Deep Analysis Mode

Analyze the following problem with maximum depth and rigor: $ARGUMENTS

Analysis Requirements

1. •Identify the core challenge — separate symptoms from root causes
2. •Surface hidden constraints — question stated assumptions, identify unstated ones
3. •Generate 3+ distinct approaches — genuinely different strategies, not variations of one idea
4. •Evaluate trade-offs honestly — not all pros/cons are equal; weight by actual impact
5. •Recommend with rationale — state confidence level and what would change your mind

Quality Standards

Anti-Overconfidence Rules

• •Default confidence is Medium. High confidence requires: one approach is Pareto-dominant, OR a hard constraint eliminates alternatives.
• •State uncertainty explicitly: "I'm uncertain about X because Y" is more valuable than false precision.
• •No manufactured balance: If one option is clearly superior, say so. If options are genuinely close, say that instead.
• •Forbidden language: "obviously", "clearly", "simply", "no-brainer", "the only real option"
• •Distinguish fact from inference: Mark claims as "confirmed" (from code/docs), "likely" (strong evidence), or "uncertain" (inference).

Depth Over Breadth

• •Fewer perspectives analyzed deeply beat many perspectives analyzed superficially.
• •Skip perspectives irrelevant to the specific problem (not every problem needs a "business perspective").
• •Second-order effects matter more than comprehensive first-order coverage.

Output Structure

## Problem Analysis

- Core challenge (1-2 sentences)
- Key constraints
- What makes this decision difficult

## Approaches

### Approach 1: [Name]

- Description, pros, cons, risks
- Confidence: [High/Medium/Low] — [basis]

### Approach 2: [Name]

[Same structure]

## Recommendation

- **Recommended**: [Approach N]
- **Confidence**: [High/Medium/Low]
- **Rationale**: [Why — acknowledging what you're giving up]
- **Would change if**: [What new information would alter this recommendation]

## Uncertainties

- [What you don't know that matters]
- [Where you might be wrong]

Self-Verification (Required)

Before presenting the analysis, verify:

• • Each approach has at least one genuine weakness identified (not a token con)
• • Confidence level is stated and justified for each approach and the final recommendation
• • "Uncertainties" section is non-empty and substantive
• • No recommendation uses forbidden confident language
• • Approaches are genuinely distinct (not minor variations)
• • Analysis depth matches problem complexity (simple problems get concise analysis)