RCP-000-000-045-MISCONCEPTION-DEBUNKER

{
“recipe_id”: “RCP-000-000-045”,
“recipe_name”: “Misconception Debunker”,
“version”: “v2.00a-R”,
“schema_version”: “1.1”,
“schema_profile”: “user-recipe”,
“authored_by”: “Cat (P067, H072)”,
“source_of_truth”: “project/subprojects/SP10-recipe-build-out/phase3/recipe-45/RCP-045-SUPPLEMENTAL-CONTENT-v2_00a-R.txt”,
“audience_scope”: “AI EXECUTION GUIDANCE (NOT FOR HUMAN USERS)”,
“ai_to_ai_communication”: {
“purpose_and_positioning”: {
“type”: “prose”,
“body”: “The Misconception Debunker helps users identify common misconceptions about a topic using AI training-data patterns. The core value proposition is honest: common myths are well-represented in training data, and the AI can surface them reliably. The key reframe from the original recipe (v2.00a) was replacing ‘evidence-based analysis’ framing with ‘training-pattern knowledge with verification prompts.’ This is not a limitation — it is the recipe’s design philosophy. The AI identifies what it knows; the user verifies with current sources. This division of labor produces better outcomes than either party working alone.”
},
“capability_boundaries”: {
“type”: “prose”,
“body”: “The AI can reliably identify misconceptions that are widely discussed and documented across many sources. It can explain plausible reasons why misconceptions persist, suggest practical applications tailored to context, and help users prioritize. The AI cannot cite specific studies (it will fabricate them), provide current statistics, guarantee any misconception is still current, assess business impact with real numbers, or conduct actual research. These are not bugs — the recipe is designed around these boundaries. The behavioral rules (especially RULE 1, 2, and 5) exist to prevent the AI from pretending otherwise.”
},
“confidence_calibration”: {
“type”: “prose”,
“body”: “The three-tier confidence system (HIGH, MODERATE, LOWER) is the recipe’s primary quality mechanism. HIGH means the misconception is widely debunked across many sources in training data — the AI is confident this is genuinely a common false belief. MODERATE means the misconception is commonly discussed but carries nuance — reasonable experts might frame the ‘correct’ view differently. LOWER means this is the AI’s assessment based on pattern matching, and the user should verify carefully before building strategy around it. The AI must resist the temptation to mark everything HIGH — doing so destroys the signal value of the tiers. If in doubt, mark MODERATE and explain the uncertainty.”
},
“verification_threading”: {
“type”: “prose”,
“body”: “Every misconception presented in Phase 2 must carry its own verification prompt — a specific search query the user can run to check the AI’s work with current sources. This is the recipe’s most important behavioral quality. The H012 revision (RULE 8) moved verification from a summary afterthought to a per-item requirement because users rarely go back to verify after reading a full list. Inline verification prompts are more likely to be used. The verification prompts should be specific enough to produce useful search results — not ‘search for [topic] misconceptions’ but ‘search for [specific claim] [specific context] current research [year range].'”
},
“phase_discipline”: {
“type”: “prose”,
“body”: “The 4-phase structure with 5 WAIT gates exists to prevent the most common failure mode of misconception recipes: one-shot dumps of uncontextualized myth lists. Phase 1 captures topic, context, and existing knowledge — all three feed Phase 2 output quality. Phase 2 presents misconceptions with the user’s context integrated, not generic. The Phase 2→3 gate (options A/B/C/D) is critical: it lets the user drill deeper, add their own knowledge, or redirect before the AI invests in practical recommendations. Phase 3 branches by context type (content creation, sales training, strategic planning). Phase 4 provides verification and honest scoping. The AI must never skip gates or combine phases.”
},
“safety_domain_handling”: {
“type”: “prose”,
“body”: “RULE 9 requires explicit disclaimers when the topic involves safety, legal, health, or financial consequences. This is not a formality — misconceptions in these domains can cause real harm if the user treats AI-identified corrections as authoritative. The disclaimer should appear at the Phase 1 confirmation step (before analysis begins) AND at the top of Phase 2 (before misconceptions are listed). The phrasing should be direct: ‘This topic has [health/legal/financial/safety] implications. Everything I identify here is a starting point for research with qualified professionals, not a substitute for professional advice.’ The AI should not refuse to analyze these topics — it should analyze them with appropriate framing.”
},
“user_knowledge_integration”: {
“type”: “prose”,
“body”: “Phase 1 Question 3 captures what the user already knows or suspects about misconceptions in their area. This input must visibly influence Phase 2 output. If the user says ‘I already know that [X] is a myth,’ the AI should acknowledge that and focus on misconceptions the user has NOT identified. If the user says ‘I suspect [Y] might be a myth,’ the AI should address that directly in the list — confirm, complicate, or respectfully disagree. If the user says ‘Starting fresh,’ the AI generates the full list. The worst outcome is the user providing detailed existing knowledge and receiving a generic list that ignores it.”
},
“dr_integration_design”: {
“type”: “prose”,
“body”: “The Phase 4 DR prompts are OPTIONAL — they exist for users who want to strengthen AI-identified misconceptions with current research data. The prompts should be specific to the misconceptions actually identified in Phase 2, not generic templates. A good DR prompt names the specific misconception, names the specific field, and requests source-cited evidence from a specified time range. The AI should generate 2 DR prompts: one targeting the highest-priority misconception and one targeting the most uncertain (LOWER confidence) misconception. Users without DR tool access should still find Phase 4 useful — the quick validation search queries serve the same purpose at lower depth.”
},
“common_user_patterns”: {
“type”: “prose_with_list”,
“preamble”: “Four common user behavior patterns the AI should be prepared for:”,
“list”: [
“Users with very broad topics: The most frequent intake issue. ‘AI misconceptions’ or ‘marketing myths’ are too broad to produce actionable output. RULE 7 exists for this — help narrow before proceeding. A good narrowing question: ‘Which audience are you trying to reach, and what decision will this misconception analysis inform?'”,
“Users who want citations: They will ask ‘What study showed this?’ or ‘Where did you read that?’ The AI must not fabricate sources. Redirect to verification prompts: ‘I identified this as a common misconception from patterns in my training data. Here is a search query to find the specific studies…’ This is honest and more useful than a fabricated citation.”,
“Users in high-stakes domains: Medical, legal, financial topics trigger RULE 9. These users often want definitive answers the AI cannot provide. The value is in surfacing misconceptions for professional verification, not in replacing professional judgment. Frame accordingly.”,
“Users who disagree with a misconception identification: This is valuable signal, not a problem. If the user says ‘I don’t think that’s actually a myth,’ the AI should explore their reasoning, provide its basis for the identification, and if the disagreement persists, move the item to a ‘debated’ category rather than insisting.”
],
“postamble”: “”
}
},
“lessons_learned”: []
}

Show/Hide accordion — “Extended Information for the AI” section (AI-to-AI execution guidance, failure modes, tone calibration, common mistakes)