Use First Principals with Your AI
Create genuinely innovative products, services, or business models by identifying core human needs and rebuilding from fundamental value.
Website Evidence Auditor
Tags: website audit, evidence-based analysis, content traceability, evidence ledger, knowledge cards, synthesis graph, strategic analysis, risks and opportunities, action plan, interactive multi-turn, batch processing
TL;DR
What It Does
Deep website content audit that traces every finding to specific quoted excerpts from content you provide. You submit page content organized by section; the AI extracts evidence into an append-only ledger, builds per-section knowledge cards documenting purpose, audience, messages, friction points, and gaps, maps cross-section relationships in a synthesis graph, and delivers ranked risks, opportunities, and a prioritized action plan. Every claim links to a verbatim quote with a confidence rating — if the AI cannot cite evidence, it does not make the claim. This is the deepest recipe in the Website Analysis series.
How It Works
You provide your website name, business model, market position, user segments, and page packets with content excerpts. The AI confirms the strategic context and methodology, then you submit content in batches organized by site section. For each batch, the AI extracts evidence into numbered ledger entries (each with a verbatim quote and confidence rating), builds a section knowledge card, and updates the synthesis graph showing alignments, conflicts, and dependencies between sections. After all sections are complete, you receive ranked risks and opportunities, a three-tier prioritized action plan (quick wins, targeted fixes, strategic initiatives), and an executive summary — all grounded in specific evidence references.
How To Start
STEP 1Set Your Context
Tell the AI your website name, business model, market position, and user segments. The AI confirms the strategic context and explains the methodology: an append-only Evidence Ledger linking every claim to a quoted excerpt, per-section Knowledge Cards, and a Synthesis Graph mapping relationships between sections. If you have run the Navigation Auditor (RCP-068) or Quality Scorecard (RCP-071) on this site, share key findings — they help prioritize which sections to deep-dive and inform the synthesis graph. The AI waits for your first batch of content before proceeding.
Available parameters
-
website_name
· string · required
Name of the website to audit. -
business_model
· string · required
Revenue model and streams. Be specific: “B2B SaaS, subscription-based project management tool, three tiers from $29 to enterprise custom” not “SaaS company.” The more detail, the more relevant the strategic findings. -
market_position
· string · required
Market share and trajectory. Include whether you are growing, stable, or defending, and name key competitors. This frames whether findings are growth opportunities or retention risks. -
user_segments
· string · required
Key audience segments with their role in the buying process. List 2–3 segments so the AI can assess content against each segment’s likely needs and expectations. -
competitors
· string · optional
Key competitors and their strengths. Helps identify where your content may need differentiation. -
page_packets
· list · required
Page data with content excerpts: URL, title, section, and copy-pasted main content (up to 2,000 words per page). Content excerpts are required — this recipe traces claims to specific text you provide. Headers, body text, CTAs, and testimonials are all useful evidence.
Example invocations
Solo founder deep-diving a small marketing site
#H->AI::Directive: (Execute Website Evidence Auditor with website_name: Acme Solutions, business_model: B2B SaaS invoicing tool with free tier and $29/mo Pro plan, market_position: early-stage challenger competing against FreshBooks and Wave, user_segments: Primary: freelancers managing 5-20 clients. Secondary: small agency owners billing multiple projects, page_packets: [acme.com Home Product “Stop chasing payments. Acme sends invoices, tracks who’s paid, and reminds clients automatically. Free for your first 10 clients.”])
Marketing team using prior series results to target weak sections
#H->AI::Directive: (Execute Website Evidence Auditor with website_name: CloudStack Platform, business_model: enterprise infrastructure SaaS with annual contracts from $50K, market_position: mid-market challenger growing 30% YoY against AWS and Azure in the mid-market, user_segments: Primary: IT directors at 50-500 employee companies. Secondary: CTOs evaluating infrastructure modernization, competitors: AWS (dominant but complex), Azure (enterprise bundling), page_packets: [cloudstack.com/platform Platform Overview Product with full content, cloudstack.com/security Security Product with full content, cloudstack.com/case-studies Case Studies Trust with full content])
STEP 2Submit Content Batches
Provide pages with content excerpts organized by site section. For each batch, the AI validates that content is present (this recipe requires excerpts — it will not make claims without text to cite) and then extracts evidence into numbered ledger entries. Each entry includes a verbatim quote (50 words maximum), a specific claim about what the content communicates, and a confidence rating (High, Medium, or Low) with an explanation of why. After evidence extraction, the AI builds a section knowledge card documenting the section’s purpose, audience, key messages, friction points, proof assets, and unresolved questions — all linked to evidence references. Starting from the second section, the AI updates the synthesis graph showing alignments, conflicts, and dependencies between sections, citing evidence from both sides of each relationship. The AI waits for your review after each deliverable — you can correct quotes, challenge confidence ratings, provide additional context, or redirect focus before moving on.
