Developing for Mizan: extending the core
For contributors who do need to touch the core: an orientation to Mizan's architecture and where new metric kinds and behaviors plug in.
ghchinoy
7 posts by ghchinoy
Turn a writing skill into an eval: grading this series with Mizan
The capstone closes the loop: take the editorial standard behind these posts and encode it as a Mizan rubric that grades the series itself.
Contribute a template pack: extend Mizan without touching the core
You do not have to fork Mizan to add value to it. This piece walks authoring, validating, and sharing a template pack that others can import.
Using Mizan in your role: per-persona playbooks, part 2 (Genmedia configurator, Brand Lab)
The second pair of playbooks: the genmedia configurator embedding an eval-set as a calling shape, and the Brand Lab user generating a rubric from a brand book and freezing it into a reusable metric.
Using Mizan in your role: per-persona playbooks, part 1 (Asset creator, Asset manager)
Two roles, two playbooks. How an asset creator finds the shortest path from one eval to a scorecard, and how an asset manager curates a set that reports each concern on its own.
One spectrum, four users: from brand alignment to technical metrics
Four Mizan users sit on one spectrum, from "is this on-brand?" to "is this grounded and accurate?" The same machinery answers both. Only the criteria change.
Why evals, and what is an LLM-as-a-judge, really?
You already judge generative output by eye. An eval turns that private judgment into something explicit, repeatable, and shareable.