{"data":{"kind":"file","path":"README.md","version_id":"z0ybrx8l8ihya3k04wtrequd","entry":{"name":"README.md","path":"README.md","is_directory":false,"size":3526,"modified_at":"2026-06-22T18:28:34.711000","content_hash":"bd82b4036c3e8305549dc482ce0d2e096e188c534068f04e3d83522d5600d6b2"},"entries":[],"content":"# Nexum Deal Debt Scenario Update\n\n## Overview\n\nThis environment is a commercial real estate diligence benchmark built from an anonymized property-level deal workflow. The main task asks an agent to review a native Excel underwriting model and a PDF debt quote summary, update the model for three bridge-financing scenarios, and decide whether the deal still clears a 20% levered IRR hurdle at a revised $132 million purchase price.\n\nThe environment is designed to test realistic financial-document reasoning rather than simple text extraction. The difficult part is selecting the correct floating-rate bridge-loan assumptions from a messy PDF extraction and applying them to the model.\n\n## What Is Exposed\n\nPackaged source files:\n\n- `data/native/Amazon Delaware Debt Guidance 3-21-22.pdf`\n- `data/native/Amazon Middletown - AREP Run -3.15.2022 (Base Case).xlsx`\n- Selected anonymized markdown deal documents under `data/markdown/`\n\nThe environment does not package prior eval outputs, scratch workbooks, virtual environments, or gold-source iteration workbooks.\n\n## Main Task\n\nThe principal eval row asks the model to:\n\n- Update purchase price to `$132,000,000`.\n- Use Section I, Initial Floating Rate Bridge Loan, from the debt guidance PDF.\n- Ignore Section II, Refinance.\n- Use the high end of each quoted range for Life Co Financing, Bank Financing, and Higher LTV Financing.\n- Run the scenario workbook tool.\n- Return levered IRR for each quote, rounded to the nearest hundredth of a percent.\n- State which quote, if any, clears 20% levered IRR.\n\nCorrect scenario outputs:\n\n| Scenario | Levered IRR | Decision |\n| --- | ---: | --- |\n| Life Co Financing | `19.57%` | Does not clear 20% |\n| Bank Financing | `19.71%` | Does not clear 20% |\n| Higher LTV Financing | `20.51%` | Clears 20% |\n\n## Why This Is Hard\n\nThe PDF extraction is intentionally column-oriented and messy. Strong models usually need to distinguish:\n\n- Floating-rate bridge financing from fixed-rate guidance.\n- Section I bridge loan assumptions from Section II refinance assumptions.\n- SOFR today from all-in rate today.\n- Lender spread from lender/origination fee.\n- Basis points as decimal model inputs.\n\nCommon failure modes observed in local benchmarking:\n\n- Using fixed-rate lender-fee rows for Life Co or Bank.\n- Treating `0.30%` SOFR as the lender fee.\n- Treating `2.80%` all-in/rate-today text as SOFR.\n- Averaging ranges instead of using the high end.\n- Producing a directionally plausible final answer after running the wrong model assumptions.\n\n## Tools\n\nThe environment exposes document tools and a scenario workbook tool:\n\n- `read_debt_guidance_pdf`\n- `inspect_original_debt_model`\n- `run_debt_quote_scenarios`\n- `search_document`\n- `read_document`\n\nWhen Windows Excel automation is available, the scenario tool recalculates copied workbooks directly. On hosted/Linux infrastructure where Excel COM is unavailable, it uses a deterministic fallback calibrated to the validated workbook outputs so the benchmark remains runnable.\n\n## Rubric\n\nThe native debt scenario task is scored as:\n\n| Component | Weight |\n| --- | ---: |\n| Correct scenario assumptions passed to the workbook tool | `45%` |\n| Correct final levered IRRs | `25%` |\n| Correct pass/fail decision | `20%` |\n| Source citations | `10%` |\n\n## Local Run\n\n```bash\nprime --plain eval run nexum_deal_folder_test --env-dir-path . --skip-upload -n 1 -r 5 -c 1 -m google/gemini-2.5-pro -t 5000 -T 0 --save-results --debug -A\n```\n\nFor public PI runs, omit `--skip-upload` as appropriate.\n\n","encoding":"utf-8","truncated":false,"total_bytes":3526},"status":null}