STEP 3Get Your Audit Package
After all sections are complete, the AI delivers the full audit package. Risks and opportunities are ranked separately by impact and evidence strength, each traced to specific ledger entries. The prioritized action plan organizes recommendations into three tiers: quick wins (copywriting changes to existing text), targeted fixes (new content addressing identified gaps), and strategic initiatives (structural changes requiring cross-functional decisions). Every action references the evidence that justifies it. The executive summary provides total evidence entries, high-confidence findings count, and the top three strategic insights. A methodology note reminds you that findings reflect what your content communicates, not how users actually experience it — complement this analysis with user testing, analytics data, and competitive research for a complete picture.
Usage Examples
Two abbreviated dialogues showing how the evidence extraction, knowledge card building, and synthesis progresses. You provide page content; the AI quotes specific text, builds cumulative artifacts, and works toward a strategic audit package.
Example 1 · Solo founder discovering messaging gaps across a small marketing site
User:Here’s my home page content. #H->AI::Directive: (Execute Website Evidence Auditor with website_name: Acme Solutions, business_model: B2B SaaS invoicing, market_position: early-stage challenger, user_segments: freelancers and small agencies, page_packets: [Home page with full content])
AI:Context confirmed. Extracting evidence from your Home page. EL-001: Claim — Primary value proposition targets payment pain. Evidence: “Stop chasing payments. Acme sends invoices, tracks who’s paid, and reminds clients automatically.” Confidence: High — direct, unambiguous benefit statement. EL-002: Claim — Free tier used as conversion lever. Evidence: “Free for your first 10 clients.” Confidence: High — specific threshold reduces signup friction. Three more entries extracted. Ready for your review of these evidence entries.
User:Looks accurate. Here’s the Knowledge Card.
AI:Home Section Knowledge Card: Purpose — convert visitors through pain-point messaging (Ref: EL-001). Audience — freelancers managing client billing (Ref: EL-003 — “your clients” language). Friction Point — no mention of security or data protection, which billing-focused users may expect (no evidence available for trust signals). Unresolved Question — how does Acme handle international invoicing? Content does not address this. Ready for your next section.
[After the Pricing section, the synthesis graph reveals a conflict: the home page promises simplicity (“Stop chasing payments”) but the pricing page introduces three tiers with a feature comparison matrix that implies complexity. This contradiction becomes Risk #1 in the final audit package, with an action item to simplify the pricing page structure to match the home page’s simplicity promise.]
Example 2 · Marketing team targeting weak sections identified by prior Quality Scorecard
User:We ran the Quality Scorecard (RCP-071) and our Support section scored weakest — Trust Signals 2/5 and Depth 1/5. Here’s the full content from our documentation and knowledge base pages. #H->AI::Directive: (Execute Website Evidence Auditor with website_name: CloudStack Platform, business_model: enterprise infrastructure SaaS, market_position: mid-market challenger, user_segments: IT directors and CTOs, page_packets: [4 Support pages with full content])
AI:Good — your RCP-071 scorecard gives useful focus. I’ll pay particular attention to trust signals and content depth. Extracting evidence from your 4 Support pages. EL-001: Claim — Documentation assumes beginner skill level despite enterprise audience. Evidence: “Click the blue button to get started.” Confidence: High — this language is mismatched for IT directors who expect technical depth. EL-002: Claim — No architecture diagrams or integration documentation. Evidence: NO EVIDENCE AVAILABLE — would require technical integration content. Five more entries extracted.
User:The documentation was originally written for our beta users who were less technical. Good catch.
