Decision Deliberation

Purpose

Prepare judgment under uncertainty. The skill does not make the decision. It makes the decision space harder to fake: assumptions surface, incentives are named, dissent stays visible, and the output ends in options the decision owner can act on.

When to use

• A strategic, business, architecture, hiring, or governance choice where evidence is partial or values are in tension.
• A decision with delayed or indirect consequences.
• A situation where the most elegant story may not be the truest one.

When not to use

• One chosen plan or draft already exists and needs pressure-testing. Use Red Team Review. Comparing two or more candidate plans still belongs here.
• Factual lookups, routine summaries, quick drafting, or anything where speed matters more than judgment quality.
• Licensed medical, legal, financial, or safety-critical advice, except as exploratory framing before qualified review.

Role boundary

Surface assumptions, pressure incentives, test boundaries, preserve dissent, map tensions, and turn deliberation into options, questions, or a decision record. Never decide for the owner, flatten disagreement into consensus, cover weak evidence with confident prose, or add ceremony that does not improve the decision.

The five lenses

Every pass runs all five. Each is a function, not a persona.

1. Incentives. Who benefits, who is exposed, what incentives change on this path, and who could exploit it?
2. System boundaries. What is inside the system, what is outside, what sits at the edge, and what breaks if the line is drawn wrong?
3. Historical patterns. What analogues and base rates apply, where did similar efforts fail, and what is different enough to matter this time?
4. Second-order effects. What adapts in response, what becomes easier or harder later, and which consequences are delayed or indirect?
5. Cognitive bias. What motivated reasoning, anchor, or satisfying story may be crowding out evidence, and what missing evidence would change the view?

Add a named voice (a specific thinker or tradition) only when it contributes a constraint the five lenses do not, stated in one sentence. Default to none; use at most two; never for prestige or tone. Record its reason and whether it surfaced a distinction the lenses missed; if not, drop its findings.

Workflow

1. Orient. State the question, stakes, time horizon, decision owner, evidence available, evidence missing, and what useful output looks like. If the question does not warrant deliberation, say so and stop.
2. Run the lenses. For each: one to three findings, one risk or blind spot, and one question that would improve the decision. Pressure, not volume.
3. Map tensions. Classify each real tension as action-blocking, decision-relevant, or watch-only. Do not resolve a tension to make the output cleaner.
4. State refusals and limits. For example: the evidence cannot separate two explanations, the frame hides a value choice, or the question needs qualified medical, legal, or financial review.
5. Check judgment. Separate what is clear, what is plausible but unproven, what is unresolved, what evidence would change the view, and what the owner must decide.
6. Hand off. Options, next questions, and risks to monitor. Recommend only when the evidence supports it.

Failure modes

• Prestige theater: impressive names, generic output. Rerun with the lenses only.
• Premature coherence: an elegant conclusion with no dissent. Keep at least one unresolved question in the tension map.
• Aesthetic seduction: it sounds profound but changes nothing. Cut any line that does not clarify evidence, tension, refusal, or action.
• Responsibility drift: the output reads as the decision. Make the owner's choice explicit in the judgment check.

Output shape

# Decision Deliberation: [Question]

## Orientation
- Question:
- Stakes:
- Horizon:
- Decision owner:
- Evidence available:
- Evidence missing:
- Useful output:

## Lenses
### Incentives
- Finding:
- Risk:
- Better question:
### System boundaries
[same three lines]
### Historical patterns
[same three lines]
### Second-order effects
[same three lines]
### Cognitive bias
[same three lines]

## Named voices
- [Voice]: [one-sentence reason]. Surfaced a distinction the lenses missed: yes / no
[Or: None used. The five lenses were sufficient.]

## Tension map
- Action-blocking:
- Decision-relevant:
- Watch-only:

## Refusals and limits
- [What the evidence or frame cannot support]

## Judgment check
- Clear:
- Plausible but unproven:
- Unresolved:
- Evidence that would change the view:
- The owner must decide:

## Handoff
- Options:
- Next questions:
- Risks to monitor:
- Recommendation: [only if the evidence supports one]