Find unknown

Find Unknown

Overview

Run a 5–10 minute, evidence-first preflight. Surface at most three direction-changing unknowns, then investigate, ask, prototype, or experiment according to where evidence can come from.

Route by Evidence

Observable condition	Action
Code, documents, data, tools, or environment can supply evidence	Investigate: inspect and report supported findings.
The gap is the user's goal, preference, authority, or trade-off	Ask: put one direction-changing decision to the user.
The user can only recognize a preference by seeing it	Prototype: produce 2–4 cheap, distinct options.
Neither party can know directly, or the claim concerns future effects	Experiment: state a falsifiable hypothesis and the smallest useful test.

Preflight

1. Name the decision at risk. Inspect available files, code, references, constraints, and evidence.
2. Inventory candidates silently. Rank by ability to invalidate the direction, then route by the table.
3. Return zero to three material unknowns; at most three is a ceiling, not a quota.
4. When preference requires recognition, Prototype before asking. Show 2–4 distinct options, let users react to what they can see, and treat their reaction as evidence.
5. Ask only when a user-only decision blocks direction. Ask one question per turn and at most three questions for the whole preflight.
6. After each answer, treat it as evidence. Update the current direction; Discard resolved or invalidated unknowns; check dependencies; Re-rank against the updated direction.
7. If one unknown is a prerequisite gate, make it the only active branch. Defer downstream branches until it passes; name them briefly as deferred, not as active unknowns or questions.

Stop when the direction is updated, no unanswered user-only decision can immediately invalidate it, and every remaining material unknown has a validation path. At the question limit, report unresolved validation paths.

Output

## Unknowns Preflight: <topic>
**Current direction:** <commitment being checked>

1. **<unknown>** — <investigate | ask | prototype | experiment>
   - Evidence: <observed evidence, or “not yet available”>
   - Direction impact: <what could change>
   - Next validation: <smallest concrete action>

When a user-only decision blocks direction, append and wait:

**One decision question:** <one direction-changing choice>
- Provisional recommendation: <best current default>
- Basis: <supporting evidence, or an explicitly labeled assumption>
- Directional consequences: <how materially different answers change the direction>

When stopping, close with:

## Decision-ready summary
**Current direction:** <updated direction>
**Material assumptions:** <remaining assumptions, clearly labeled>
**Next validations:** <up to three concrete actions>

Guardrails

• Separate facts from hypotheses. Keep recommendations provisional and expose their basis.
• Inspect before asking. Prototype taste. Experiment on future effects.
• Do not force four categories, three unknowns, or three questions.
• If no material unknown remains, say so and continue.