A reproducible measurement of the terms you agree to — arbitration, data use, content licensing, and how often the rules change — read directly from the public archive. Every figure below counts distinct platforms, shows its confidence where coverage is thin, and can be reproduced from the same free evidence. No estimates.
352 platforms tracked · counts are distinct entities, not provisions · how it's measured →
Sharper, where we've normalized the underlying stance (and only counted platforms whose stance is actually stated — never inferred from silence): of the 150 platforms with a normalized arbitration stance, 114 impose forced arbitration and 133 waive your right to a class action; of the 52 with a normalized data-use stance, 39 train AI models on your content.
Distinct platforms carrying at least one active provision of each type, out of 352 tracked.
History starts 2026-02. The current month is in progress (shown hatched).
Ranked among the 43 platforms scored on 5–6 of 6 governance dimensions — enough coverage to rank fairly. Each row shows its coverage. A further 148 platforms are scored on only 3–4 dimensions (provisional) and are deliberately not ranked here, so a thin read never outranks a complete one.
Across all 191 scored platforms: average 31.4/100 · median 30.8 · quartiles 18.4 / 42.3. Higher = more protective of you, the user.
Every number on this page is counted from the same public archive you can browse for free — no proprietary panel, no survey, no estimate. Prevalence and headline figures count distinct platforms (a platform with ten arbitration provisions counts once). The User-Protection Index is deterministic: the same cited governance stances always produce the same score, and no dimension we haven't confidently read is ever counted. Where coverage is thin, we say so and hold it back rather than guess.