Capital Refinery
About Capital Refinery

Built for the moment a prior decision stops being correct.

We are a small, deliberate team building decision integrity infrastructure for private markets — the layer that keeps the IC thesis alive after the deal closes.

Why this exists

Private markets software has spent fifteen years getting better at collecting data. CRMs track the relationship. Data rooms hold the diligence. Monitoring tools render the KPIs. Accounting systems close the books.

What none of them do is govern the investment decision after it is made. The thesis that closed the deal lives in a memo on a shared drive. The assumptions that justified the price are not testable. The conditions the team relied on never get checked against the live operator data.

$3.6 trillion in unrealized value sits on GP balance sheets — and the quarterly reconstruction exercise to check it is still done by hand.

Bain & Company, 2025 Global Private Equity Report

What we are building

Capital Refinery is a structured record of every IC decision, alive across the life of the position. The thesis, the assumptions, and the conditions are stored in a way the platform can re-test continuously against every operator update. When the conditions stop being met, the team finds out — with the time still left to act.

Underneath it is a deterministic-first extraction kernel, a financial truth engine with full provenance, and a closed operating loop from accounting to action. None of it relies on a chat window over your files.

How we work

We are in private access. Every engagement starts with a real deal from your portfolio — your CIM, your credit agreement, your model — and a working session where we walk back the decision validity together. No demo data. No procurement loop. If the platform does not survive that session on your own position, it should not be in your stack.

Principles

What we will and will not build.

01 / 03

Deterministic before generative

The kernel is regex sweepers and deterministic constructors.

A small local LLM adjudicates between candidates the deterministic layer already produced. The model never invents a KPI. Every figure has a method, a candidate ID, and a provenance trail back to the page it came from.

02 / 03

Decisions, not dashboards

We measure the platform by whether it changes a decision.

If the team sees a screen and does not act differently than they would have without it, we cut the screen. Visibility is not the goal — decision validity is.

03 / 03

Earned, not promised

Private access is the entire sales motion.

We do not run a procurement loop. We do not ship a sandbox. We work from your real artifacts and earn the seat one position at a time. If the loop fails on your portfolio, that is the answer.

Bring us a deal.

The fastest way to know whether decision integrity is real is to see it on a position you already hold.