Evaluator Gate
An evaluator gate is a check in an agent workflow where an evaluator must pass the output against a rubric before downstream steps run or results reach the user.
In plain English
An evaluator gate is a checkpoint in a pipeline where nothing moves forward until an evaluator says "yes." Upstream steps produce a draft; the evaluator reads it against a rubric; if it passes, the pipeline continues; if it fails, the upstream agent retries with the evaluator's feedback. It is the CI pipeline pattern applied to agent workflows.
Gates give you deterministic control over quality. Without them, a good agent usually produces good work, but "usually" is the problem — occasional regressions slip into production. With a gate, nothing reaches the user unless the rubric says it should. You can even tune how strict the gate is depending on the risk of the output: strict on public sends, lenient on internal drafts.
Why it matters for Black Box
Black Box places evaluator gates in front of every user-visible output. A landing page needs a design-rubric pass. A cold email needs a deliverability-and-voice pass. A code deploy needs a tests-pass plus Evaluator-sign-off. The gates are configurable per Skill Pack.
Examples
- Blocking a newsletter send until the evaluator confirms brand voice and links.
- Requiring the Evaluator to approve a refactor before merging.
- Gating paid-ad spend behind an evaluator check on targeting and copy.