Glossary: the memoryfixtures vocabulary

Every term the evaluator uses, in plain language — the failure-mode categories, the fixture checks, the tri-state verdicts, and the fixture kinds.

Failure-mode categories

forgetting
A stale or lost fact is returned in place of the current one — the system "forgot" an update and surfaced an outdated value.
chronology
Facts are returned in the wrong time order. Even when every item is individually correct, a mis-ordered history makes downstream reasoning unsafe.
entity-confusion
A fact is attributed to the wrong entity — the right information is pinned to the wrong person, account, or object.

Checks

answerEquals
Asserts the system-under-test’s answer exactly equals an expected value (case- and whitespace-insensitive).
mustContain
Asserts every listed value appears in the answer — used to confirm a current or retained fact was surfaced.
mustNotContain
Asserts no listed value appears in the answer — used to confirm a known-stale value did not leak back in.
order
Asserts the actual sequence deep-equals the expected sequence — the chronology check. Any reordering fails.
attributedTo
Asserts the answer’s entity matches the expected entity (case- and whitespace-insensitive) — the entity-confusion check.

Verdicts

pass
At least one check passed and none were failed or unknown.
fail
At least one check failed. Fail takes precedence over every other verdict.
unknown
No check failed, but the evidence needed to assign a pass was missing (for example, no answer was supplied). A first-class verdict: missing evidence never fabricates a pass or a fail.

Fixture kinds

positive
A fixture whose correct system behavior should score a pass.
negative
A fixture that encodes a real regression — the system behaving wrongly should score a fail.
malformed
A valid fixture-run that exercises a malformed memory scenario (distinct from an invalid input envelope, which is a contract error).
missing-data
A fixture where required evidence is absent, exercising the unknown verdict path.
adversarial
A fixture built to trip a naive scorer — look-alike values, partial matches, or injection-shaped content — that the evaluator must still score honestly.

See also

The methodology page walks through how these checks combine into a verdict. Contract version 1, evaluator 1.0.0.