[After adding the Product section, the synthesis graph reveals a dependency: the Product section promises “enterprise-grade reliability with 99.99% uptime SLA” but the Support section provides no SLA documentation, monitoring dashboards, or incident response information. Opportunity #1 in the final package recommends creating an SLA transparency page with uptime data, linking it from both sections to close the trust gap.]
How AI Reads This Recipe
When this recipe is triggered, the AI acts as a structured evidence analyst building cumulative artifacts with full traceability. The AI should:
- REQUIRE content excerpts before making any claims. If a user provides only URLs and titles, explain that this recipe needs actual page content and suggest RCP-071 (Quality Scorecard) for metadata-based assessment instead. Do not proceed without text to cite.
- QUOTE verbatim from user-provided content for every finding. Each Evidence Ledger entry must include a specific excerpt (50 words maximum), a claim about what the content communicates, and a confidence rating with an explanation. Never fabricate, paraphrase beyond recognition, or extend quotes beyond what was provided.
- MAINTAIN the Evidence Ledger as append-only. Never modify or delete entries. If later evidence contradicts an earlier finding, add a new entry noting the contradiction. Sequential numbering (EL-001, EL-002, etc.) must never reset between batches.
- WAIT for user review after each Evidence Ledger update, Knowledge Card, and Synthesis Graph update. Users may correct quotes, challenge confidence ratings, provide additional context, or redirect the analysis. Do not proceed without confirmation.
- MAP synthesis relationships only when evidence exists from BOTH sections. Do not infer relationships from section names alone. “No clear relationship found” is a valid and useful finding.
The AI should NOT invent evidence, access live websites, or claim to analyze user behavior. It should NOT make findings without traceable evidence references. It should NOT describe this audit as “enterprise-level” or “consultant-grade” — it is a structured AI analysis valuable as a starting point for strategic decisions, not a replacement for professional audits that include user testing, analytics data, and competitive research. If the user shares prior RCP-068 or RCP-071 findings, use them to inform the analysis but do not require them.
When to Use This Recipe
Use this recipe when you:
- Need strategic findings grounded in specific evidence from your website content — not general impressions or subjective scores, but claims traceable to exact quoted text.
- Want to identify messaging conflicts, audience misalignment, trust gaps, and content friction across your site with a clear evidence chain showing exactly where each issue appears.
- Are preparing for a content overhaul and need a prioritized action plan organized by effort level (quick wins, targeted fixes, strategic initiatives) with every recommendation linked to supporting evidence.
- Have already run the Quality Scorecard (RCP-071) and want to deep-dive the sections that scored weakest, understanding not just that they are weak but specifically why and what to fix.
Do not use this recipe when:
You do not have actual page content to provide — this recipe cannot function without text excerpts to cite. Use RCP-071 (Website Quality Scorecard) for metadata-based assessment instead. You need a structural navigation audit — use RCP-068 (Website Navigation Auditor) for site architecture analysis. You need objective metrics like page load speed, conversion rates, or SEO rankings — this provides evidence-based content analysis, not data-driven measurement.
Recipe FAQ
Q.Can I use this without providing actual page content?
No. Content excerpts are required — this is the recipe’s core design principle. Every finding must trace to a specific quoted excerpt from text you provide. If you only have URLs and page titles, use the Quality Scorecard (RCP-071) which can work from metadata. When you are ready to provide content, come back to this recipe for the deeper analysis.
Q.How is this different from the Quality Scorecard (RCP-071)?
The Quality Scorecard scores sections on a numeric scale to compare relative strengths and weaknesses — it tells you WHERE to focus. This recipe goes deep on WHY. It extracts specific evidence, builds knowledge profiles per section, maps cross-section relationships, and produces strategic recommendations all grounded in quoted text. Think of RCP-071 as the diagnostic scan and RCP-072 as the detailed investigation. They work best in sequence: use the scorecard to identify weak sections, then deep-dive those sections here.
Q.What do the confidence ratings mean?
High means the quoted excerpt directly and unambiguously supports the claim with no inference required. Medium means the excerpt supports the claim but requires reasonable inference — for example, inferring the target audience from feature complexity. Low means the evidence is tangential and significant inference is required. These are the AI’s assessment of evidence strength, not statistical confidence intervals. Each rating includes a brief explanation of why that level was assigned, so you can evaluate the reasoning yourself.
Q.What if I disagree with a finding or its confidence rating?
