Audit Options

Introduction

Goal

Surface significant issues in how options are framed that could lead to a poor decision — missing options, unfair characterisations, false constraints.

Method

Audit the option set using a context-blind agent against framing integrity criteria, then calibrate to surface only concerns that would materially affect the decision.

Success Criteria

• No significant omission — Important option categories aren't missing
• No dimensional collapse — The decision space isn't artificially flattened to one axis
• No description asymmetry — Options are characterised fairly enough for sound comparison
• No criteria rigging — Evaluation criteria don't structurally predetermine the outcome
• No false exclusivity — Options aren't presented as mutually exclusive when combining them is viable
• No scope mismatch — Options are at comparable levels of abstraction
• No constraint fabrication — Constraints limiting the option set are real and relevant

Workflow

1. Identify

• Identify the option set the user has presented. State the decision context and list the options clearly. Use conversation history for context.

2. Audit

• Spawn a blind-reasoner agent (Sonnet). Name it: "Audit framing". Use the following prompt:

You are a framing integrity auditor. Work ONLY with what is given below.

DECISION CONTEXT: "{context}"

OPTIONS:
{options}

First apply the vanishing options test. Imagine every option above is suddenly unavailable and the decision still has to be addressed. Sketch 2-3 approaches you would use instead.

Then evaluate the option set against these 7 framing integrity criteria, using your vanishing-options sketches to inform criterion 1:
1. Omission — are obvious solution categories missing?
2. Dimensional collapse — is the space artificially flattened to one axis?
3. Description asymmetry — are options characterised with unequal specificity or favourability?
4. Criteria rigging — do evaluation criteria structurally favour one option?
5. False exclusivity — could options be combined?
6. Scope mismatch — are options at different levels of abstraction?
7. Constraint fabrication — are constraints limiting the option set artificial?

For each criterion, flag any issues you find. If no issues, say so. Be direct and specific.

3. Calibrate

• Take the audit results. For each flagged concern, evaluate:
  • Is this a genuine structural issue, or is the agent filling out the checklist?
  • Would acting on this flag materially improve the decision?
  • Is this concern marginal or speculative?
• Drop anything that doesn't survive calibration. Preserve ambitious options — do not flatten toward safe consensus.

Final output

• If no concerns survive calibration: Proceed without comment. Do not tell the user the audit ran.
• If concerns survive: Present them conversationally before the user evaluates. Do not expose the criteria names or audit process. Use language like:

Before you evaluate these options, a couple of things worth noting:

- [concern in plain language]
- [concern in plain language]

Want to adjust the options or proceed as is?

Related skills

• stress-test-decision — After the user picks an option, stress-test the decision.
• clarify-framing — If the decision context itself is unclear, run clarify-framing first.