Finding

A finding is one rule firing against one piece of code. A grade returns a list of findings; the verdict is computed from their severities.

Shape

{
  "rule_id": "builtin.size.fn-length",
  "severity": "warn",
  "message": "Function `dispatch` is 78 lines (warn at 60)",
  "path": "src/handlers.rs",
  "line": 88,
  "end_line": 165
}

Field	Type	Notes
rule_id	string	The id of the rule that fired. Pass to explain_rule for full context.
severity	info | warn | error	See Severity below.
message	string	Human-facing description of the violation, often with a fix hint.
path	string	File path relative to the repo root.
line	number?	First line of the violation. Optional — file-scope findings may omit it.
end_line	number?	Last line of the violation, inclusive. Optional.

Severity

Severities are ordered: info < warn < error.

Severity	Meaning
info	Informational signal. Never blocks a verdict. Use for “good to know” findings.
warn	Advisory. Counts toward [verdict] max_warns. Default thresholds let warns through.
error	Blocks the default verdict (max_errors = 0). Use sparingly — for rules you’d block a PR over.

The default [verdict] thresholds are max_errors = 0 and max_warns = u32::MAX. The agent surfaces all of them; CI gates only on errors. Tune to taste in .sextant/config.toml.

Sort order

Findings within a report are deterministically sorted by:

1. Severity, descending — errors first.
2. Path, ascending.
3. Line, ascending.

That ordering is preserved across all output formats (JSON, markdown, SARIF) so diffs between reports are stable.

In diff and PR mode

Diff-mode grades attach a finding only when its line range overlaps with the diff. PR-mode grades go further: they compare findings to a baseline-graded base SHA and report just the new ones — see Scopes and Verdict → regression mode.

See also

• Report — the bundle of findings.
• Verdict — how findings translate to approve/request_changes.