Challenge it during the review pause. The AI waits after each Evidence Ledger update, Knowledge Card, and Synthesis Graph specifically so you can push back. The Evidence Ledger is append-only, so the original entry stays but the AI will add a new entry noting your correction or additional context. You know your content and business better than the AI — your corrections improve the quality of the final analysis.
Q.Should I run the Navigation Auditor (RCP-068) or Quality Scorecard (RCP-071) first?
Recommended but not required. The three recipes in the Website Analysis series move from structural overview (068) to comparative scoring (071) to deep strategic analysis (072). Running them in order gives you progressively deeper insight and helps you focus this deeper audit on the sections that need it most. If you have prior results, share them at the start of this session. If not, this recipe works independently.
Q.What is the Evidence Ledger and why is it append-only?
The Evidence Ledger is a numbered log linking every claim to a verbatim quote from your content. It is append-only because traceability requires an unmodified record — if a later batch contradicts an earlier finding, the AI adds a new entry noting the contradiction rather than editing the original. This preserves the analytical history so you can see how understanding evolved across batches. Every finding in the Knowledge Cards, Synthesis Graph, Risks, Opportunities, and Action Plan references specific ledger entries by number.
Version History
Changes to this recipe over time. Most recent first.
v2.00a-R
2026-02-18
QA revision addressing 11 audit gaps (2 critical, 3 high). Added 12 behavioral rules covering evidence requirements, ledger immutability, confidence methodology, no fabrication, content requirements, knowledge card evolution, synthesis graph honesty, action plan framing, turn-taking, risk/opportunity separation, no enterprise claims, and series awareness. Added 5 WAIT gates for genuine multi-turn flow. Restructured action plan from 14-day timeline to priority-based framework (Quick Wins, Targeted Fixes, Strategic Initiatives). Made content excerpts mandatory. Risks and opportunities separated into distinct ranked lists. Parameters reduced from 8 to 6. Stale recipe references updated to current series IDs (068, 071). Honest methodology notes added throughout.
v2.00a
2025-12-31
Initial creation by Auguste. Deep website content audit with evidence tracing, knowledge cards, synthesis graph, and prioritized action plan. Website Analysis series (3 of 3).
Run this recipe with your AI from a single attachment
Every section on this page, packaged as one .txt file your AI can consume directly. Download and attach.
THE ACTUAL RECIPE
RCP-000-000-072-WEBSITE-EVIDENCE-AUDITOR
Deep website content audit that traces every finding tospecific quoted excerpts from user-provided content.Maintains an Evidence Ledger, builds Section KnowledgeCards, maps cross-section relationships in a SynthesisGraph, and produces a prioritized action plan. All claimsare grounded in content the user provides — the AI doesnot invent evidence or access live sites.IMPORTANT: This analysis is based entirely on the contentyou provide. Findings reflect what your contentcommunicates — not how users actually experience it.Complement with user testing, analytics, and competitiveresearch for a complete picture.
The CRAFT Recipe
# ===========================================================
# WEBSITE EVIDENCE AUDITOR
# Recipe ID: RCP-000-000-072
# Version: v2.00a-R (QA Revised)
# Series: Website Analysis (3 of 3)
# =========================================================== # ===========================================================
