Detection Concepts

A test’s identity is its body

Two tests that do the same thing under different titles are duplicates. Two textually identical bodies that run under different hooks, mocks, or fixtures are not.

To make that comparison, bodies are normalized before they are compared: titles are ignored, and the shape of what a test does is what counts. The context a body runs in — hooks, mocks, fixtures — participates in the comparison, which is why identical text under different circumstances is left alone.

The rules

Rule	Default	Meaning
duplicate-test	Error	Two tests have the same normalized body under an equivalent context.
duplicate-test-name	Warning	Two tests in one file share a full title.
parameterization-candidate	Warning	Two tests have the same structure and differ only in literal values.
near-duplicate-test	Warning	Two tests perform the same behavior under an equivalent context, but their normalized bodies differ.
cross-file-duplicate	Warning	The duplicate-test criteria without the single-file restriction.
subsumed-test	Off	One test’s behavior is a strict ordered subsequence of another’s. Off by default: its findings are frequently distinct tests rather than redundant ones.
duplicate-suite	Warning	A whole describe block is duplicated within one file, reported once instead of per test.
duplicate-each-row	Warning	Two rows of one parameterized test, benchmark, or suite carry the same case.
redundant-hook	Warning	A nested lifecycle hook repeats work an ancestor hook already performs.

A rule configured at severity off is not computed, so a run never reports a finding for it.

Verdicts

Each finding carries a verdict naming its remedy, because the remedies differ:

• Redundant — one copy can go. duplicate-test, cross-file-duplicate, duplicate-suite, duplicate-each-row, redundant-hook.
• Consolidatable — every case is needed, and they belong in one test.each table. parameterization-candidate. Deleting a row here would lose coverage.
• A review prompt — near-duplicate-test and subsumed-test flag tests that look equivalent without establishing that either adds no coverage. Review decides; for near-duplicate-test the usual outcome is consolidation into one test.each table.
• Name collision — two tests share a title; rename one.

cross-file-duplicate reports at Warning rather than Error because duplication across unit and integration layers is sometimes deliberate.

What comparison cannot see

Some tests cannot be compared: an unreadable config, a dynamic title, a table the analyzer cannot resolve statically. The run is then marked incomplete and says why, because the absence of a finding is no longer evidence of its absence. --fail-on-incomplete turns that into a gate.

Imported helpers are compared by their resolved body when the import resolves to a readable ES-module file with a plain parameter list; unresolved imports, defaulted or rest parameters, and helpers that call helpers fall back to comparing by name. Dynamic test generation that no static reader can follow is reported as unresolved rather than guessed at.