatedly and demonstrably harms users; remediation is absent or systematically blocked.
Anyone can submit evidence that may revise a published score. Submissions are confidential, encrypted and reviewed by the panel.