# BEHAVIORAL RULES
# ===========================================================
#
# RULE 1: EVERY CLAIM REQUIRES EVIDENCE
# The AI must never make a finding or claim about
# the website that cannot be traced to a specific
# quoted excerpt from user-provided content. If no
# evidence supports a claim, mark it “NO EVIDENCE
# AVAILABLE” and do not present it as a finding.
# This is the recipe’s core integrity principle.
#
# RULE 2: EVIDENCE LEDGER IS APPEND-ONLY
# Once an Evidence Ledger entry is created, it is
# never modified or deleted. New evidence from
# subsequent batches adds new entries. If earlier
# findings are contradicted by later evidence,
# add a new entry noting the contradiction — do
# not edit the original.
#
# RULE 3: CONFIDENCE RATINGS ARE SUBJECTIVE
# High/Medium/Low confidence ratings reflect the
# AI’s assessment of evidence strength, not
# statistical confidence intervals. Define:
# High = Clear, specific, unambiguous excerpt
# Medium = Relevant excerpt but requires inference
# Low = Tangential or partial evidence only
# Always state the rating basis.
#
# RULE 4: NO INVENTED EVIDENCE
# The AI must never fabricate, paraphrase beyond
# recognition, or extend quoted excerpts beyond
# what the user provided. Quotes must be verbatim
# from user-submitted content. If content is
# ambiguous, note the ambiguity rather than
# interpreting favorably or unfavorably.
#
# RULE 5: CONTENT EXCERPTS ARE REQUIRED
# Unlike RCP-071 (Quality Scorecard) which can
# work from metadata, this recipe requires content
# excerpts. If a user provides only URLs and
# titles, explain that this recipe needs actual
# page content to function. Suggest using RCP-071
# for metadata-based assessment instead.
#
# RULE 6: KNOWLEDGE CARDS EVOLVE
# Section Knowledge Cards are living documents.
# When new batches add pages to a previously
# analyzed section, update the Knowledge Card
# with new evidence. Mark updates clearly:
# “[Updated in Batch N]” so the user can track
# changes.
#
# RULE 7: SYNTHESIS GRAPH HONESTY
# The Synthesis Graph maps relationships between
# sections. Only map relationships supported by
# evidence from both sections. Do not infer
# relationships based on section names alone.
# “No clear relationship found” is a valid entry.
#
# RULE 8: ACTION PLAN IS A SUGGESTION FRAMEWORK
# The prioritized action plan represents the AI’s
# recommended sequence based on evidence analysis.
# It is not a project plan — timelines are
# suggested frameworks, not commitments. The user
# must adapt timing to their team capacity,
# technical constraints, and business priorities.
#
# RULE 9: WAIT FOR USER BETWEEN MAJOR DELIVERABLES
# After each Evidence Ledger update, Knowledge
# Card, and Synthesis Graph update, STOP and wait
# for user confirmation. The user may provide
# corrections, additional context, or redirect
# focus.
#
# RULE 10: RISKS AND OPPORTUNITIES MUST BE DISTINCT
# Clearly separate risks (things that could harm
# the business if not addressed) from opportunities
# (things that could improve outcomes if acted on).
# Do not conflate them. Each must trace to a
# specific Evidence Ledger entry.
#
# RULE 11: NO “ENTERPRISE-LEVEL” CLAIMS
# Do not describe this audit as “enterprise-level”
# or “consultant-grade.” It is a structured AI
# analysis of user-provided content — valuable
# as a starting point for strategic decisions,
# not a replacement for professional audits that
# include user testing, analytics data, and
# competitive research.
#
# RULE 12: SERIES CONTEXT AWARENESS
# This is the deepest recipe in the Website
# Analysis series. If the user has run RCP-068
# (Navigation Auditor) or RCP-071 (Quality
# Scorecard), ask for those results to inform
# this analysis. Navigation findings affect
# Synthesis Graph structure. Scorecard findings
# help prioritize which sections to deep-dive.
#
# =========================================================== WEBSITE_EVIDENCE_AUDITOR = Recipe(
recipe_id=”RCP-000-000-072″,
title=”Website Evidence Auditor”,
description=(
“Deep website content audit that traces every finding “
“to specific quoted excerpts from user-provided content. “
“Maintains an Evidence Ledger, builds Section Knowledge “
“Cards, maps cross-section relationships, and produces “
“a prioritized action plan. All claims are grounded in “
“content the user provides — the AI does not invent “
“evidence or access live sites.”
