Datasets · Q&A · JSON · Re-checked
Open records. Open JSON. Check the math.
A small published sample from our commercial real estate question-and-answer dataset, shown the way we think datasets should be shown: the question, the answer, the raw record, and an independent re-check.
Read this first. These records are model-generated training data, not extracted from real transactions. Property and tenant names are synthetic. The sample is a curated slice, not the whole corpus.
The cook
810,097 rows in. 706,915 kept as CRE training rows.
Every row passes through gates before it’s kept. Rows that fail, duplicate another row, or aren’t CRE are split out instead of silently mixed in. Counts below are from the dataset card.
- In-domain CRE706,915
- Eval split3,539
- Off-domain98,731
- Quarantined912
- Underwriting271,094
- Comp analysis102,252
- Valuation92,990
- Tax assessment74,913
- Rent roll56,519
- Debt analysis42,190
- Capital markets25,818
- New economy19,911
- 1031 exchange18,665
The samples
Open a record.
10 cold-storage records from the eval split. Real company names excluded. Switch between the Q&A view and the raw JSON, or download the whole sample.
- grade genesis
- score 95
- length short
- json n/a
- concepts rich
- numeric rich
- degenerate pass
- relevance high
System prompt
You are SwarmCRE. Perform the requested underwriting calculation. Show all steps. Round money to the nearest dollar, ratios to 2 decimal places. For data center deals, account for power costs ($/kWh), PUE, and per-MW economics. When economic incentives apply (TIF, abatements, OZ), model their impact on returns. For 1031 exchange calculations, track adjusted basis, deferred gain, boot, and new basis. For tax analysis, apply correct depreciation schedules (39-year commercial, cost segregation), capital gains rates (20% LTCG + 3.8% NIIT + 25% Section 1250 recapture), and entity-level impacts. State assumptions explicitly.
Perform a complete underwriting analysis for Interchange Yards, a 93,000 SF cold storage facility in Airport, Salt Lake City. Rent roll: - Patriot Packaging (Entire Building): 93,000 SF at $6.58/SF, modified_gross, cpi escalation, expires 2033-12-01 Vacancy: 7.7% Expenses (per SF): property tax $208,320, insurance $39,990, CAM $87,420, utilities $115,320, reserves $17,670 Management fee: 4.6% of EGI Cap rate: 7.43% Debt: SBA 504 — $712,515 at 6.11%, 25-year amort, 0-year IO, 25-year term Calculate all metrics: PGI, EGI, NOI, value, price/SF, annual debt service, DSCR, debt yield, LTV, equity, and cash-on-cash return.
REVENUE PGI: $611,940 Vacancy loss (7.7%): $47,119 EGI: $564,821 EXPENSES Management fee: $25,982 Total OpEx: $494,702 NOI: $70,119 ($0.75/SF) VALUATION Cap rate: 7.43% Value: $943,728 Price/SF: $10.15 DEBT METRICS Loan amount: $712,515 Annual debt service: $55,665 DSCR: 1.26x Debt yield: 9.84% LTV: 75.5% RETURNS Equity: $231,213 Cash-on-cash: 6.25%
Re-check recomputed here from the rent roll in the question
| Line | Stated | Recomputed | Result |
|---|---|---|---|
| PGI | $611,940 | $611,940 | ✓ matches |
| Vacancy loss | $47,119 | $47,119 | ✓ matches |
| EGI | $564,821 | $564,821 | ✓ matches |
Provenance
What this is, and isn’t.
- Model-generated training data (lineage gen_model: various). Not extracted from real transactions.
- Eval-split rows; cold-storage facilities; real company names excluded; score >= 89; one per city.
- The re-check recomputes revenue, vacancy loss and effective gross income from the rent roll in each question, and flags negative values. It doesn’t validate every line of every answer.
- No license has been chosen for these samples yet. Treat them as view-only until one is published in the repository.
Found a bad record? Good. Tell us.
Corrections, better checks, and real cold-chain examples all make the data worth more.