Real estate software development.

In most industries the rules govern how you handle data. In housing, a large share govern the decision the software makes, so the model's output is itself the regulated act. That changes the engineering problem. Encrypting a database is verified once and asserted afterward, while showing that a model does not produce a disparate impact is a statistical question that has to be re-answered as the model and its inputs change. Reveneau builds that measurement into the delivery process rather than adding it at the end.

The interagency rule applies to institutions using automated valuation models in certain credit decisions or securitization determinations, and it took effect on October 1, 2025. It requires control systems ensuring the models meet quality control standards, including random sample testing and reviews and compliance with applicable nondiscrimination laws. Whether your specific product is covered is a question for your counsel, because it turns on how the estimate is used downstream rather than on how the model is built.

Yes. Housing law is concerned with outcomes rather than intentions, so a discriminatory result does not require a discriminatory input. A model with no protected attribute anywhere in its training data can still produce a disparate impact through features that correlate with one, and postal code is the most common example. This is why Reveneau treats outcome testing as a build requirement rather than relying on an inspection of the input list.

HUD issued guidance on May 2, 2024 stating that the Fair Housing Act applies to tenant screening and to the advertising of housing opportunities through online platforms, including when algorithms and artificial intelligence perform those functions. The practical reading is that using a model does not move the obligation onto the model vendor. Reveneau builds screening workflows so the reason for a decision is recorded at the moment it is made, since an applicant may ask for it later.

Your counsel and policy owners decide that. Different statistical definitions of fairness conflict mathematically, so a product generally cannot satisfy all of them at once, and choosing among them depends on your specific context and legal exposure. Reveneau's role is to make the chosen measurement continuous, visible in the check suite, and hard to disable quietly. Picking whichever metric a library implements by default is the failure mode to avoid.

State and local variation is held as configuration a non-engineer can read and change, rather than as rules hardcoded across the codebase. The core obligations are often similar between jurisdictions while thresholds, timelines, and exemptions differ, and that variance is itself a software design problem. Building it as config means a rule change is an edit and a re-run of the check suite, instead of an engineering project scheduled against everything else in the queue.

Both. A common starting point on an existing product is establishing what the current system actually does, since the decision logic has usually spread across several services and nobody holds the whole picture. From there the specification gets written down, the eval suite gets built against it, and gaps surface as failing checks rather than as opinions. Our guide on compliance gaps in real estate software covers what that audit usually turns up.

The AVM quality control standards call for random sample testing and reviews, which is a description of a running test suite rather than of a document. Because Reveneau writes an eval suite from the specification and requires code to pass it before shipping, the normal output of the delivery process is the same artefact the rule describes. That does not make an engineering process a legal opinion, and the scope of what must be tested is still your counsel's call.

You own it. Reveneau's standard arrangement is that the client owns the product, the code, the designs, and the intellectual property outright, with no license held back and no requirement to keep working with us afterward. For proptech specifically that ownership includes the eval suite and the testing configuration, so your own team can run the fair housing checks, read the results, and extend them once the engagement ends.

Reveneau scopes to weeks from an agreed specification to a first production release, and the specification work is where the early effort goes. In proptech that scoping stage carries more weight than usual, because deciding what the system is allowed to decide, and what has to be explainable afterward, shapes the architecture. Work then ships in increments so the testing can be checked against real data at each stage rather than only at launch.