Evidence - AwareXone Docs

The Evidence & Confidence engine builds a deduplicated evidence graph per finding: supporting evidence, counter-evidence, reuse, conflicts, and an explainable confidence level.

CLI

axguard evidence .
# .findings/axguard/evidence.json
# .findings/axguard/evidence.md

Also soft-run during axguard audit. Diagnostic exit 0.

MCP

Tool	Use
`axguard_get_evidence`	Evidence for a finding
`axguard_get_evidence_chain`	Ordered chain supporting a conclusion
`axguard_get_counter_evidence`	Why a candidate may be FP / blocked

Confidence

Quality is scored from provenance, exact location, and analysis strength — not an arbitrary LLM percentage. Unknowns pull confidence down; a single strong signal never hides a critical unknown. Typical levels: VERY_HIGH · HIGH · MEDIUM · LOW · UNKNOWN.

Discipline

Repository comments are not evidence of safety
Name-only controls do not raise confidence
Agents must never invent evidence from model speculation

Slash command: /axguard-evidence. See also Judge and False Positive Adversary.