Skip to main content
The tenant verdict API returns projected findings in signals. Each finding has this shape:
Use id, tier, and family as structured fields. Treat headline and detail as prose. Read the tier from each finding instead of assigning one from the ID. Four IDs can use different tiers under different conditions. The current catalogue has 44 finding IDs across six evidence families. The catalogue also has three concealment-only IDs that never change the verdict.

Network geometry

Geometry derives a conservative lower distance bound to the claimed timezone from an enclosing radius and a safety-margined bounding box. It does not treat the reference point as the exact client location.

Transport stack

Transport reachability

Any parsed QUIC attempt suppresses reach.no-quic-attempt, including shape-only evidence. A same-address match does not authorize rules that require a confirmed join.

TLS construction

The TCP construction rules do not apply their GREASE, h2, or TLS-version expectations to a QUIC-only ClientHello.

HTTP construction

The two absolute refusal IDs are identity.measurement-declined and identity.measurement-withheld. A non-refusal absolute finding still takes precedence when both kinds are present.

Compute

gpu.below-claimed-model is a conditional population comparison. It is not a within-session invariant and cannot make a verdict conclusive.

Concealment-only IDs

These IDs appear in the full concealment report, not as verdict-bearing findings merely because the indicator fired. The concealment report can also repeat selected finding IDs as path context. Concealment indicators and network classification do not change the verdict.

Tier effect

Several findings in one family still count as one family. See Verdicts and coverage for precedence and family status.