Methodology
How Origentic computes every number.
Origentic's underwriting runs on a single deterministic calculation engine written in code and verified against a frozen suite of test vectors. The same inputs always produce the same outputs, and every figure — in a pro forma, waterfall, sensitivity table, or investment memo — traces back to the assumption or source-document field it came from. This page explains how.
A deterministic engine, not a spreadsheet
The math lives in code, not in cells you can drag or break. A frozen suite of golden test vectors pins the engine's outputs, so a change that would alter a result is caught before it ships. The practical effect: numbers are reproducible and defensible in front of an investment committee.
- •Formulas run in code and reproduce exactly
- •Golden test vectors guard against silent drift
- •No language model performs the arithmetic
Monthly cash flows, annual reporting
Every cash flow is computed monthly and rolled up to an annual pro forma, so lease-up timing, ramp, refinance events, and mid-year changes are captured precisely rather than smoothed into an annual average.
Opt-in institutional structures
The capital stack is modeled the way real deals are structured — senior debt sized by LTV, LTC, DSCR, or debt yield, plus optional junior debt, a mid-hold refinance, funded reserves, sponsor fees, and a preferred-equity tranche — each layer opt-in, so a simple deal stays simple and a complex one reconciles.
- •Debt sizing by LTV / LTC / DSCR / debt yield
- •Preferred equity, junior debt, and mid-hold refinance
- •GP/LP waterfalls with preferred return, promote, and clawback
Glass-box AI, human sign-off
AI is used only to read documents — rent rolls, T-12s, and offering memoranda — and to propose structured inputs. Each proposed field shows its provenance and confidence, and a human must review and approve it before it enters the model. Nothing is auto-applied, so automation never introduces a number no one checked.
Reconciliation and tie-out checks
The engine runs built-in reconciliation checks — sources equal uses, closing equity ties to hold-period equity, the preferred is fully redeemed, and the waterfall conserves cash — so a model that doesn't tie is flagged rather than quietly wrong.
Why teams choose Origentic
Frequently asked questions
- Does AI calculate the returns?
- No. AI only reads documents and proposes inputs, which a human approves. All underwriting math — pro formas, waterfalls, sensitivity, and returns — is computed by a deterministic engine, so no figure in the model is produced by a language model.
- How do you know the numbers are correct?
- The engine is deterministic and pinned by a frozen suite of golden test vectors, and every model runs reconciliation checks (sources equal uses, equity ties, waterfall conserves cash), so errors surface rather than hide.
- Can I trace a number back to its source?
- Yes. Every figure derives from an assumption or an AI-extracted field that carries its provenance, so you can trace any output back to the input or source document it came from.
Keep exploring