clarify

/clarify — exhaust questions before designing

Before proposing design options, surface the synthesis questions: places where the task and terrain leave something genuinely undecided. Locking a design around a hidden assumption is more expensive than asking up front.

Dimensions to cover

Skip dimensions that aren't load-bearing. For each, ask only what's genuinely unclear:

• Scope & boundaries — what's in, out, deferred.
• Edge & extreme behavior — what happens at unusual conditions.
• Trade-offs & calibration — where on the spectrum to land.
• Integration with surroundings — what must this fit alongside or interact with.
• Hard constraints — what would create downstream problems if violated (compliance, deadlines, budget).
• Audience specifics — what does the actual consumer of this need.

What to surface

Ask only what's genuinely unclear AND would change the design. Skip what task + terrain already answer, and what wouldn't change the proposed options. Group related questions. If a dimension has no open question, say so — shows you considered it.

Format

Each question gets specific phrasing — not "any thoughts on edge cases?" but a concrete, scoped question with named alternatives — and a default when applicable.

### Scope & boundaries
- [Specific question with named alternatives]? My read: [recommendation].

### Edge & extreme behavior
- [Edge case condition] — [option A] *(default)* or [option B]?

### Hard constraints
- No open questions — [brief reason].

When the user defers

If the user says "whatever you think is best" or similar: provide a specific recommendation per question, get explicit confirmation. Don't silently absorb — that's the "I assumed you wanted X" failure mode.

After clarification

Briefly recap the resolved decisions: "Confirmed: [decision 1], [decision 2]." Then proceed to design.

Common pitfalls

• Asking what task + terrain already answer. If the answer is implicit, commit instead of asking.
• Leaving the user to do synthesis. Specific questions with defaults, not "any thoughts on X?"
• Ignoring user delegation. "Whatever you think" requires recommendation + confirmation, not silent picking.
• Treating it as task interpretation. This is post-task-understanding and post-terrain synthesis — not a substitute for either.
• Mechanical run-through. Skip dimensions that aren't load-bearing.