Severity & Confidence

Every Chainvet finding carries two independent ratings. Keeping them separate is what makes triage precise.

Severity — potential impact

high · medium · low. Severity is the potential impact of the issue if exploited. It is assigned per detector when a finding is constructed — for example, a reentrant ETH transfer or an arbitrary selfdestruct is high, while a block.timestamp-based decision is low.

Confidence — the detector's precision

high · medium · low. Confidence is the reporting detector's own estimate of how likely the finding is a true positive — not how bad it is.

• High — high-precision detection; false positives are rare.
• Medium — plausible but pattern/heuristic-based; warrants manual triage.
• Low — speculative or false-positive-prone.

How it's resolved: a detector uses a per-finding override when it has local evidence (a guard it saw, a confirmed taint path, an ETH-sending sink); otherwise it falls back to a per-kind default. Roughly:

• Near-syntactic checks (e.g. tx.origin, default visibility, unprotected selfdestruct, weak PRNG) default to high confidence.
• Dataflow/heuristic checks (several reentrancy variants, shadowing, tainted call, missing input validation) default to low.
• The rest default to medium.

Across engines the meaning is consistent: static confidence is heuristic, symbolic execution derives it from solver evidence, and fuzzing from concrete reproduction — all on the same scale so the orchestrator can rank and merge findings.

Triage order

Filtering by each axis

Both axes are independently filterable in the CLI — by a floor (at or above) or by exact values:

chainvet scan -s high contracts/          # severity >= high
chainvet scan -c high contracts/          # confidence >= high (most precise only)
chainvet scan --severity medium contracts/    # exactly medium severity
chainvet scan --confidence high contracts/    # exactly high confidence

See Filtering findings for the full rules, and the CI gating page for how both axes combine to fail a build.