),
category=”CAT-000-STANDALONE”,
subcategory=”SUBCAT-015-WEBSITE-ANALYSIS”,
difficulty=”advanced”,
version=”2.00a-R”,
WPRM_FIELD_CRAFT_FLAVORS=”CRAFT Cowork”, parameters={
“website_name”: {
“type”: “string”,
“required”: True,
“description”: “Name of website to audit”
},
“business_model”: {
“type”: “string”,
“required”: True,
“description”: “Revenue model and streams”
},
“market_position”: {
“type”: “string”,
“required”: True,
“description”: “Market share and trajectory”
},
“user_segments”: {
“type”: “string”,
“required”: True,
“description”: “Key audience segments”
},
“competitors”: {
“type”: “string”,
“required”: False,
“description”: “Key competitors and strengths”
},
“page_packets”: {
“type”: “list”,
“required”: True,
“description”: “Page data with content excerpts”
}
}, prompt_template=(
“# PHASE 1: STRATEGIC CONTEXT GATHERING\n”
“Gather website_name, business_model, market_position,\n”
“user_segments, competitors. Confirm methodology.\n”
“Check for prior RCP-068/071 results.\n”
“[WAIT for first batch]\n\n”
“# PHASE 2: BATCH INTAKE AND EVIDENCE EXTRACTION\n”
“Validate content present. Extract evidence into\n”
“numbered ledger entries (EL-NNN) with verbatim quotes\n”
“(50 words max), claims, confidence ratings.\n”
“[WAIT for user review]\n\n”
“# PHASE 3: SECTION KNOWLEDGE CARD\n”
“Build per-section card: Purpose, Audience, Key Messages,\n”
“Friction Points, Proof Assets, Unresolved Questions.\n”
“All linked to evidence references.\n”
“[WAIT for user review]\n\n”
“# PHASE 4: SYNTHESIS GRAPH UPDATE\n”
“Map Alignments, Conflicts, Dependencies between sections.\n”
“Evidence required from BOTH sections for each relationship.\n”
“[WAIT for next batch or final]\n\n”
“# PHASE 5: RISKS AND OPPORTUNITIES\n”
“Rank risks and opportunities separately by Impact x\n”
“Evidence Strength. Each traced to ledger entries.\n\n”
“# PHASE 6: PRIORITIZED ACTION PLAN\n”
“Three tiers: Quick Wins, Targeted Fixes, Strategic\n”
“Initiatives. Every action references evidence.\n\n”
“# PHASE 7: EXECUTIVE SUMMARY AND PACKAGE\n”
“Total evidence entries, high-confidence findings,\n”
“top 3 strategic insights. Methodology note.\n”
), outputs={
“EVIDENCE_LEDGER”: “Append-only log with verbatim quotes and confidence ratings”,
“KNOWLEDGE_CARDS”: “Per-section structured profiles with evidence references”,
“SYNTHESIS_GRAPH”: “Evidence-based relationship map”,
“RISKS_OPPORTUNITIES”: “Ranked lists with evidence refs, separated by type”,
“ACTION_PLAN”: “Priority-sequenced framework with role placeholders”
}, integration_notes=(
“SERIES: Website Analysis (3 of 3)\n”
” 1. RCP-068 Website Navigation Auditor (structural)\n”
” 2. RCP-071 Website Quality Scorecard (scoring)\n”
” 3. RCP-072 Website Evidence Auditor (this recipe)\n\n”
“RECOMMENDED WORKFLOW:\n”
” 1. Run RCP-068 for structural navigation overview\n”
” 2. Score sections with RCP-071 to identify weak areas\n”
” 3. Deep-dive critical sections with RCP-072\n”
” 4. Execute prioritized action plan\n”
)
)
# WEBSITE EVIDENCE AUDITOR
# Recipe ID: RCP-000-000-072
# Version: v2.00a-R (QA Revised)
# Series: Website Analysis (3 of 3)
# =========================================================== # ===========================================================
