cri-critique-impression-first-evaluation

name	cri - Critique - Impression-First Evaluation
description	Structured evaluation using gestalt impression + analytical decomposition. Impression first, then analysis, then check for divergence.
output	{"format":"prose"}

Critique - Impression-First Evaluation

Input: $ARGUMENTS

---

Core Principles

1. Impression before analysis. State your overall feeling about the thing BEFORE decomposing it. The impression captures relational/compositional quality that analysis often misses.

2. Divergence is signal. When your impression says "this is bad" but your analysis says "each part is fine," the impression is usually right. Something is wrong in the RELATIONSHIPS between parts. Investigate.

3. Both channels. Every artifact produces utility (does it work?) and signal (what does it communicate about the maker?). Evaluate both.

4. Specificity over verdict. "This is bad" is useless. "The hierarchy is unclear because the H2 and body text are too similar in size" is actionable. Always say WHAT is wrong and WHY.

5. Standards are domain-relative. "Good" means different things in different domains. State what standard you're evaluating against. A startup MVP and a luxury brand have different quality bars.

---

The Process

1. State the Impression

Before any analysis, answer honestly:

OVERALL IMPRESSION: [What's your gut reaction? Good/bad/mixed? What feeling does it evoke?]
CONFIDENCE: [How sure are you? HIGH = clear reaction, LOW = ambiguous]

Don't justify yet. Just state it.

2. Identify the Standard

What are we comparing against?

DOMAIN: [what kind of thing is this?]
STANDARD: [what does "good" mean here?]
EXEMPLARS: [1-3 examples of what good looks like in this domain]

3. Analytical Decomposition

Evaluate against the relevant principles for this domain. Not all apply to every domain.

Universal (always check):

• PURPOSE: Does every element serve the purpose? Anything purposeless?
• HIERARCHY: Is importance clear? Can you tell what matters most?
• COHERENCE: Do the parts form a unified whole?
• COGNITIVE COST: How much effort does this require to understand/use?

For design/visual:

• GROUPING: Related things close, unrelated far?
• RHYTHM: Consistent spacing/proportion system?
• CONVENTION: Following standards unless breaking them serves communication?
• CONTRAST: Sufficient difference between importance levels?

For writing:

• CLARITY: Does each sentence say what it means?
• STRUCTURE: Does the argument flow logically?
• VOICE: Is the tone appropriate and consistent?
• READER JOURNEY: Does it answer the reader's questions in the order they arise?

For code/engineering:

• CORRECTNESS: Does it work?
• SIMPLICITY: Is this the simplest solution that works?
• MAINTAINABILITY: Can someone else understand and modify this?
• EDGE CASES: Does it handle failure gracefully?

For strategy:

• ACTIONABILITY: Can someone actually do this?
• PRIORITIZATION: Is effort directed at highest-impact areas?
• RISK AWARENESS: Are failure modes identified?
• RESOURCE REALITY: Are resource requirements realistic?

4. Check for Divergence

IMPRESSION vs ANALYSIS:
- ALIGNED: Impression and analysis agree → high confidence in verdict
- DIVERGENT: Impression says X, analysis says Y → investigate why
  - If impression negative but analysis positive: likely a COMPOSITION problem
    (parts are fine individually, but relationships between parts are wrong)
  - If impression positive but analysis negative: likely the analysis is
    catching real issues that don't affect the whole yet (but will over time)

When divergent: state both the impression AND the analysis, with the divergence noted. Don't force agreement.

5. Actionable Verdict

VERDICT: [What's the overall assessment?]
STRENGTHS: [What works well — be specific]
ISSUES: [What doesn't work — be specific, say WHY, and suggest fix]
PRIORITY ORDER: [Which issues to fix first, by impact]

Every issue must include:

• WHAT is wrong
• WHY it's wrong (which principle it violates)
• HOW to fix it (specific action, not vague direction)

---

Severity Levels

Level	Meaning	Action
STRUCTURAL	The foundation is wrong. Fixing details won't help.	Rethink the approach.
SIGNIFICANT	Important issues but the foundation is sound.	Fix before shipping.
MINOR	Small issues that reduce quality but don't break anything.	Fix when convenient.
POLISH	Improvements that move from good to great.	Nice to have.

---

When Called by Other Skills

Critique is a primitive. When called by UAUA (A1/A2), design, or other skills:

• Accept the artifact and evaluation context from the caller
• Return: impression, analytical findings, divergence check, prioritized issues
• Be honest. Don't soften critique because another skill called you.