# Microsoft — We can build it. Can we judge it?

- Company: Microsoft (microsoft.com)
- Announced: 2026-10-20T14:53:11+00:00
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://commandline.microsoft.com/mini-foundry-design-journey-evaluation/
- Record: https://forck.live/items/25262-we-can-build-it-can-we-judge-it
- Subject: Foundry / Responsible AI

Microsoft has moved a team of designers into VS Code with GitHub Copilot and built Mini-Foundry, a lighter modular version of its product now used by more than 100 designers and product managers. The team says loading and testing the full production application was slowing exploration and stakeholder feedback, so UX engineers created the lighter version with shared components, relevant data and repository-specific agent skills. Designers learn Git through that work, creating branches, making commits and building on one another's changes. The UX engineering team is also developing a journey rubric Lab around the principle "checked when possible, judged when needed", where each rubric packages criteria, examples, executable checks or judging prompts, and a report. Deterministic checks verify specific requirements while model-based assessments interpret evidence against criteria and labeled examples, and those judgments can vary between runs, making repeated evaluation and human calibration important. Setup starts with a brief covering where the journey begins, who attempts it, what they need to accomplish and what evidence counts as success, then the application URL and task boundaries are supplied and the rubric runs against that flow. The team says an automated run can surface a concern but does not establish how people experience it, so colleagues inspect the evidence, try to reproduce the issue and use research to verify or challenge the finding.

## Evidence

Verbatim from https://commandline.microsoft.com/mini-foundry-design-journey-evaluation/:

> Our UX engineering team has been developing a journey rubric Lab around a principle that leaves room for both rigor and judgment: checked when possible, judged when needed. Each rubric packages criteria, examples, executable checks or judging prompts, and a report.

---

Record: https://forck.live/items/25262-we-can-build-it-can-we-judge-it
Catalogue: https://forck.live/llms.txt
Current issue: https://forck.live/feed.md