Underwriting that survives close.
Most underwriting software is built to get the deal team to IC approval. Capital Refinery is built to keep the underwriting alive across the life of the position — so the assumptions the team relied on at entry remain testable against the operator data the firm is already collecting.
What underwriting tools say they do, and what actually happens.
The standard workflow ends at IC approval. The structural problem starts the day after.
Three things the dominant tools do not ship.
A structured investment record at IC
Not a Word document. A typed object.
Thesis, assumptions, conditions, KPIs — bound at IC approval as separate, individually-testable rows. From the moment the wire clears, the platform has a reference point against which every operator update can be tested. Decision validity stops being a memory exercise and starts being a property the platform can compute.
Deterministic-first extraction with provenance
Every figure traces back to the source page.
A regex-and-constructor sweep produces candidate figures from the CIM, the financials, and the credit agreement. A small local LLM adjudicates conflicts. Every number has a candidate ID and a provenance trail. When the assumption gets re-tested in year three, the team can trace it back to the line in the source document — not to the LLM that drafted the model.
A continuous binding from operator data to assumption
Quarter after quarter, for the life of the position.
Each accounting feed and KPI report is mapped to the entry assumption it tests. The connection is structural, not generated. When conditions change enough that the original IC decision is no longer defensible, the platform surfaces it — with the assumption that broke and the time-to-consequence ranking against the rest of the portfolio.
Where the lines fall.
| Capability | Generic underwriting tools | Capital Refinery |
|---|---|---|
| Financial extraction from CIMs and statements | LLM-first, citation-style provenance | Deterministic-first with candidate IDs and source-page provenance |
| Draft model generation | Yes — strong in the new AI category | Yes — but as input to the structured record |
| Structured investment record at IC | No — output is a model file or a Word memo | Yes — bound at approval with thesis, assumptions, conditions |
| Operator data → entry assumption mapping | No — handoff at IC | Yes — the data model |
| Decision validity scoring across the position | No | Yes — live across the life of the deal |
| Covenant + breach probability for credit positions | No — not in the underwriting layer | Yes — for credit positions |
| Time-to-consequence ranking | No | Yes |
A walkthrough of the workspace where the IC memo is a structured object, the assumptions are typed rows, and every figure traces back to the source page in the diligence pack. The same workspace stays alive after close.
The research behind this.
- The Decision Integrity Gap
Why the underwriting layer in private capital does not survive close — and what would have to be true for it to.
Read → - Your Data Room Dies at Close
The most analytically dense work the firm did on this asset becomes inaccessible the moment it matters most.
Read → - What Is an IC Memo, and Why Does It Go Stale?
The structural reason every IC memo dies the day after approval — and what to do about it.
Read →
Bring us the underwriting from a deal you closed last year.
The fastest way to feel the gap is to compare what your current underwriting tool produced for the IC pack to what a structured investment record looks like in year three. Real document, real position, 60–90 minutes.