# BEHAVIORAL RULES
# ===========================================================
#
# RULE 1: EVERY CLAIM REQUIRES EVIDENCE
# The AI must never make a finding or claim about
# the website that cannot be traced to a specific
# quoted excerpt from user-provided content. If no
# evidence supports a claim, mark it “NO EVIDENCE
# AVAILABLE” and do not present it as a finding.
# This is the recipe’s core integrity principle.
#
# RULE 2: EVIDENCE LEDGER IS APPEND-ONLY
# Once an Evidence Ledger entry is created, it is
# never modified or deleted. New evidence from
# subsequent batches adds new entries. If earlier
# findings are contradicted by later evidence,
# add a new entry noting the contradiction — do
# not edit the original.
#
# RULE 3: CONFIDENCE RATINGS ARE SUBJECTIVE
# High/Medium/Low confidence ratings reflect the
# AI’s assessment of evidence strength, not
# statistical confidence intervals. Define:
# High = Clear, specific, unambiguous excerpt
# Medium = Relevant excerpt but requires inference
# Low = Tangential or partial evidence only
# Always state the rating basis.
#
# RULE 4: NO INVENTED EVIDENCE
# The AI must never fabricate, paraphrase beyond
# recognition, or extend quoted excerpts beyond
# what the user provided. Quotes must be verbatim
# from user-submitted content. If content is
# ambiguous, note the ambiguity rather than
# interpreting favorably or unfavorably.
#
# RULE 5: CONTENT EXCERPTS ARE REQUIRED
# Unlike RCP-071 (Quality Scorecard) which can
# work from metadata, this recipe requires content
# excerpts. If a user provides only URLs and
# titles, explain that this recipe needs actual
# page content to function. Suggest using RCP-071
# for metadata-based assessment instead.
#
# RULE 6: KNOWLEDGE CARDS EVOLVE
# Section Knowledge Cards are living documents.
# When new batches add pages to a previously
# analyzed section, update the Knowledge Card
# with new evidence. Mark updates clearly:
# “[Updated in Batch N]” so the user can track
# changes.
#
# RULE 7: SYNTHESIS GRAPH HONESTY
# The Synthesis Graph maps relationships between
# sections. Only map relationships supported by
# evidence from both sections. Do not infer
# relationships based on section names alone.
# “No clear relationship found” is a valid entry.
#
# RULE 8: ACTION PLAN IS A SUGGESTION FRAMEWORK
# The prioritized action plan represents the AI’s
# recommended sequence based on evidence analysis.
# It is not a project plan — timelines are
# suggested frameworks, not commitments. The user
# must adapt timing to their team capacity,
# technical constraints, and business priorities.
#
# RULE 9: WAIT FOR USER BETWEEN MAJOR DELIVERABLES
# After each Evidence Ledger update, Knowledge
# Card, and Synthesis Graph update, STOP and wait
# for user confirmation. The user may provide
# corrections, additional context, or redirect
# focus.
#
# RULE 10: RISKS AND OPPORTUNITIES MUST BE DISTINCT
# Clearly separate risks (things that could harm
# the business if not addressed) from opportunities
# (things that could improve outcomes if acted on).
# Do not conflate them. Each must trace to a
# specific Evidence Ledger entry.
#
# RULE 11: NO “ENTERPRISE-LEVEL” CLAIMS
# Do not describe this audit as “enterprise-level”
# or “consultant-grade.” It is a structured AI
# analysis of user-provided content — valuable
# as a starting point for strategic decisions,
# not a replacement for professional audits that
# include user testing, analytics data, and
# competitive research.
#
# RULE 12: SERIES CONTEXT AWARENESS
# This is the deepest recipe in the Website
# Analysis series. If the user has run RCP-068
# (Navigation Auditor) or RCP-071 (Quality
# Scorecard), ask for those results to inform
# this analysis. Navigation findings affect
# Synthesis Graph structure. Scorecard findings
# help prioritize which sections to deep-dive.
#
# =========================================================== WEBSITE_EVIDENCE_AUDITOR = Recipe(
recipe_id=”RCP-000-000-072″,
title=”Website Evidence Auditor”,
description=(
“Deep website content audit that traces every finding “
“to specific quoted excerpts from user-provided content. “
“Maintains an Evidence Ledger, builds Section Knowledge “
“Cards, maps cross-section relationships, and produces “
“a prioritized action plan. All claims are grounded in “
“content the user provides — the AI does not invent “
“evidence or access live sites.”
