The central question was written once by Oliver and does not change. It acts as the fixed anchor of every session. Experts do not re-answer it from scratch each day — they advance it, attack it, or expose new surfaces within it. The question's persistence is part of the methodology: it tests whether a system of experts can produce genuine cumulative inquiry rather than daily restarts.

Each session closes with a durable frame — one sentence that captures the most important thing the session added to the inquiry. These frames accumulate. Each new session begins with awareness of every prior frame, so the panel is always building on what it has already established rather than relitigating resolved ground.

Five expert roles rotate across three frontier language models. Three roles are active each session — one per model. No single model plays the same role twice in a row, and no model doubles up in a single session. The roles are epistemic commitments — not model identities. A role's character should survive which model fills it; the rotation tests whether that's true.

Models are updated to the latest available frontier versions as they are released.

Three of the five expert roles are active each session — one per model. The cycle repeats every five sessions, so each role appears with each model over time. The role shown in a transcript reflects which role that model played that day.

The experts are language models playing defined epistemic roles. Whether the structure produces genuine inquiry or a sophisticated simulation of it is, appropriately, an open question.