Decision IntelThe people presenting the deal also write the case for it
That is not a flaw in private equity. Somebody has to compress ten thousand pages into a decision. But it means individual facts get challenged hard while the architecture of the reasoning does not: which assumptions are load-bearing, which failure mechanisms fire together, and where the committee’s confidence exceeds its evidence.
The three governed records
Every audit produces forwardable records bound to the filing they came from, so a reader who was not in the room can check the reasoning rather than take it on trust.
The Decision Brief
Two pages a committee can forward. The dependency the deal is betting on, the structural risks in order, and the protections the document does not contain.
The Outcome Record
The audit graded against what happened, with misses carried at the same weight as hits. A record that only counted its wins would not be worth keeping.
Decision Provenance Record
The full evidence dossier: verbatim citations, the cohort the deal was matched against, and a hash chain bound to an independent timestamp you can verify without us.
RFC 3161 TOKEN: third-party attested
COHORT: 1,307 registrations
The one test that costs you nothing to run
Submit a closed deal blind and compare the day-one read against what actually happened. It asks for no forecast and no faith — only that you read the past.
MODE: blind · name-blinded · isolated
STATUS: outcomes withheld
Retrieval off, names redacted, no account memory. Three independent toggles, each verifiable on the artifact itself.
The structural layer is deterministic, so the same document produces the same read every time it is run.
You already know the answer, so you are grading us rather than trusting us. That is the whole point of starting here.
What happens to your document
Written to be checked rather than believed. Each claim below is one a security reviewer can verify, and the last one they can verify without us entirely.
Zero-retention routing
Requests are served only by providers holding zero-retention agreements. If none can serve a model, the request fails rather than quietly falling back.
Never used for training
Your deal data is routed only to no-training provider tiers and does not enter any training corpus.
Stripped before inference
A GDPR and NDPR anonymisation layer removes personal data before any model call. AES-256-GCM at rest with versioned key rotation.
Verify without us
A SHA-256 chain binds sources, claims and delivered bytes, and an RFC 3161 token from an independent authority is stored whole.
Before you sign
The questions deal teams and general partners actually bring before capital moves. Select a segment to read the specification behind each answer.
How do you guarantee proprietary deal theses and confidential CIM materials never leak or train a model?
Engagements run with zero-data-retention model routing enforced: requests are served only by providers holding zero-retention agreements, and if no such provider can serve a model the request fails rather than silently falling back. Your deal data is never used to train any model. Personal data is stripped by an anonymisation layer before any model call, and documents are encrypted at rest so the audit stays re-readable by you and nobody else.
Start with a deal you already know the answer to
A blind audit on a closed transaction costs you one document and tells you, in your own terms, whether this would have told you anything your process did not. That is the honest way to evaluate it.