Report

A report is what every grade returns. It bundles findings with top-level metadata so a caller (agent, CLI, Action) can render a verdict without recomputing anything.

Shape

{
  "summary": "12 findings (0 errors, 7 warns, 5 info). Verdict: approve.",
  "verdict": { "kind": "approve" },
  "counts": { "info": 5, "warn": 7, "error": 0 },
  "findings": [
    {
      "rule_id": "builtin.size.file-length",
      "severity": "warn",
      "path": "src/parser.rs",
      "line": 412,
      "message": "File length 412 exceeds warn threshold (400)"
    },
    …
  ]
}

Field	Type	Notes
summary	string	One-line synopsis suitable for an agent or a chat reply. Read this first.
verdict	object	{ "kind": "approve" } or { "kind": "request_changes", "reasons": [...] }. See Verdict.
counts	object	{ info, warn, error } totals.
findings	array	All findings, sorted by (severity desc, path, line). See Finding.

PrReport (PR mode)

sextant grade --pr returns a richer wrapper around two reports:

{
  "head": { /* full Report for the PR head */ },
  "baseline": { /* full Report for the base SHA */ },
  "delta": {
    "new": [ /* findings only in head */ ],
    "fixed": [ /* findings only in baseline */ ],
    "unchanged_count": 12,
    "new_counts": { "info": 0, "warn": 1, "error": 0 },
    "fixed_counts": { "info": 0, "warn": 2, "error": 0 }
  },
  "verdict": { "kind": "approve" },
  "summary": "1 new finding (1 warn). 2 findings fixed. Verdict: approve."
}

The verdict on a PrReport is computed against delta.new_counts, not against the head report’s totals — a PR that merely exposes an existing finding doesn’t block. See Verdict → regression mode.

Output formats

Format	What you get
human	Coloured terminal output for sextant grade (default).
json	The structured Report (or PrReport) above.
markdown	A PR-comment-friendly markdown rendering. PR mode only.
sarif	SARIF 2.1.0 for GitHub Code Scanning.
review-json	A Review payload ready to POST to the GitHub PR Reviews API. PR mode only.

The --report-json <PATH> CLI flag is a side-channel: it always writes the structured report to a file, even when --format is markdown or SARIF. The Action uses this to render a markdown review while still parsing the JSON for verdict / counts.

See also

• Finding — what’s inside findings.
• Verdict — how verdict is derived.
• Output formats — picking between them on the CLI.