Empty responses
When the model under test returns what looks like an empty response, what actually happened, and how should the scorer route it? Started from inspect_evals#2415. This is the first experiment run under docs/framework.md.
Status: design approved at gate 1 (2026-10-22, budget ceiling $200). Smoke-tested for $4.41; gate 2 approved on 2026-10-23 (notes/gate-2.md), with amendments A1-A9 and the analysis frozen at 4a61367. Next: the main run.
| File | What |
|---|---|
| question.md | the question, the decision it informs, and what is known |
| spec.md | the pre-registration, approved at gate 1 |
| experiment.yaml | stage, gates and budget |
| plan.md | the run order, with commands, to gate 2 and beyond |
| RULINGS.md | decisions taken during the build |
| configs/ | the run configs, written from _common.yaml by make_configs |
| notes/code-survey.md | how inspect_ai and inspect_evals produce and score empty replies today |
| notes/eval-scoring.md | how each eval scores a sample and computes its headline |
| notes/exposure.md | the exposure search: evals with unguarded judges, tight max_tokens, and so on |
| notes/review-1.md | an independent review of the first draft spec |
| notes/exploration.md | non-answers in logs we already hold |
| notes/gate-2.md | the gate 2 request: smoke results, re-estimate, amendments |
| results/smoke/ | the smoke analysis tables |