Open underwriting interoperability standard

`.uw.md` is the canonical data contract for underwriting systems.

AI-ready. Deterministically computed. Verifiable to the digit.

UW Markdown gives AI agents, calculation engines, internal platforms, and any compatible editor one stable underwriting record to exchange and extend. AI never does the financial math: every NOI, DSCR, IRR, and waterfall split is recomputed by deterministic engines, attested by verification receipts, and portable into any data lake.

CANONICAL RECORD Deal facts + provenanceDETERMINISTIC Validation + calculationVERIFIABLE Receipts + signatures + digestsREPRESENTATIONS Markdown / JSON / XML / CSVINTEROPERABLE Agents / tools / data lakes

One canonical record

The same underwriting data works everywhere.

Open complete example

Parkview-Apts-Glendale-AZ.uwx.mdVALIDCOMPUTED

Canonical record

portable source

---
uw_version: "2.0"
deal_name: "Parkview Apartments"
asset_class: "multifamily"
quick_metrics:
  purchase_price: 7200000
  noi_underwritten: 396635
  dscr: 1.109
  ltv: 0.70
---

## Underwritten NOI

Vacancy is normalized to 7%. Management
fees use the new agreement at 5.5% of EGI.

```json uw:section=noi_model
{
  "effective_gross_income": 630936,
  "total_operating_expenses": 234301,
  "net_operating_income": 396635
}
```

Compatible consumer

underwriting view

Parkview Apartments

48 units · Glendale, Arizona

Purchase price$7,200,000

UW NOI$396,635

DSCR1.109×

LTV70.0%

Operating statement	T-12	UW
Gross potential rent	$655,200	$655,200
Vacancy / credit loss	($32,760)	($45,864)
Effective gross income	$640,440	$630,936
Total operating expenses	($228,344)	($234,301)
Net operating income	$412,096	$396,635

Deterministic calculation complete
No AI arithmetic. Inputs resolved by the multifamily calculation pack.

FORMAT 2.0 · analyst tier · append-only historyOpen in reference viewer

What the format provides

A durable interchange layer for underwriting.

Code-native

Typed sections, schemas, and a 377-assertion conformance corpus make the record dependable software input.

AI-ready

Agents extract data and write narrative against one shared contract — and never do the financial math.

Verifiable

Capital stacks, cash-flow series, and waterfalls are stated in the document and recomputed in full — verifiers never trust the stated numbers. Receipts and signatures carry the proof.

Interoperable

The same record loads in any compatible platform, converts to JSON, XML, and CSV, and flattens straight into a data lake.

How it works

One record in. Any compatible underwriting system out.

Build your first file

1
Ingest the record
Any compatible tool, service, or agent loads the same .uw.md deal.
2
Parse shared facts
Structured sections give every consumer a stable underwriting vocabulary.
3
Validate and compute
Conforming hosts enforce rules and run deterministic calculation packs.
4
Preserve the record
Changes supersede prior blocks without destroying provenance or history.

Standard and reference tools

Open rules, practical implementations.

Format specificationNormative structure of a .uw.md filev2.0ProtocolRequirements for readers, editors, calc hosts, and agentsv2.3.0Conformance corpusFixtures that prove implementation behavior377 assertionsCore library and CLIParse, validate, calculate, verify, convert, and renderv2.3.0 · npmData-lake on-rampDeals to DuckDB with digests and receipts intactguideReference viewerRead a file locally in a browserTier 1Reference editorEdit, validate, calculate, diff, and exportpublic preview

Start with the artifact