How to choose the level of impact and confidence? Request for definitions of High/Medium/Low criteria

[KaifeinDev]

The Slither documentation and detectors use impact (High/Medium/Low/Informational) and confidence (High/Medium/Low) to classify findings. However, I couldn’t find formal definitions for what qualifies each level.

[adilburaksen]

Here's how Slither defines the levels in practice:

Impact

• High — direct fund loss, complete access control bypass, or contract takeover
• Medium — unexpected state changes, partial fund loss, temporary DoS, privilege escalation in non-critical paths
• Low — best practice violations, minor logic issues with limited blast radius
• Informational — code quality, gas optimizations, style issues with no security consequence

Confidence

• High — very low false positive rate; the pattern almost always indicates a real issue
• Medium — some false positives expected; manual review needed to confirm
• Low — the pattern matches but is often benign; treat as a lead, not a finding

Slither doesn't combine them into a single score internally — it just surfaces both so you can triage. In practice, High/High findings go straight to the top, Informational/Low can usually be skimmed or suppressed.

For writing custom detectors, the detector API docs have the canonical definitions with examples.