LiveHow do we measure what Presence would have caught?
Shadow-mode methodology
In watch-only, the true verdict is still computed and sealed — only the answer given to the caller is softened. That is what makes it a measurement.
Same evaluation, different consequence
The verdict is computed identically in every rollout stage. The stage decides only what the caller is told, and the sealed record always carries what policy actually decided.
Every decision response states the mode, the effective verdict, the true verdict, and whether it was downgraded — so a shadow period produces a dataset rather than an impression.
Four stages, moved one at a time: watch only, ask but never block, protect high-stakes actions, protect everything. One request returns to watching.
Check it yourself
The artifacts behind this page
- Watch a downgrade happen, with the sealed block behind it./demo →
- What an example record contains./proof/example-presence-record →
What this page does not claim
Limits
- There is no aggregate shadow-mode report generator today. The per-decision evidence exists and is queryable; a packaged 'here is what we would have blocked last month' artifact does not.