Evidence graph (v6)

What is proven, what is unresolved, surface by surface.

arkheionx evidence-graph classifies every important review surface into one evidence state, with an evidence strength, the missing-evidence gaps, and a next local test direction. It distinguishes "the file is referenced" from "the right edge is actually tested". An evidence state is not a vulnerability claim.

+

tested

A direct local test exists on a structurally simple surface.

+

needs-human-review

A test references the surface but coverage is shallow (often file-level) on a high-impact surface — the "looks tested" trap.

+

unresolved

A high-impact surface with weak, missing, or unclear local evidence. Not vulnerable — just unproven.

+

insufficient-evidence

Tests exist but do not prove the relevant assumption or counterfactual.

+

rejected-with-evidence

A local hypothesis log records this as tested and rejected with evidence.

+

confirmed-candidate

Local memory marks a candidate. A review candidate, not a confirmed vulnerability.

Run it

Classify the surface, see what is missing.

arkheionx evidence-graph .
arkheionx evidence-graph . --json
arkheionx evidence-graph . --out .arkheionx/evidence-graph
arkheionx evidence-graph . --only-unresolved

Boundary

An evidence state, not a verdict.

Evidence strength is a heuristic. A confirmed-candidate is not a confirmed vulnerability, and the two strongest states are only ever assigned from explicit local research memory. ArkheionX does not confirm vulnerabilities. Human review is required. See docs/EVIDENCE_GRAPH.md.

See the interaction matrix