What Computation Provenance Means for Lending

A credit committee is reviewing a $50 million facility request. The dossier shows a debt-service coverage ratio of 2.61x. A committee member asks: “Where does the income number come from?”

In most institutional lending workflows, the answer requires the analyst to open a separate spreadsheet, find the cell, trace it to a tab that references a PDF, open that PDF, and locate the line item. If the analyst built the model two weeks ago, they may need to reconstruct the logic. If a different analyst built it, the reconstruction may be impossible without starting over.

This is the provenance problem in lending. Every number in a credit analysis is a derived figure — computed from inputs that were extracted from documents that were provided by the borrower. The chain from source document to final metric can pass through five or six transformations. When any link in that chain is opaque, the entire analysis becomes a matter of trust rather than verification.

Computation provenance solves this by making every link in the chain explicit and traceable.

In practice, this means: the DSCR of 2.61x is computed from total recurring income of $10,748,000 divided by debt service of $4,123,700. The income figure is the sum of six sources — GRAT annuity distributions ($3,360,000, sourced from the trust accounting statement, page 3), family trust distributions ($2,840,000, sourced from the brokerage statement, page 4), and four other sources each traced to specific documents and page numbers. The debt service figure includes the proposed facility interest ($3,500,000, computed from $50M at 7.0%) and the existing bank facility ($623,700 annual P&I, sourced from the credit agreement, Section 2.3).

Every figure in the waterfall links back to a specific document, a specific page, and a specific element on that page. The credit committee member who asks “where does this come from?” gets an answer in seconds, not minutes. The compliance officer reviewing the file six months later can reconstruct every computation without the original analyst’s involvement.

This capability fundamentally changes the relationship between AI-generated analysis and institutional trust. The concern with AI in lending has always been the black box problem — if an AI model produces a DSCR, how does the credit committee know it’s correct? Computation provenance eliminates this concern by making the AI’s work fully auditable. The model is not asking to be trusted. It is showing its work.

This is what happens when proprietary institutional knowledge — encoded in the Kalynto Genome — is paired with frontier AI. The AI has the reasoning capability to extract, cross-reference, and compute across 20 documents simultaneously. The Genome has the domain knowledge to know what to extract, how to cross-reference it, and which computations matter for each borrower archetype. The provenance chain is the output of both systems working together — not a feature bolted on after the fact.

Kalynto embeds computation provenance throughout every surface of the platform. The lender dossier, the Excel credit model, the PowerPoint deck — every metric in every format traces to its source. When a lender drops the Excel model into Copilot, Claude, or ChatGPT and asks “where does this DSCR come from?”, the AI tool can follow the provenance chain to the source documents because the chain is embedded in the artifact itself.

This is what institutional-grade AI looks like. Not smarter models producing more impressive numbers. Proprietary knowledge paired with frontier AI, producing numbers that can be independently verified — every step of the way. And the provenance system compounds: with every deal, the extraction patterns that feed the provenance chain are validated and refined.