Multi-source property scoring, designed to be defended.
A walkthrough of how we design decision intelligence systems — seven independent data signals, a configurable weighting layer, and a deterministic recommendation a stakeholder can stand behind in any review.
A walk score is a 0–100 lifestyle metric. A crime rate is incidents per 1,000 residents. A rent-delta is a market signal in dollars. A school rating is categorical.
None of these live on the same axis — and yet they all feed the same decision. The system below is one way to bring them together honestly.
Blunt averaging hides the trade-offs.
A black-box model obscures the policy — which is exactly the part a stakeholder needs to see.
What people actually need is a system where the inputs are visible, the weighting is a policy decision, and the recommendation can be defended.
Seven signals, normalized to one scale.
Each source adapter handles one external signal — rate-limited, cached, and tolerant of partial data. Normalizers convert raw values (a $400/mo rent delta, a 12-point walkability shift) to a comparable 0–100 scale.
The seven signals to the right are the property-investor configuration. The weights are an answer, not the answer — an investor focused on cash flow uses different weights than a housing analyst focused on affordability, and the same engine serves both.
Composed of small, named, replaceable parts.
Sources feed adapters, adapters feed normalizers, normalizers feed the scoring engine, and the engine reads weights from a versioned config to produce a composite score. The decision layer maps that score to an action band.
Every stage is replaceable. Adding flood risk or climate exposure means writing one adapter and one normalizer — no retraining, no model versioning, no statistical justification needed.
Three bands, one recommendation.
People don’t act on “73.4.” They act on “this is a BUY.” The decision layer commits the system to a recommendation, which forces honesty: if your bands are wrong, the system makes obviously wrong calls.
That visibility is the point. A junior analyst sees the same recommendation legal does, and both can trace it back through the weights to the inputs that produced it. Confidence scores travel with every output — when the data is thin, the system says so.
Four decisions that make the system defensible.
None of these are about accuracy. They’re about whether the output can be reviewed, questioned, and changed without rewriting the system.
Weights as policy
The weights live in a versioned config a non-engineer can read. When the call is questioned, the answer is the inputs and the weights — both visible.
Configuration over coding
Adding a new factor means one adapter, one normalizer, one weight. No retraining. The system grows with the decision-maker’s understanding of the problem.
Graceful fallback
When ZIP-level data is missing, fall back to city. Mark the output with a confidence score. Never silently fabricate — if a signal is unavailable, say so.
Bands, not raw scores
Humans don’t act on 73.4. They act on BUY. The decision layer commits the system to a recommendation — which forces the failure mode to be visible.
The architecture isn’t about property.
The data sources change, the weights change, the stakeholder changes. The pattern — multi-source, weighted by explicit policy, with transparent reasoning — doesn’t.
Vendor risk scoring
Compliance, financial health, operational continuity, and security signals → approve / review / decline, with an audit trail.
Grant funding prioritization
Impact, alignment, capacity, and risk → fund / hold / decline, with reviewer confidence per criterion.
Site selection
Demographics, infrastructure, regulation, and competitive density → ranked candidate list weighted by strategic priority.
Loan underwriting policy
Credit, capacity, collateral, and conduct → approve / review / decline, with every decision auditable on demand.
A methodology, not a deployed engagement.
This piece describes a methodology and a system architecture, not a customer engagement. The property-scoring system the architecture is drawn from is in active development — some source adapters are live, others are stubbed pending data-source selection against budget and reliability constraints.
A production deployment for any specific stakeholder would involve a data-source selection workshop, weight calibration with subject-matter experts, integration with existing systems of record, and a thin analytics layer for ongoing decision review. Those steps are part of the engagement, not skipped over.
The reason this writeup exists at all is that the thinking — multi-source, weighted, explainable, defensible — is portable. The data sources change. The architecture doesn’t.
Multi-signal decisions, made defensible.
If you’re making decisions from multiple signals — or you should be, and right now it’s spreadsheets and judgment calls — a short conversation is enough to know whether this approach fits.