{"data":{"kind":"file","path":"README.md","version_id":"olb498drlwziwbrt2evaua61","entry":{"name":"README.md","path":"README.md","is_directory":false,"size":2725,"modified_at":"2026-08-28T12:54:16.206000","content_hash":"4551fbb468fc06614abda5dd742baf8d0ba7dcb20962b4d7bb227c3d491c2239"},"entries":[],"content":"# Real-estate Listing Analyzer & Matcher\n\nA comprehensive environment for evaluating LLM capabilities in real-estate domain reasoning, including property analysis, buyer-seller matching, comparative evaluation, and negotiation strategy.\n\n## Task Types\n\n### 1. Listing Analysis (8 tasks)\nGiven a property listing, the model must provide:\n- Price analysis (price/sqft, vs median, rental yield)\n- Strengths and red flags identification\n- Investment metrics (cap rate, cash-on-cash return)\n- Overall recommendation with confidence level\n\n### 2. Buyer-Seller Matching (4 tasks)\nGiven a buyer profile and 8 available listings, the model must:\n- Rank all listings by fit\n- Check deal-breaker compliance\n- Identify must-have and nice-to-have matches\n- Recommend top 3 with reasoning\n\n### 3. Comparative Analysis (4 tasks)\nGiven a buyer and 3 properties, the model must:\n- Build a side-by-side comparison table\n- Analyze financial implications\n- Assess lifestyle fit and risks\n- Make a final recommendation\n\n### 4. Negotiation Strategy (4 tasks)\nGiven a buyer and target listing, the model must:\n- Analyze market position and leverage\n- Recommend offer price with reasoning\n- Develop specific negotiation tactics\n- Create contingency plans\n\n## Reward Functions (5 metrics)\n\n| Metric | Weight | Description |\n|--------|--------|-------------|\n| Format Completeness | 0.25 | All required sections present with substantive content |\n| Numerical Accuracy | 0.20 | Correct calculations (price/sqft, yields, percentages) |\n| Red Flag Detection | 0.20 | Identifies risks: flood zones, assessments, structural issues |\n| Reasoning Quality | 0.20 | Evidence-based, logical, specific, balanced analysis |\n| Buyer Alignment | 0.15 | Recommendations match buyer needs and constraints |\n\n## Dataset\n\n- 8 diverse property listings (single family, condo, duplex, townhouse)\n- 4 buyer profiles (family, investor, luxury, first-time)\n- 20 total evaluation tasks across 4 task types\n- Realistic Austin, TX market data with nuanced red flags\n\n## Anti-Gaming Design\n\n- **Multi-signal scoring**: Each reward checks multiple independent conditions\n- **Evidence requirement**: Must cite specific numbers from listings, not generic advice\n- **Red flag traps**: Listings contain subtle issues (reserve funding %, special assessments, flood zones) that require careful reading\n- **Buyer-specific logic**: Same listing can be good for one buyer and bad for another\n- **Numerical verification**: Calculations must be within tolerance of ground truth\n\n## Usage\n\n```python\nimport verifiers as vf\n\nenv = vf.load_environment(\"real-estate-listing-analyzer\")\n# or\nfrom real_estate_listing_analyzer import load_environment\nenv = load_environment()\n```\n\n## License\n\nMIT\n","encoding":"utf-8","truncated":false,"total_bytes":2725},"status":null}