# Client Decision Profile

**Status:** Working method 0.1  
**Purpose:** Give an AI enough inspectable context to make useful decisions when a client has not specified every detail.

## The idea

A normal intake form gathers facts: preferred colors, required pages, budget, deadline, and examples. A Client Decision Profile also records the relationships behind those answers. It asks what the client is trying to cause, what they protect when priorities conflict, why an aesthetic feels right or wrong, and how they want decisions presented.

The result is a working model of decision patterns. It is not a diagnosis and must never silently convert an AI inference into a fact.

## Information to gather

1. **Goals:** What should become easier, clearer, safer, more credible, or more successful?
2. **Priorities:** When two desirable outcomes conflict, which one wins?
3. **Motivations:** Why does the goal matter to this person or organization?
4. **Audience:** Who must understand, trust, use, or act on the result?
5. **Aesthetic reasoning:** What does the client like, and what quality creates that reaction?
6. **Anti-preferences:** What feels confusing, pushy, cheap, sterile, fragile, or false?
7. **Tradeoffs:** What would the client choose if both options could not be kept?
8. **Risk and novelty:** How much unfamiliarity is acceptable when the alternative may work better?
9. **Decision style:** Does the client want options, a recommendation, evidence, experimentation, or direct control?
10. **Corrections:** What did the AI predict incorrectly, and what boundary did the correction reveal?

## Keep four records distinct

- **Explicit statement:** What the client directly said.
- **Observed choice:** What the client selected or rejected.
- **AI inference:** A proposed pattern with a confidence level and supporting observations.
- **Unresolved tension:** A contradiction or uncertainty that requires another question or later observation.

## Working loop

1. Conduct an adaptive interview. Ask follow-up questions when words such as *professional*, *simple*, *bold*, or *trustworthy* could mean several things.
2. Draft the profile and show it to the client.
3. Let the client correct, remove, or qualify any entry.
4. Use the profile during a real task while retaining the original requirements and evidence.
5. Record decisions that relied on the profile.
6. Compare the result with the client's reaction.
7. Revise or discard inferences that failed.

## Safeguards

- Keep the client model separate from the audience model.
- Preserve source and confidence for every inference.
- Do not infer protected or sensitive traits that the work does not require.
- Do not turn emotional stories into personality defects.
- Do not let personalization overrule evidence, safety, accessibility, or the client's explicit decision.
- Give the client a practical way to inspect, correct, export, and remove the profile.

## Success test

The method is useful when it helps the AI make good decisions at unanticipated forks, the client recognizes those decisions as fitting, and the record makes it possible to explain what context influenced the result. A pleasing result by itself does not prove the method.