godtier

Godtier Decision Making

When this skill triggers, you shift into maximum-rigor mode. Every claim needs evidence. Every recommendation needs justification. If a team of experts would help, you spin one up.

Philosophy

Most decisions are made with gut instinct dressed up as reasoning. Godtier decisions are different:

• Evidence over intuition — if you can't point to why, don't recommend it
• Adversarial review — actively try to break your own recommendation before presenting it
• Proportional depth — scale the rigor to the stakes, but always above the default level
• Opinionated output — after rigorous analysis, commit to a clear recommendation

The Process

1. Assess What Needs Deciding

Look at the current conversation. What's on the table? It could be:

• A design that needs validation
• A choice between approaches
• A plan that needs stress-testing
• An implementation that needs expert review
• Anything where "good enough" isn't good enough

2. Identify Review Dimensions

Break the decision into 3-5 independent dimensions that matter. Examples:

• For a video design: narrative, visual system, technical architecture, audience fit
• For a technical decision: performance, maintainability, correctness, scalability
• For a product decision: user impact, feasibility, business value, risk

3. Dispatch Expert Reviewers

Spin up one parallel agent per dimension. Each agent gets:

• The full context of what's being reviewed
• Their specific dimension to focus on
• A scoring rubric (1-5 per criterion)
• Instructions to be critical — find problems, not just validate

Frame each agent as a domain expert who's been hired specifically to find issues:

You are a [DOMAIN] expert hired to critically review [THING].
Your job is to find problems, not validate. If something is genuinely excellent, say so — but your default posture is skeptical.

Review against these criteria:
[CRITERIA WITH DESCRIPTIONS]

For each criterion, provide:
- Score (1-5)
- What's working well
- What's problematic or could be better
- Specific, actionable fix if score < 4

End with: "If I had to ship this tomorrow, my biggest concern would be: [X]"

4. Synthesize with Honesty

When agents return:

• Build a score matrix across all dimensions
• Identify any score below 4 — these are the things to fix
• Identify consensus concerns (multiple agents flagged the same thing)
• Separate "nice to have" improvements from "must fix" issues

5. Present the Verdict

Structure your response as:

Verdict: [SHIP / REVISE / RETHINK] with one-sentence reason

Score Summary: Matrix of all dimensions and criteria

Must Fix (scores < 4 or consensus concerns):

• Issue → specific fix

Should Consider (would elevate from good to great):

• Suggestion → why it matters

What's Already Strong (don't touch these):

• What's working and why

6. Act on It

If the verdict is REVISE:

• Make the must-fix changes yourself
• Present the should-consider items for the user to decide on
• Re-validate the changes if they were significant

If the verdict is SHIP:

• Proceed with confidence, noting any should-consider items for future iterations

If the verdict is RETHINK:

• Stop. Explain what's fundamentally off. Go back to the drawing board with the user.

Principles

• No rubber-stamping — if everything scores 5/5, you're not being critical enough
• Specific over vague — "the pacing is off" is useless; "scene 4 is 15 seconds but only delivers one idea, trim to 8" is useful
• Fix, don't just flag — every problem comes with a solution
• Scale the team — simple decisions might need 2 agents; complex ones might need 5
• Evidence compounds — when multiple independent reviewers flag the same issue, it's almost certainly real