How the Household Stability Index™ is designed.
HSI is explainable behavioral intelligence — not a replacement for traditional underwriting. This page describes how the score is constructed, what behavioral inputs it evaluates, how we surface explanations, and how we scope its limitations. It is written for the compliance, procurement, and investment teams who need to understand the model at the level of principles rather than parameters.
Proprietary weights, thresholds, and category coefficients are intentionally excluded from this document; those are shared under NDA with pilot partners and their reviewers.
Measure what most closely tracks the next 12 months of rent.
The Household Stability Index™ answers a narrower question than a credit score: given what a housing provider can verify about an applicant today, what is the direction and strength of the evidence that this household will pay rent on time and remain in place through the term of the lease?
The design objective is to organize behavioral evidence — payment history, tenancy stability, income adequacy, financial resilience, and disclosed context — into a single, explainable, reproducible signal. HSI does not attempt to predict credit-card default, auto-loan risk, or lifetime creditworthiness. It stays inside the housing decision.
HSI is designed to complement, not replace, traditional underwriting. Housing providers continue to run their existing screening steps (identity, criminal history where applicable, income verification, prior-landlord references). HSI adds behavior-first signal on top of those steps — and produces the reason codes needed to defend the resulting decision.
Five categories of behavior, each with structured sub-signals.
HSI groups behavioral evidence into five weighted categories. Each category is composed of structured sub-signals derived from what the applicant submits, what the housing provider verifies, and (when explicitly consented) financial evidence from open-banking sources.
Payment
On-time rent history across prior tenancies, payment consistency, recovery behavior after documented hardship. Emphasizes verified housing payments over inferred credit-file behavior.
Income
Income adequacy relative to the specific unit under consideration. Focus is on the sufficiency of documented income — not headcount, employer identity, or occupation.
Financial
Consented open-banking evidence of financial resilience: cash-flow patterns, savings signal, and stability of the recurring income deposit relative to the applicant's stated income.
Support
Household support systems and stability of context — length of tenancy history, responsible transitions between housing situations, presence of cosigner or guarantor when applicable.
Behavioral
Disclosure honesty, application completeness, response timing, and consistency across the fields the applicant submits directly.
Category weights and individual sub-signal coefficients are proprietary and versioned. They are shared with pilot partners under NDA and referenced in the model card attached to every recommendation.
Every score exposes its own reasoning.
Explainability is treated as a product surface, not a slogan. Four commitments hold across every HSI recommendation:
- Category breakdown per decision. Every score returns the contribution of each of the five categories to the final result. Reviewers see which categories drove the recommendation, not just the aggregate number.
- Machine-readable reason codes. Each recommendation ships with a fixed set of reason-code tokens that map to the specific sub-signals that most influenced the outcome. The same tokens appear in the operator UI and on any downstream adverse-action-format report.
- Version-stamped model card. Every decision records the exact HSI model version and weight configuration used to produce it, so recommendations are reproducible weeks or years later under internal or external review.
- Human authority preserved. HSI recommends. The housing provider decides. The system does not take autonomous adverse action; overrides and the rationale for them are captured in the audit log.
Behavior-first inputs, protected-class exclusion, and monitored outputs.
HSI is designed to support fair, defensible rental decisions. The fairness posture is implemented at three layers of the model lifecycle:
Protected-class attributes under the Fair Housing Act (race, color, religion, sex, familial status, national origin, disability) and applicable state or local protected classes are not used as HSI inputs. HSI operates on behavioral evidence, verified financial signals, and applicant-submitted context — not demographic identifiers.
Where a candidate input could plausibly act as a proxy for a protected class, that input is either excluded, weighted down, or transformed to preserve the behavioral signal while reducing proxy leakage. Every such decision is documented in the model card.
Recommendation-distribution monitoring is part of the operator workspace. Housing providers can review aggregate approval and decline rates and reason-code frequencies over configurable time windows, which supports their own fair-housing compliance review.
Fairness posture is an engineering commitment. It is not a claim of certification and does not substitute for the housing provider's own fair-housing compliance program or independent legal review.
What HSI is not.
Institutional buyers make better decisions when the limits of a decision-support tool are stated as clearly as its capabilities. HSI has three categories of limitation that reviewers should understand up front:
- HSI is not a replacement for traditional underwriting. HSI does not evaluate criminal-history considerations, does not perform identity verification, and does not itself confirm employment. It is designed to run alongside the housing provider's existing screening stack, not to replace it.
- HSI does not attempt to predict credit-card, auto-loan, or lifetime-credit behavior. The model is scoped to the housing decision. Buyers should not repurpose HSI outputs for underwriting decisions outside of rental housing without a separate suitability review.
- HSI cannot recover signal that does not exist in the submission. A thin submission with limited housing history produces a lower-confidence recommendation than a submission with verified positive history. Confidence bands are surfaced alongside the score so reviewers do not treat all recommendations as equally certain.
- HSI does not eliminate the housing provider's compliance obligations. Adverse-action delivery, record retention, fair-housing monitoring, and applicable state or local screening requirements remain the responsibility of the housing provider. HSI produces the outputs (reason codes, audit trail, adverse-action-format reports) that support those obligations.
Backtest, human-review, and versioned re-evaluation.
Before any HSI model version is exposed to a pilot partner, it is evaluated internally against three sources of ground truth:
Score is applied retrospectively to de-identified historical applications with known lease outcomes. We compare recommendation vs. observed outcome across category breakdowns and reason-code distributions.
A panel of experienced operators reviews a sample of borderline and high-impact recommendations. Disagreements are cataloged and reviewed as candidate inputs for future model revisions.
Every model version is re-evaluated against the previous version's decisions before roll-out. Material shifts in category weights or reason-code frequency are flagged for review before release.
Independent review is on the roadmap. Dates are targets, not certifications.
Third-party validation is a commitment, not a claim. The items below are engineering targets — none should be read as an achieved certification until the underlying report or attestation has been published.
- External model-card publication (Q3 2026 target). A public model card describing the current HSI model version, input categories, evaluation approach, and known limitations, in a format aligned with community model-card practices.
- Third-party fair-housing review (Q4 2026 target). Engagement with an independent reviewer to examine HSI's fair-housing posture, protected-class exclusion, and monitoring approach. Findings will be published in redacted form.
- SOC 2 Type I readiness assessment (Q2 2026 target) and Type I attestation (Q3 2026 target). Audit-focused controls review; the auditor engagement will be published upon signature. See the Trust Center and the Enterprise Security Overview PDF for the current controls posture.
- Peer-reviewable methodology write-up (2027 target). A longer-form methodology paper suitable for review by third-party researchers in housing policy or fair-lending analysis. Draft circulation begins with pilot partners under NDA before publication.
Dates reflect current engineering intent and will be updated as engagement letters are signed. StabilityLogic does not claim any regulatory certification or independent validation that has not been formally achieved.
For weights, coefficients, and portfolio backtests — let's talk.
Enterprise pilot partners receive the full model card, the current weight configuration, and a portfolio-specific backtest under a signed DPA and mutual NDA. This public methodology page describes the principles; the pilot scoping call opens the specifics.