How this site is built
You came here for honest guidance. Here is the honest version of how the site itself gets made.
Chris co-builds this site with AI systems he directs. Every number, claim, and sentence passes human review gates before it ships. And when the AI gets it wrong — it does — the corrections are public. That last part is the whole philosophy: the machine does the assembly, a human who has actually walked these blocks owns the truth, and you get to see both.
What the AI does
The neighborhood pages
The 85 neighborhood guides are assembled by AI from local data, then read by a person before they ship. The imagery decision, the taglines, and the 'Chris Take' on every page are Chris's.
The answer engine
281 plain-language answers to real buyer and seller questions, drafted by AI against verified sources, with human review gates on every one that touches fair housing, schools, or neighborhoods.
Topher, the voice
The site's AI guide is Chris's cloned voice (ElevenLabs). It says so on the advocate page, and it will connect you to the actual Chris the moment you ask.
The tools
What's My Number, the Reality Check, and the rest run their math in code you could audit — the numbers come from your inputs, not a database of leads.
What the human does
The review gates are the site. Four of them, all standing:
Fair Housing screens
Every AI path passes a compliance screen before anything reaches you. Words that steer — by race, religion, familial status, or any protected class — get deleted at the gate, and the worst offenders go to a human review queue.
The corpus audit
A standing audit verifies the site's claims against primary sources. 87 of 638 records verified in the first pass; corrections ship when a claim fails, not after.
Audio provenance
Every audio track carries an identifier and a provenance record. 29 of 29 verified against the catalog. Recorded audio is never quietly re-written — it gets re-recorded.
Human verification queues
When the machine flags a rewrite — school claims, neighborhood character, market numbers — it waits for a human. Some of the best lines on this site came out of that queue.
What our AI got wrong
A running log, newest first. Most sites bury their corrections; this one publishes them, because a correction you can see is worth more than an accuracy claim you can't check.
The Corrections Desk — an agent that re-checks our citations against the statutes themselves — found our long-hold guide describing the one-year residence rule as covering owners who 'moved into licensed care.' The statute is narrower: it requires becoming incapable of self-care, and it lives at IRC §121(d)(7), not the (d)(8) we cited. Four more citations pointed at the wrong corner of their statute. Five corrected; sixteen others came back clean. A human read every fix before it shipped.
The AI gave 26 neighborhoods the same dream-lake hero image — vintage station wagons, 'PRIVATE DRIVE' signs, golden light no Minnesota afternoon ever produced. A full visual census caught it; the dedupe is underway.
Two neighborhood pages said 'for under k' and 'pay k-k less than Hale' — a templated dollar figure lost its number. Caught by the same census, fixed from the site's own market data.
Lilydale's hero showed the Minneapolis skyline. Lilydale's bluffs face St. Paul. A local would have caught it in three seconds; the census caught it in three days. On the fix list.
22 statutory citations in the transition guides re-checked against primary sources, 5 corrected. Every verdict is filed with its evidence — the same receipts that wrote the entries above.
The rules the AI must follow
Fair Housing compliance outranks everything. No aggregate ratings, anywhere. Recorded audio never changes underneath a transcript. And nothing the machine writes about a protected class ships without a human reading it first. The full policies live in the site's compliance docs — ask Chris and he'll walk you through any of them.