Nabu — Credit Matrix & Portfolio Synthesis Engine

Most construction lending platforms treat credit policy and fund economics as separate conversations. This model unifies them into one directional flow: a change to the box propagates all the way to investor returns.

↻  Performance feeds reinvestment and AUM growth, closing the loop back to origination volume.

Every factor is quantified, bucketed into scored tiers, and carries a probability distribution that shapes what the book looks like at scale. The bars show each factor’s tier distribution: the deliberate shape of the target portfolio. The percentage in each card is that factor’s weight in the composite risk score.

Loan-to-Cost leads at 14.0%, ahead of Loan-to-Value at 8.5%. In construction lending, cost basis governs risk more than appraised value: the money going into the ground is what you can lose. Reshaping any distribution deliberately reshapes the book, and the effect flows straight through to yield.

Asset-type multipliers apply on top of base pricing to reflect inherent product risk. Ground-up construction prices in execution, timeline, and draw complexity that stabilized perm does not carry.

These distributions are targets, not fixed weights. The synthesizer applies deliberate wobble to the Monte Carlo draws, so the realized book varies around each target the way a real portfolio does rather than snapping to an exact percentage. That variance is what makes the 10,000-loan output behave like an actual book instead of a spreadsheet average, and it is where cushion and contingency get shaped in: you can widen the wobble on the factors you are least certain about and tighten it where your policy is firm.

This is the closed loop made operational. Score a loan against the mandate, and the composite weighted score maps into the pricing ladder to produce the term sheet: margin, origination fee, and required guaranty. No back-and-forth between origination and credit. Once it’s scored, the LOI is priced.

Nine risk bands map score to margin, origination fee, and a guaranty requirement that escalates from elective to full personal guaranty. The band your loan falls into is highlighted as you score it above.

The mandate and synthetic portfolio flow into a full fund model. The commitment base builds, peaks, and winds down: the “so what” of the entire engine.

The demo runs on assumed distributions. The point of the architecture is to replace them with your actual loan tape, so the model reflects the book you really originate rather than a theoretical one.

Calibration also surfaces where your current guidelines produce a different book than you intended, and it is where cushion gets deliberate. Fit the distributions to what you have done, then widen the wobble to build in contingency on the factors that scare you and tighten it where your policy is firm. The credit committee can test a change, say loosening LTC on ground-up from 80% to 83%, and watch the score distribution, weighted yield, and expected loss move before a single guideline goes live.