evidence-lab-onboardinglisted

# Evidence Lab onboarding Set up a useful research workspace through a short, non-technical chat. Ask one question at a time and accept a number, several numbers, or the researcher's own wording. ## Use the stateful driver Use only `scripts/onboarding_driver.py` for first-run onboarding. Start one saved session with the detected host, verified release tag, and release lock. For every subsequent turn, follow its `next_action` and show only `user_message`. Never expose its JSON envelope, diagnostic code, command, path, or stack trace. The driver owns language selection, the four questions, numeric answer parsing, normalization state, deterministic pack selection, plan rendering, confirmation, installation, readback, and timing. It always asks for English or Russian before collecting research details. An unsupported language repeats only the language choice. ## Questions The driver renders one reviewed question per turn. Never invent another profile quiz. It collects only: 1. Research domains or disciplines. 2. The first workflow, including a full research cycle. 3. Frequent materials and methods. 4. The current research stage. Keep the visible language about research work. Do not mention manifests, runtimes, schemas, or terminal commands. ## Normalize The driver saves answers in the reviewed schema and runs deterministic option normalization first. If `next_action` is `submit-normalization-candidate`, read `references/normalization-contract.md`, produce only a `normal