CMD+RVL — Structured finance data intelligence

We turn primary sources — SEC filings and the disclosures behind structured finance deals — into machine-readable data with the provenance kept intact, so every number can be traced back to where it came from.

The datasets on this page are open samplers of that work. They're built to be downloaded, modeled, and pulled apart. The full, continuously-updated data that backs them is available on the Snowflake and Databricks marketplaces — these Hugging Face releases are the taste, not the whole table.

Explore the data live → dealcharts.org

DealCharts is our public window into structured finance: thousands of CMBS deals, mapped funds, and BDCs, each backed by the SEC filings it came from. Ask Cairn anything about a deal, fund, or entity and get an answer with its sources. Start there — it's the fastest way to see what this data can do.

On Hugging Face

Available now — CMBS special-servicing early warning. A leakage-safe population framed around "which CMBS assets are at elevated risk of transferring to special servicing within the next 12 months," using only point-in-time information from SEC filings, in three shapes:

• cmbs-special-servicing-transfer — flat supervised table for tabular classification, AutoML, XGBoost, and tabular foundation models like TabPFN and TabFM.
• cmbs-special-servicing-sequences — the same population as ordered per-asset sequences, for sequence and time-series models.
• cmbs-special-servicing-timeline — a text companion with a natural-language timeline per observation, for LLM prompting, NLP, and text classification.

What's next. This is one use case in one asset class. More CMBS use cases, Auto ABS, CLO, and other structured finance datasets are on the way.

How to use these

Treat them as starting points, not answers. CMBS is hard, and these tables are unlikely to be predictive on their own — even cleaned up. They're more interesting as the shape of a problem: a defined population, an honest temporal split, and traceable provenance to build on. Stitch your own signals onto the spine — rates, spreads, macro, property-level data, your own features — and see what actually moves the label.

The point of the provenance is that you never have to take a number on faith: every observation carries the SEC filing and reporting period it came from, so you can trace and check any value yourself. We're not claiming the labels are clean or the feature set complete — we're claiming they're auditable. That's the same discipline we bring to client work: finished results with proof that survive without us.

Terms

License: CC-BY-NC-4.0 — open for non-commercial use. For commercial use, get the full data on Snowflake or Databricks, or reach us at cairn@cmdrvl.com.

dealcharts.org · Snowflake Marketplace · Databricks Marketplace · cmdrvl.com · cairn@cmdrvl.com