The evidence layer for private markets.
Capital Refinery turns the documents in your data room into a decision-grade record — every KPI cited, every covenant bound, every IC anchor preserved. The record stays testable as the deal moves. Twelve months later, you can prove what you said, what changed, and why.
We are not a dashboard. We are not an LLM agent. We are the layer underneath both — the one that makes “trace this number” a one-click answer instead of a forensic project.
Three things that get easier the moment you have an evidence layer.
IC-grade output on day one.
Send the deal materials. We return a lane-aware IC memo, a financial workbook, an executive report, and a KPI summary — each carrying an evidence fingerprint. Every figure on every page traces back to the page in the document it came from.
Drift surfaces against the underwrite, not against last quarter.
When new evidence lands, we re-run the record and show you what moved relative to what was originally underwritten — not just relative to the prior period. Covenant cushion, customer concentration, lease expense, add-back quality. If the thesis broke, you see exactly which assumption broke and where the proof is.
Anchored to what you knew when you knew it.
The IC anchor is preserved. Every export carries the fingerprint of the evidence state it was built from. When an LP asks, an auditor reviews, or a portfolio committee second-guesses — the answer is a downloadable record, not an argument from memory.
We don't ask you to trust us. We ask you to download the proofs.
Three artifacts on this site, all downloadable, no login required, no email gate.
Three deal packs.
Pre-close IC, post-close watch, post-close add-back exposure. Each pack contains every KPI with its source citation, every covenant, every evidence document the platform read. Open kpis.csv, follow the citation, the number matches. We dare you to ask the same of any other vendor in the category.
A signed sample pack.
An IC memo, an executive report, a financial workbook, a KPI summary — produced by the live pipeline against a composite portfolio position. Every artifact carries an evidence fingerprint. The fingerprint is reproducible: run the pipeline twice on the same evidence state, get the same fingerprint. That is what makes the record defensible.
Open the signed pack →A diagnostic on your deal.
Send a CIM, a credit agreement, a quarterly pack. We return a diagnostic in 2–4 hours — same shape as the packs above, against your real evidence. Under NDA. No obligation.
Run the diagnostic →The category is crowded. The differences are not subtle.
Not an LLM agent.
Agentic-AI investing tools are loud right now. They demo well; they fail in production where conservation arithmetic, subject identity, and lane shape decide the outcome. Our output is reproducible byte-for-byte from the same evidence state. Theirs is not.
Not a dashboard.
Dashboards show you what is happening. They do not show you whether what is happening is what you said would happen. Our surfaces are anchored to the original underwrite — drift is always measured as a delta, not as a snapshot.
Not a data provider.
Pitchbook, Allvue, DealCloud and the rest are valuable for what they cover. They are not designed to bind a specific IC decision to a specific evidence state and keep that bond testable through a hold period. We are.
Not a workflow rewrite.
We do not ask your team to abandon the spreadsheet, the memo template, or the IC ritual. We sit underneath those artifacts and make them defensible. The spreadsheet is not the enemy. The spreadsheet without an evidence layer is.
Fifteen minutes. No call. No deck.
Open the three deal packs. Pick a row in kpis.csv. Follow the evidence_citation column to a specific cell or section in evidence/. The number matches. Repeat for any other row.
That is the entire evaluation. If it works for those three deals, it will work for yours. Bring us a CIM and we will show you on your data, under NDA, in 2–4 hours.
Don't trust the dashboard. Trace the numbers yourself.
Three downloadable deal packs across the lifecycle. Every KPI cites its source cell. Open kpis.csv, follow the citation, the number matches. No app, no login required.