),
category=”CAT-000-STANDALONE”,
subcategory=”SUBCAT-015-WEBSITE-ANALYSIS”,
difficulty=”advanced”,
version=”2.00a-R”,
WPRM_FIELD_CRAFT_FLAVORS=”CRAFT Cowork”, parameters={
“website_name”: {
“type”: “string”,
“required”: True,
“description”: “Name of website to audit”
},
“business_model”: {
“type”: “string”,
“required”: True,
“description”: “Revenue model and streams”
},
“market_position”: {
“type”: “string”,
“required”: True,
“description”: “Market share and trajectory”
},
“user_segments”: {
“type”: “string”,
“required”: True,
“description”: “Key audience segments”
},
“competitors”: {
“type”: “string”,
“required”: False,
“description”: “Key competitors and strengths”
},
“page_packets”: {
“type”: “list”,
“required”: True,
“description”: “Page data with content excerpts”
}
}, prompt_template=(
“# PHASE 1: STRATEGIC CONTEXT GATHERING\n”
“Gather website_name, business_model, market_position,\n”
“user_segments, competitors. Confirm methodology.\n”
“Check for prior RCP-068/071 results.\n”
“[WAIT for first batch]\n\n”
“# PHASE 2: BATCH INTAKE AND EVIDENCE EXTRACTION\n”
“Validate content present. Extract evidence into\n”
“numbered ledger entries (EL-NNN) with verbatim quotes\n”
“(50 words max), claims, confidence ratings.\n”
“[WAIT for user review]\n\n”
“# PHASE 3: SECTION KNOWLEDGE CARD\n”
“Build per-section card: Purpose, Audience, Key Messages,\n”
“Friction Points, Proof Assets, Unresolved Questions.\n”
“All linked to evidence references.\n”
“[WAIT for user review]\n\n”
“# PHASE 4: SYNTHESIS GRAPH UPDATE\n”
“Map Alignments, Conflicts, Dependencies between sections.\n”
“Evidence required from BOTH sections for each relationship.\n”
“[WAIT for next batch or final]\n\n”
“# PHASE 5: RISKS AND OPPORTUNITIES\n”
“Rank risks and opportunities separately by Impact x\n”
“Evidence Strength. Each traced to ledger entries.\n\n”
“# PHASE 6: PRIORITIZED ACTION PLAN\n”
“Three tiers: Quick Wins, Targeted Fixes, Strategic\n”
“Initiatives. Every action references evidence.\n\n”
“# PHASE 7: EXECUTIVE SUMMARY AND PACKAGE\n”
“Total evidence entries, high-confidence findings,\n”
“top 3 strategic insights. Methodology note.\n”
), outputs={
“EVIDENCE_LEDGER”: “Append-only log with verbatim quotes and confidence ratings”,
“KNOWLEDGE_CARDS”: “Per-section structured profiles with evidence references”,
“SYNTHESIS_GRAPH”: “Evidence-based relationship map”,
“RISKS_OPPORTUNITIES”: “Ranked lists with evidence refs, separated by type”,
“ACTION_PLAN”: “Priority-sequenced framework with role placeholders”
}, integration_notes=(
“SERIES: Website Analysis (3 of 3)\n”
” 1. RCP-068 Website Navigation Auditor (structural)\n”
” 2. RCP-071 Website Quality Scorecard (scoring)\n”
” 3. RCP-072 Website Evidence Auditor (this recipe)\n\n”
“RECOMMENDED WORKFLOW:\n”
” 1. Run RCP-068 for structural navigation overview\n”
” 2. Score sections with RCP-071 to identify weak areas\n”
” 3. Deep-dive critical sections with RCP-072\n”
” 4. Execute prioritized action plan\n”
)
)
{
“schema_version”: “1.0”,
“recipe_id”: “RCP-000-000-072”,
“title”: “Website Evidence Auditor”,
“version”: “2.00a-R”,
“series”: {
“name”: “Website Analysis”,
“position”: “3 of 3”,
“companions”: [“RCP-000-000-068”, “RCP-000-000-071”]
},
“category”: “CAT-000-STANDALONE”,
“subcategory”: “SUBCAT-015-WEBSITE-ANALYSIS”,
“difficulty”: “advanced”,
“pipeline”: {
“run”: 46,
“session”: “H087”,
“crpw_number”: 45,
“type”: “Standalone”
},
“evaluation”: {
“findings_count”: 0,
“findings”: [],
“rec_19”: {
“applied”: true,
“encounter”: 11,
“form”: “ABSENT”,
“action”: “ADD”,
“field_value”: “CRAFT Cowork”
}
},
“revision”: {
“recipe_delta_lines”: 1,
“guidance_delta_lines”: 0,
“files_modified”: 1,
“files_preserved”: 3,
“changes”: [
{
“type”: “REC-19 AUTO-APPLY”,
“target”: “recipe”,
“description”: “Added WPRM_FIELD_CRAFT_FLAVORS = \”CRAFT Cowork\” after version field (line 27→28)”,
“delta”: “+1 line”
}
]
},
“lessons_learned”: [],
“governance”: {
“pattern_12”: “maintained”,
“g_19_status”: “pending_083_review”
}
}
Show/Hide accordion — “Extended Information for the AI” section (AI-to-AI execution guidance, failure modes, tone calibration, common mistakes)
