# Candidate Work-Sample Rubric

Use this rubric when reviewing early Edxperimental Labs candidates. The goal is to select people who can produce research, benchmark artifacts, and client-facing analysis that improves real AI decisions.

| Dimension | What Strong Looks Like |
| --- | --- |
| Evidence taste | Cites primary sources, shows uncertainty, and keeps claims attached to inspectable proof. |
| Builder speed | Turns an idea into a memo, harness, analysis page, script, or demo without waiting for perfect instructions. |
| Buyer clarity | Explains the decision, the risk, the tradeoff, and the next test in language an operator can use. |
| Scope discipline | Chooses a narrow useful artifact instead of pretending to solve a whole category in one pass. |
| Aesthetic judgment | Ships work that feels polished enough for a serious AI buyer to trust and share internally. |

## Review Notes

- Prefer inspectable artifacts over polished but unverifiable claims.
- Ask how the candidate would improve the artifact after one client conversation.
- Look for a clear decision, not just information density.
- Route sales-engineering candidates to Saujas after the first screen if the artifact shows client-facing judgment.