Skip to content

Work

We rarely start from zero.

Our extensive catalog of public data gives us something to work with from the start. Each project also leaves behind useful pieces we can reuse. That is why the next hard question usually moves faster.

Contracts. Regulatory filings. Investment books. Bond ownership. Vendor spend. Different questions, same goal: get to an answer you can check and use.

Current work

What we are working on now.

Running means there is a working system. Building means we are still putting it together.

  1. 01 / Running

    2026-08-13

    Procurement

    Which executed statements of work fail our policy today, and why?

    documents → narrow extraction → deterministic score → quoted evidence → portfolio re-score

    A working contract-risk system separates reading from judgment. Change a policy lever and the portfolio re-scores without another model call.

  2. 02 / Building

    2026-08-12

    Banking / community development

    Which communities are underserved, how do true peers compare, and what evidence supports the conclusion?

    five public data families → entity resolution → evaluation corpus → peer evidence → geography-level signal

    The source universe is mapped. The hard part is joining incompatible identifiers, historical geography, lending data, and regulator narratives without losing the as-of date.

  3. 03 / Running

    2026-08

    Insurance / structured finance

    What does this insurer actually hold?

    statutory filings → layout fingerprint → deterministic parse → reconciliation gates → security and property holdings

    The parser accepts a value only when the filing evidence and reconciliation checks support it. Partial and refused parses remain visible instead of being guessed through.

  4. 04 / Building

    2026-07

    Structured credit

    Who holds a given CLO tranche, and what changed this week?

    fund filings + daily holdings + statutory sources → identity graph → holder observations → change signal

    The visible holder base is scattered across incompatible sources. The work is to reconstruct the observable slice and make coverage explicit.

How we work

Different problems. Same basic work.

Find the right sources, make sense of them, show what supports the answer, and turn it into something the team can use again.

For the practitioner

  1. 01Command
  2. 02Reveal
  3. 03Run
  4. 04Keep

Under the hood

  1. 01Source
  2. 02Structure
  3. 03Judgment
  4. 04Receipt

Every problem should leave the practice more capable than it found it.

Past questions

Questions we have worked on.

Some became running systems. Some became reusable pieces. Some simply showed us where the public record stops.

  1. 01Structured financeWho actually holds this bond?
  2. 02ProcurementShould we accept this vendor increase?
  3. 03ContractsWhich statements of work fail our policy?
  4. 04InsuranceWhat does this insurer actually hold?
  5. 05BankingWhich communities are materially underserved?
  6. 06Structured creditWho holds this CLO tranche, and what changed?
  7. 07Commercial real estateWhat does this CRE book most likely contain, property by property?
  8. 08Multifamily financeWhich agency-backed loans cost more than they should?
  9. 09Point-in-time researchWhat could actually have been known on that date?
  10. 10Vendor spendWhat are we paying for that we can reproduce or replace?
  11. 11Public filingsCan this model answer be reconstructed from the record?
  12. 12Data operationsWhich source changed, what did it change, and who needs to know?

Public corpus · 107+ providers · 169+ channels · 811+ tables

A lot of the groundwork is already there.

Public data often gets us most of the way. Add your contracts, positions, invoices, policies, or internal records and the answer becomes specific to you.

Source check

Ask a question, then check the sources.

Loom is our question-answering service, built on public data. Each response shows the source, what changed, and the date the data was true. Ask for evidence and you can open the full source trail.

Source
Public data, traced to source on request
Coverage
Finance and procurement datasets

Source check

What evidence supports the overnight rate?

markets.newyorkfed.org
  • Public fixed-income data

    DealCharts

    A working structured-finance intelligence service built from public filings. Search deals, funds, and BDCs, then inspect the source trail behind each result.

    Source
    Public SEC filings and remittance reports
    Coverage
    Deals, funds, BDCs, and downloadable datasets

    Public SEC filings and remittance reports

    Deals, funds, BDCs, and downloadable datasets

  • Procurement signal

    When does a resin price increase actually stick?

    We built this signal from public data in days. It combines energy prices, producer prices, and industry use to show when an increase may hold. A team's contract data makes it more precise.

    Source
    Henry Hub, BLS PPI, and WTI, all public
    Coverage
    213 scored events, 56.8% directional match

    Henry Hub, BLS PPI, and WTI, all public

    213 scored events, 56.8% directional match

  • Contract risk

    Which contracts would not survive a dispute?

    Executed statements of work, pulled from public SEC exhibits and scored against one policy: pass, review, or fail. Move a threshold and the whole book re-scores instantly. Every verdict carries a receipt back to the filing it came from.

    Source
    Executed SOWs filed as public SEC exhibits
    Coverage
    Policy-scored, with a receipt on every verdict

    Deterministic verdict

    review · 0.55

    quote evidence
    source unchanged
    policy re-score without another model call

  • Holder map

    Who actually holds this bond?

    One class of a live auto ABS deal, traced through public filings. Two funds are named, together about 0.6% of the class. Sponsor risk retention accounts for roughly 5% more. No public filing names a holder for the rest — yet. Every source added closes the gap.

    Source
    SEC 424B5, 10-D, N-PORT, and Reg RR
    Coverage
    One class traced, sources still being added

    SEC 424B5, 10-D, N-PORT, and Reg RR

    One class traced, sources still being added

  • SEC filing tools

    SEC filings, available to AI workflows.

    A bounded SEC EDGAR tool surface that lets MCP-compatible agents query filing metadata inside governed workflows.

    Source
    SEC EDGAR filing metadata
    Coverage
    MCP-compatible clients

    SEC EDGAR filing metadata

    MCP-compatible clients

  • Public MCP server

    Claim in, receipt out.

    Paste in commentary, an article, or text from Claude. The server checks the CMBS claims it supports, links to the SEC filings, and shows what it could not check.

    Source
    SEC ABS-EE and 10-D filings
    Coverage
    519 CMBS deals, latest held filing period per deal

    Public verification surface

    https://receipts.cmdrvl.com/mcpclaude mcp add --transport http receipts https://receipts.cmdrvl.com/mcpcodex mcp add receipts --url https://receipts.cmdrvl.com/mcp

    Connect, call quickstart, then send the text to verify_claims. Call coverage before reading an UNVERIFIED result.

  • Point-in-time benchmark

    What was knowable on that date?

    KnownBy is a point-in-time benchmark for financial systems. It tests whether a system can return the value, citation, and data date that were publicly available then, or correctly decline when the fact was not knowable. Arithmetic and schema validation score each response, and every answer key carries a recipe for rebuilding it from an official archive.

    Source
    Official public archives
    Coverage
    Rates and Calendars, 96-task demo fixture

    Point-in-time check

    What was knowable on that date?

    decline: post-cessation

    96-task demo fixture

Bring us the question that does not fit neatly in a software category.

We will identify the sources, the evidence boundary, and the smallest working result worth building first.

Talk with us