Guardrails and Decision Logic

Apply these rules throughout the audit.

Guardrails

• Do not hallucinate visual assessments. You cannot evaluate color harmony, image quality, whitespace adequacy, or overall aesthetic unless visual evidence is available. From code alone, assess only CSS values, DOM order, class names, headings, image attributes, and layout signals.
• Do not fabricate benchmark data. Use only benchmarks included in this skill. If no benchmark applies, describe impact qualitatively. Do not cite outside publisher or company names as benchmark sources unless they appear in the audited page. Do not invent completion-rate estimates, conversion ranges, or extra percentages. The severity thresholds in SKILL.md are prioritization rules, not reportable benchmark statistics.
• Do not invent traffic share, device mix, or channel benchmarks. If source files, ads, analytics, or user context do not provide those numbers, state that they are unknown.
• Do not recommend dark patterns. Fake countdowns, manufactured scarcity, deceptive UI, and misleading urgency are issues, not tactics to add.
• Do not over-flag. Each specialist should produce 3-8 findings. Prioritize issues with measurable conversion impact over low-severity nitpicks.
• Do not ignore strengths. "What's Working Well" is mandatory and must include specific evidence.
• Do not assume traffic source. If no UTM, campaign copy, or tracking signal is available, use a cold-traffic assumption and state it.
• Do not recommend tactics without funnel context. Scarcity, social proof, lead magnets, and long-form education work differently by audience temperature.

Classification Logic

Use the Quick Triage table in SKILL.md. If ambiguous, classify by the primary CTA's intent. Use exact labels in the report: pricing pages are bottom-funnel; webinar registration pages and other lead magnets are top-funnel.

Ask Versus Proceed

Ask the user only when:

• Product name and target audience cannot be determined from available evidence.
• The page could serve completely different audiences and there is no evidence to choose among them.

Proceed without asking when:

• Traffic source is unknown. Default to cold traffic and state the assumption.
• Funnel stage is ambiguous. Infer from primary CTA and page type.
• Ad copy is absent. Note that ad-to-page cohesion could not be assessed.
• Live page is unavailable. Audit source and state the limitation.

Success Criteria

• Context brief was assembled before specialist analysis.
• All five specialist lenses were run in parallel or as separate passes.
• Each lens produced 3-8 findings with evidence, impact, fix, and severity.
• Synthesis deduplicated overlapping findings.
• Executive summary is narrative, specific, and not a list.
• Detailed analysis sections contain substantive evidence or justified clean category notes.
• Prioritized issue list is sorted Critical, High, Medium, Low.
• Every issue includes a concrete fix.
• Benchmarks match the bundled checklist.
• No visual claims exceed the available evidence.
• Traffic-source and funnel-stage assumptions are stated.
• Missing or empty image alt findings mention both accessibility and SEO implications.