runHarness — the harness-scope entry of the revamped API (Phase 2 of
research/testing-api-design.md). The harness has two execution scopes, hook
(runHook) and harness (the whole assembled agent); today's integration /
e2e / eval are all the harness scope under realness flags. This entry is
the deterministic harness run (model: "mock", the default) — the
workhorse you gate every commit, with no key. A real-model harness run is
non-deterministic by definition, so you don't assert a single one — you
measure() it across trials (the eval scope). egress is a capability of this
scope (the e2e tier), not a separate tier.
Behaviour is identical to runHarnessTest (which it wraps); the new name +
model flag make the scope/realness explicit and steer real-model runs to the
right tool.
runHarness— the harness-scope entry of the revamped API (Phase 2 ofresearch/testing-api-design.md). The harness has two execution scopes,hook(runHook) andharness(the whole assembled agent); today'sintegration/e2e/evalare all the harness scope under realness flags. This entry is the deterministic harness run (model: "mock", the default) — the workhorse you gate every commit, with no key. A real-model harness run is non-deterministic by definition, so you don't assert a single one — youmeasure()it across trials (the eval scope).egressis a capability of this scope (the e2e tier), not a separate tier.Behaviour is identical to
runHarnessTest(which it wraps); the new name +modelflag make the scope/realness explicit and steer real-model runs to the right tool.