Diagnostic records

Two opt-in flags add records to the jsonl stream. Both require --reporters jsonl.

--emit-test-fingerprints

Adds one record per test and per hook: the comparison keys themselves, for debugging why two tests did or did not match.

{"type":"test","schemaVersion":"0.1","path":"cart.test.ts","line":13,"column":1,"title":"totals a two-line cart","alpha":"24ca2dec36cb546b","structural":"61d3c3582d21bc51","behavior":"6ace5d410965ead9","trivial":false,"comparable":true,"contextKey":"9b9573371b170f09","modifierKey":"09612b07b5ecb5a5","truncatedFields":[]}

Comparable tests from distinct declarations registered in one file are a duplicate-test when alpha, contextKey, and modifierKey all match. If a pair you expected to match did not, diff those three to see which one disagrees.

"test" records

Field	Meaning
registrationFile	Optional caller path when different from the declaration’s path; otherwise path also identifies the registering file.
alpha	Identifiers normalized to their resolved symbol; literal values preserved, except inert literal-data properties that a const object destructure does not select from its own object-literal initializer.
structural	alpha with literal values collapsed to their kind.
behavior	The ordered behavior events: callee paths, assertions, awaits, control flow.
trivial	The body is empty or a recognized stub, and is excluded from comparison whatever the hashes say.
comparable	The context resolved well enough to participate in comparison. false means the test was not compared, not that it was verified distinct.
contextKey	The ambient facts that gate comparison: hook chain, vi.mock registrations, test.extend fixtures, imports, module-scope setup, per-directory config scope.
modifierKey	Modifiers that change what the test does (.each data, .skipIf/.runIf, .fails), never ones that only change whether or when it runs.

"hook" records

Hook records carry kind, alpha, structural, behavior, and trivial. redundant-hook only compares hooks of matching kind.

--debug-memory

Adds one run-level memory record: the bytes retained after extraction, split by the field holding them, with the number of tests and bytesPerTest, which is their average. The buckets are disjoint and sum to the reported total, so a consumer can both trust the total and attribute it.

Shared text is apportioned rather than multiplied: a body referenced by three tests is counted once and divided among them, so the per-test figure is an average rather than an overcount.

peakRssBytes is a process high-water mark for sanity only. It includes the allocator’s slack and the parser’s transient arenas, neither of which is retained.

Back to Reports.