Reviewing Mori results

Mori produces a shortlist for source review. A high score means two normalized feature multisets overlap strongly; it does not establish equivalent behavior.

Keep a complete report while showing a bounded review summary:

mori scan --profile review --format agent --output /private/report.json .

Choose a private report path outside tracked source. Inspect retained JSON identities and source ranges without repeating analysis. The agent summary shows at most 25 groups; an unreviewed remainder is still unreviewed, even when the command succeeds. Repetitive initializers and UI shapes can be intentional; review their effects and ownership before suggesting consolidation.

If a report was not retained, copy a content-pair ID and rescan explicitly with the same scan settings:

mori explain aaaaaaaaaaaaaaaa:bbbbbbbbbbbbbbbb --profile review .

explain repeats the scan without the presentation limit, then renders only the selected identity and its structural evidence. It does not infer behavior or retrieve a result from a remote service.

Narrow the source policy

Mori honors nested .gitignore and .moriignore files. Add explicit exclusions only for reviewed project policy:

mori scan --exclude '**/*_test.go' --exclude '**/*.test.ts' .

Use --exclude-generated for conservatively recognized generated headers. Those files remain visible as excluded_generated in JSON coverage evidence. Keep exploratory exclusions separate from an inclusive staged gate: excluding a supported staged file does not count as analyzing it, and can fail focused coverage. A source root may still include colocated tests; review paths and both function bodies before narrowing further.

Use review ranking

mori scan --ranking review .

Review ranking prioritizes disclosed source-location signals before ordinary structural ordering. It does not change scores, fingerprints, or eligibility.

In Mori v0.34.0 or later, repeated small call wrappers lose the name/repetition boost when the same name occurs in at least four directories. The signal is repeated-small-wrapper(-7).

Repeated short accessors, forwarding calls, throwing wrappers and single-assignment setters can also receive repeated-small-boilerplate(-N). This requires at least three distinct source occurrences across two files. Both representative bodies must have at most 80 normalized tokens, no nested functions, and only a small straight-line operation shape. Arithmetic, branches, loops and bindings keep normal priority. The deduction is at most seven points, never makes the base priority negative, and does not stack with the existing wrapper deduction. Configured path priorities are added afterward.

These are conservative presentation heuristics, not evidence that duplication is intentional. They help prevent repeated scaffolding from crowding the shortlist when lowering --min-tokens, but do not guarantee that every useful finding is retained. Use structural ranking, a larger --max-groups, or explicit priority paths to inspect deprioritized groups. No findings are automatically accepted or removed, and scores and identities are unchanged.

Agent and compact reports mark groups whose retained occurrences exclude nested bodies with outer body only: nested functions evaluated separately. Inspect those separate functions before interpreting a high parent score. This warning uses existing JSON occurrence metadata and does not change the report schema.

Projects can add deterministic presentation-only path priority:

mori scan --profile review --priority-path '**/auth/**=25' .

A priority path says where reviewers want to look first. It is not evidence that a finding is risky or actionable.

Focus change review without narrowing comparison coverage

mori scan --changed-since origin/main --threshold 0.85 .

The revision must already exist locally. Mori compares changed and unchanged source together, then moves groups touching changed files forward. It includes staged, unstaged, and untracked non-ignored files and never fetches a remote.

Use mori review staged check for a commit decision. This canonical command reads the repository index only and enforces focused inclusion and coverage. Inspect path_evidence and the separate policy/analysis outcome before calling the review complete. Review policy describes explicitly opting into advisory enforcement; existing gates stay strict by default.

Use repeated --focus-path for explicit paths. For nested worktrees, give each root its own locally available revision with repeated --changed-worktree PATH=REVISION; one parent revision cannot describe several Git histories safely.

Investigate partial duplication deliberately

mori scan --statement-blocks --block-statements 3 .

Statement blocks are fixed-size windows inside functions. They are a separate fragment kind, never compare with whole functions, and exclude overlapping same-file candidates. This mode is off by default because it increases candidate counts and can surface ordinary local symmetry.

Read the evidence

For every relevant group:

1. Open both retained source ranges.
2. Compare identifiers and literal values in context.
3. Inspect types, control flow, data flow, side effects, callers, error paths, transactions, permissions, schemas, and tests.
4. Classify it as likely duplication, intentional structural similarity, or a false positive.
5. Refactor only when source and behavioral evidence justify it.

100% means normalized feature identity only. Nested functions are independent comparison units and are excluded from their parent function's score.

Use JSON for the complete deterministic evidence contract and SARIF for editor or code-scanning consumers:

mori scan --format json .
mori scan --format sarif .

See How Mori scores fragments and Scan selection for the detailed contracts.