Most real-estate reports tell you what already happened.
Prices went up. Sales went down. Rents changed. Permits were issued.
We’re building something different: a market-intelligence engine designed to preserve those observations over time, connect datasets that normally live apart, and ask what is changing before the headline makes it obvious.
Sioux Falls is our first test market. And one of the first combinations worth investigating is in apartments.
The interesting part is not one number
In the South Dakota Multi-Housing Association’s July 2026 Sioux Falls survey, 869 of 14,706 surveyed units were vacant. That is a 5.91% vacancy rate, down from 6.62% in January. Market-rate vacancy also improved, from 6.34% to 5.76%.
By itself, falling vacancy is useful.
But then we put it beside the supply data.
Sioux Falls permitted 1,168 multifamily units in 2025. That was down from 1,256 in 2024, 1,930 in 2023, and 3,343 in 2022.
So the market appears to be absorbing apartment slack while the pipeline of newly permitted multifamily housing has fallen sharply from its 2022 peak.
That does not prove rents are about to surge. Permits are not deliveries. Survey samples change. Rent datasets measure different things.
It does give us a much better question:
Is renter demand absorbing new apartment supply faster than developers are replacing it?
That is the kind of question this system is being built to answer.
Why we keep incompatible rent series separate
Another useful lesson appeared immediately.
Different Zillow products can show different rent numbers because they measure different populations and use different methodologies. HUD contract rents are something else entirely. A portal’s advertised availability is not physical vacancy.
A normal dashboard is tempted to blend those numbers into one clean chart.
We are deliberately not doing that.
Every observation keeps its geography, period, source, source URL, ingestion date, source vintage when available, sample size when available, quality flags, and the original record or snapshot reference where practical.
If two sources disagree, the disagreement can be information.
Then we add the debt
This is where the project gets more interesting.
We are also collecting public multifamily credit records from HUD and FHFA: FHA-insured mortgages, maturity dates, original mortgage amounts, recorded interest rates, lenders and servicers, Enterprise multifamily acquisition history, and FHA production records.
Imagine two otherwise similar apartment properties.
Both are well occupied. Both have similar rents. Both sit in a submarket where vacancy is improving.
One has long-dated debt.
The other has a low-rate mortgage approaching maturity.
The real estate may be healthy while the second owner’s capital structure becomes uncomfortable.
That distinction matters. A future version of the engine should be able to identify a property where operating fundamentals are improving but refinancing pressure could still force new equity, a recapitalization, a sale, or another capital event.
In other words:
the debt can be distressed without the building being distressed.
What we’re actually building
The valuable part is not an AI-generated market opinion.
The valuable part is the market memory underneath it.
Public source data flows into immutable raw observations. We preserve source and vintage history. Normalized metrics sit above the raw layer. Deterministic calculations compare periods and geographies. Anomaly detection looks for relationships that deserve investigation. Only then does an AI research layer explain what changed and trace the conclusion back to the observations.
We want the system to eventually answer questions such as:
- Which Sioux Falls ZIP is strengthening before prices respond?
- Where is new apartment supply being absorbed unusually quickly?
- Which apparently strong market is actually weakening underneath the headline price?
- Which apartment properties have improving fundamentals but an approaching refinancing problem?
- What information was genuinely available six months ago, and what happened afterward?
The last question is especially important. We are preserving vintages so the system can eventually be backtested without pretending it knew information that had not been published yet.
The hypothesis we’re watching now
The current Sioux Falls apartment hypothesis is simple:
Vacancy has been improving while the future multifamily pipeline has contracted substantially from its peak.
Now we watch what happens next.
If vacancy continues falling, concessions retreat, effective rents strengthen, and new supply remains restrained, the evidence gets stronger.
If vacancy reverses, deliveries overwhelm absorption, or rent weakness broadens, the thesis breaks.
Either outcome is useful.
The point is not to make the story true.
The point is to build a system good enough to tell us when it stops being true.
That is Market Radar: data first, intelligence second, interface last.