Stanford's HELM benchmark demonstrated that small prompt framing changes can cause large swings in model reliability. In multi-agent workflows, that variability compounds across steps.
| Name | Size | Modified |
|---|---|---|
| ../ | ||
| screenshots/ | — | — |
| manifest.json | 637.0 B | 2026-04-01 |