Concepts

Sextant has a small, deliberately flat data model. Everything you’ll encounter in JSON output, MCP responses, and PR comments comes from these six types:

Concept	What it is
Rule	A single check, identified by a dotted id (e.g. builtin.size.fn-length).
Finding	One match of a rule against a file, at a specific line range.
Report	The output of a grade — findings, severity counts, verdict, summary.
Verdict	approve or request_changes, derived from configured thresholds.
Evaluator	The kind of check a rule performs: builtin, regex, or llm.
Scope	The slice of code a grade covers: diff, file, or repo.

The types live in the sextant-core crate and have no I/O dependencies — the same definitions back the CLI’s JSON, the MCP server’s tool results, and the GitHub Action’s review comment.

Recommended reading order

If you’ve never used Sextant before, read in this order:

1. Rule — what gets checked.
2. Evaluator — how it gets checked.
3. Finding — what comes out.
4. Report — how findings are bundled.
5. Verdict — how a report becomes approve / request_changes.
6. Scope — where in the codebase a grade looks.

After that, the CLI reference and MCP tool reference read like API docs.