Report Trust Index

How much should I trust THIS report — before I believe any verdict in it?

AgentSmack ships five independent trust meta-lenses, each answering one orthogonal “can I trust this report?” sub-question: coverage (which production dimensions were never tested), validity (did the harness actually exercise the agent, or was the run vacuous), precision (does AgentSmack cry wolf or wrongly hard-block a benign run), corroboration (is each finding cross-surface confirmed or contested), and calibration (did the pre-flight forecast under-predict your real risk). The deployment gate folds only a subset into a binary go/no-go. Nothing produced a single, graded, explainable trust NUMBER over the full set — the meta-meta lens a CISO reads BEFORE believing any clean/dirty verdict. This is that artifact. It composes the five reports into one auditable 0–100 Report Trust Index with per-component status chips and a closed verdict. The killer property nothing else enforces: a disqualifying component is non-launderable — an overblocking scanner (FP.1), a vacuous run (TV.1), or a dangerously-optimistic forecast (FC.1) caps the index ≤39 / trust_compromised no matter how high every other lens scored, because a high blend cannot wash out a corrupted trust signal. A coverage blind-spot (TH.1 shallow/minimal) degrades to trust_qualified, never silently report_trustworthy. An absent lens is excluded (not_assessed) — never silently scored 0 or 100. Honest-empty: a run with zero lenses reports insufficient_trust_signal — trust is never fabricated. It is a PURE aggregator over already-computed reports (no re-scoring, no new surface, no DB). Fill a sample below to watch the index swing with no live infra.