Understanding Results

Every finding carries the same structure across all report formats:

• Rule and severity — which of the eight rules fired, and whether it counts as an error or a warning. Exact collisions default to errors; heuristic findings default to warnings.
• Primary location — the path:line:column span where the finding is anchored.
• Related locations — the other definitions (or feature steps) involved in a pair or cluster finding.
• Evidence — why the finding exists: matcher differences, similarity scores, handler behavior, or usage facts.
• Suggested action — the remediation direction, such as keeping one definition, making matchers mutually exclusive, or replacing literal differences with a step parameter.
• Semantic fingerprint — a location-independent identity used by baselines and suppression accounting.

Severity and exit codes

Findings alone do not fail a run. Exit code 1 is produced by the duplication threshold, an active error-level rule, or the new-finding allowance. Warnings alone never produce exit code 1. Operational problems — discovery, parsing, or configuration failures — exit with 2. See CI and baselines.

Reading the duplication rate

The summary line reports how many definitions participate in active, error-level duplication findings, as a percentage of every discovered definition:

Duplication: 2 of 11 definitions (18.18%), threshold 5.00% — FAIL

Suppressed and baselined findings do not count toward the numerator, and a definition counts once even when it participates in several findings. ambiguous-step, overlapping-matcher, and unused-definition errors are independent policies and remain fatal regardless of the threshold.

Incomplete analysis

CukeDedup reports incomplete analysis instead of treating missing evidence as clean. Machine reports carry corpus completeness signals and truncation counts, and the HTML report shows an incomplete-run alert. When findings depend on a complete corpus, existing findings remain valid — but the absence of a finding proves nothing while the run is incomplete.

Use --fail-on-incomplete when CI must reject degraded analysis, and --explain-discovery to inspect what was actually discovered.

Suppressing accepted findings

Two suppression mechanisms exist, both requiring an explicit reason:

• Configured suppressions in .cuke-dedup.json — select by rule plus path and/or matcher. For pair findings, a path must contain every involved definition, so a directory exception cannot hide a conflict crossing into maintained code.
• Inline directives above a registration — // cuke-dedup:ignore RULE -- reason (JS/TS) or # cuke-dedup:ignore RULE -- reason (Ruby). The directive applies only to findings involving its attached definition.

Suppressed findings are listed in reports with their suppression state and do not count toward the duplication threshold. See Configuration.

Baselines and changed-file analysis

For established suites, baselines suppress findings that already exist in a committed semantic fingerprint file, and --changed-since reports only findings involving files changed since a Git revision — while still comparing those files against the complete corpus.