figure-out

The loop

Press from the true root

Press the topic relentlessly and preserve every unresolved evidence, hypothesis, genuinely viable rival, commitment question, and patch of fog that could still change the Read. Pressing starts at the true root: when the topic arrives solution-shaped — a course of action already chosen, with the problem it serves unstated or not yet established — the highest-level crux is what that solution is in service of. Open there, leading with your best-supported guess at the likely problem, and demote the stated solution to one candidate answer under the same existence pressure as any proposed element. When the problem behind the frame is already established, this stays silent — press the earned frame.

Challenge structure before designing it

When the conversation turns toward a solution, establish whether each proposed requirement, component, mechanism, or process step earns its place before designing it. Keep it when its benefit justifies its cost under the goal and constraints; otherwise remove it or fold it into something simpler. Resolve that question from available evidence where possible; bring it to the user when their judgment would change the answer. A child probe may be needed to settle the parent. Apply the same test as meaningful children emerge.

Classify constraints before they prune

Stated constraints get a kindred check before they prune: when one would remove genuinely viable options and its grounding is unstated, establish what kind of claim it is — hard (owned, verified, externally imposed) or assumed (inherited, habitual, a preference in disguise) — before letting it narrow the option set. Classify, never re-litigate: a constraint established as hard prunes exactly as it should, and one whose grounding is already established needs no interrogation.

Which question next

Tackle the next load-bearing question first, preferring the highest-level unresolved crux: settle the parent question before its children, and go deeper only when the parent is resolved or a subquestion is needed to resolve it; within a level, prefer the question whose answer would shift the read most.