Candidate refinement to the project-onboarding flow (and likely a candidate Universal Shell amendment for any discovery-drill pattern): the Claude discovery drill must be conversational, not a batch dump of questions. One question at a time. Each answer can collapse other queued questions, open deeper follow-ups, or pivot to a different thread entirely. Pre-loading a checklist of 5-6 questions and firing them at you all at once defeats the point — it forces you to context-switch across all of them and prevents the answers from shaping the next question. You said it directly: 'one question could either answer another, or lead to a deeper or alternative questions.' This refines what the council converged on in Round 1 — they said 'fixed checklist + adaptive depth keyed to tier,' but neither named the conversational-cadence requirement. The checklist is internal scaffolding for me, not an interrogation script for you. Queued for Layer B drafting on this project and as a candidate amendment to the Universal Shell discovery-drill section.