CMD+RVL FOUNDRY

For financial analysts

One document is chat.
The whole corpus is a tournament.

Claude can answer one document. Foundry takes that same question, runs it across all the documents, and turns the answers into a dataset with proof.

Foundry is not a better way to rip one document. It is a way to keep doing it across all of them.

Representative data. Public names withheld.
Pool 01BDC & SEC FILINGS4 tournaments / BDC / Holdings / Footnotes
Pool 02CMBS FINANCIALS4 tournaments / CMBS / Financials / Commentary
Pool 03LEASE & APPRAISAL DOCS3 tournaments / Lease / Appraisal / Terms
3tenant pools
11live tournaments
1running now
2xlsx ready
Tournament running nowFootnotes428 docs / proof attached
Ask once

Start with one analyst question such as holdings, footnotes, rates, or lease terms.

Use AI to build

Use models to help build and improve routes, not to answer every row on every run.

Run at scale

The winning route runs across the corpus, works in zero-retention environments, and keeps cost and latency down.

Chat works on one file.
Analysts need the answer across all of them.

01

A single-document answer is useful, but it is still a one-off.

02

Re-asking a model on every file is slower, more expensive, and harder to trust at scale.

03

A tournament builds a reusable route so the answers can land in a dataset and compound over time.

One question.
One tournament.
One compounding dataset.

01
Ask onceChoose the question

Pick one answer you want back from the corpus: holdings, footnotes, rate terms, lease terms, or another document-bound field.

02
Run the tournamentModels help build routes

Several routes try to answer that same question across the same private corpus without forcing a model to do all the work every time.

03
Keep the winnerSave result and proof

The winning result lands with its status, its proof, and a route you can run again.

04
Let it compoundOld and new files

Once the route is good, run it across older files, keep using it on new files, and keep building the dataset.

Start with one question.
End with a dataset.

What are the current holdings?

Run a tournament over filings and schedules to surface row-level holdings.

Which rows have footnotes?

Attach footnote markers to the exact rows analysts care about.

What is the rate and spread?

Break debt rows into reference rate, spread, floor, and coupon.

What are the lease terms?

Pull term, rent, escalators, and allowances from lease documents.

See the same question running across a corpus.

Anonymized tournament registerOpen tenant view
TournamentQuestionStatusUpdatedExport
BDC HoldingsHoldingsready2h agoXLSX
BDC Footnote MarkersFootnotesrunning4m ago
BDC Rate ComponentsRate Termsneeds review1d ago
BDC Obligation StateStateready6h agoXLSX
Current specimen
Document setSEC 10-Q / SOI inline markers
Docs428
Fields2,107
Last run2026-03-13 17:58 UTC