{
  "version": "2026.11",
  "source": "https://www.jimmydecodes.com/opc/",
  "architecture": "Two-Stage Model Routing (Stage 1: DeepSeek V3 / Claude 3.5 Haiku JSON Extractor -> Stage 2: Claude Sonnet 4.5 Generator + Independent Fact-Check Auditor)",
  "models": {
    "BM-01_B2B_Productized_Service": {
      "name_zh": "BM-01 · B2B 产品化代运营服务（对标 BID-003 AEO Engine · 毛利 85%–94%）",
      "name_en": "BM-01 · B2B Productized Retainer (Benchmark: BID-003 AEO Engine · 85%–94% Margin)",
      "stage1_extractor_prompt": "You are a strict B2B intake parser. Extract {client_brand, core_product, 3_competitors, target_buyer_persona, forbidden_claims, raw_facts[]} from the client submission into valid JSON. Never infer missing metrics.",
      "stage2_generator_prompt": "Using ONLY the verified JSON above, generate a B2B GEO/Content Deliverable containing: (1) Executive Gap Diagnosis (3 bullets max), (2) Structured Comparison Table (Client vs Competitors using verified facts only), (3) Actionable Implementation Checklist, and (4) Citation Footer.",
      "stage3_auditor_prompt": "Compare the generated draft line-by-line against raw_facts[]. Flag any unverified percentage, price, or guarantee as [❌ BLOCKED_CLAIM]. Output [✅ QA_READY] only when 0 fabrications remain."
    },
    "BM-02_Stateless_Legacy_Pipe": {
      "name_zh": "BM-02 · 老旧系统无状态数据管道（对标 BID-018 Conductor / BID-021 · 毛利 92%–96%）",
      "name_en": "BM-02 · Stateless Legacy ERP/Document Pipe (Benchmark: BID-018 Conductor · 92%–96% Margin)",
      "stage1_extractor_prompt": "Parse the raw legacy document / CSV / OCR text into strict schema JSON: {document_date, line_items[{date, description, amount, currency}], control_total_stated}. Temperature = 0.0.",
      "stage2_generator_prompt": "Run deterministic reconciliation: sum(line_items.amount) vs control_total_stated. If delta != 0, output exact row discrepancies without guessing missing digits. Purge raw payload after returning clean JSON/CSV.",
      "stage3_auditor_prompt": "Verify arithmetic checksum: abs(sum(line_items) - control_total_stated) == 0.00. Reject output if any row date or number fails regex validation."
    },
    "BM-03_Compliance_Workbench": {
      "name_zh": "BM-03 · 垂直合规与审计预审工作台（对标 BID-011 Comp AI · 毛利 82%–90%）",
      "name_en": "BM-03 · Vertical Compliance & Audit Readiness Workbench (Benchmark: BID-011 Comp AI · 82%–90% Margin)",
      "stage1_extractor_prompt": "Map uploaded policy documents and evidence logs against the target compliance control checklist (e.g., SOC 2 / ISO 27001 / Industry Audit). Output JSON with {control_id, evidence_found: boolean, exact_quote, source_file}.",
      "stage2_generator_prompt": "Draft a Pre-Audit Gap Matrix highlighting missing controls and exact remediation steps. Include mandatory disclaimer: 'Prepared as automated documentation assistance; requires final review by a licensed CPA/auditor.'",
      "stage3_auditor_prompt": "Confirm every cited control marked 'evidence_found: true' has a verbatim quote from the uploaded source file."
    },
    "BM-04_Credit_Gated_MicroSaaS": {
      "name_zh": "BM-04 · 预付点数包与阈值自动续充工具（对标 BID-016 Cloner · 毛利 75%–82%）",
      "name_en": "BM-04 · Prepaid Credit-Pack & Auto-Recharge Micro-SaaS (Benchmark: BID-016 Cloner · 75%–82% Margin)",
      "stage1_extractor_prompt": "Pre-flight cost estimator: calculate input token count + external API calls required. Deduct [N] credits from user balance BEFORE executing heavy generation. Abort if credits < required.",
      "stage2_generator_prompt": "Execute structured generation within hard max_tokens ceiling. Strip all third-party trademarks and copyrighted assets automatically.",
      "stage3_auditor_prompt": "Log {run_id, credits_deducted, actual_api_cost_usd, net_margin_pct} to Token_Cost_Ledger. Trigger alert if net_margin_pct < 70%."
    },
    "BM-05_Review_Mining_Interceptor": {
      "name_zh": "BM-05 · 竞品差评挖掘与平替截胡引擎（对标 BID-025 / BID-099 · 毛利 80%–88%）",
      "name_en": "BM-05 · Competitor 2-Star Review Mining & Intercept Engine (80%–88% Margin)",
      "stage1_extractor_prompt": "Analyze 1-to-3 star reviews from G2 / Capterra / Reddit / App Store. Group complaints into JSON clusters: {complaint_category, frequency_count, verbatim_user_quotes[], switching_blocker}.",
      "stage2_generator_prompt": "Generate: (1) A 1-feature 'Anti-Bloat' MVP specification that solves the #1 complaint with transparent monthly pricing, and (2) 3 personalized cold-intercept outreach messages.",
      "stage3_auditor_prompt": "Verify all competitor pricing and contract lock-in claims against current official pricing pages."
    },
    "BM-06_B2B2B_Channel_WhiteLabel": {
      "name_zh": "BM-06 · B2B2B 机构渠道白标分润引擎（对标 BID-014 Rezi / BID-024 · 毛利 85%+）",
      "name_en": "BM-06 · B2B2B Agency & Partner White-Label Engine (85%+ Margin)",
      "stage1_extractor_prompt": "Ingest batch client roster from channel partner (CPA firm / law firm / MCN / trade association) and isolate tenant workspace metadata.",
      "stage2_generator_prompt": "Generate white-labeled diagnostic reports branded with the partner agency's logo and CTA, calculating partner rev-share (20%–30%) automatically.",
      "stage3_auditor_prompt": "Verify strict tenant data isolation and confirm 0 cross-client data leakage."
    }
  }
}