After the ninth home on the listings page there is a box offering four smarter ways to sort, in exchange for a free account. None of the four are wired up. This page works out which of them are true, which need rewording, and what else we could offer out of data we already hold. Every number on it was measured against the live databases on 16 August 2026.
The live copy, word for word, so we are arguing about the real thing.
Make a free account and the same homes re-order by the things a portal will not show you: cheapest per square foot, cheapest per unit, price cuts this month, and which ones have an open permit on file.
The paragraph promises open permits. The bullets swap it for time on market. Four things are promised twice and they are not the same four.
Run against the listings feed and the county records on 16 August 2026. Three of the figures I wrote down first were wrong until I re-ran them.
Ranked by how much a normal buyer would care, times how much of the data we genuinely have, divided by how badly it could look wrong. The first four are the four we already promise. The last four are new.
The cheapest signal there is that somebody wants out: they have already cut their own price.
Order every home by how far it has come down from the highest price it ever asked. Not from today's price — from its own peak.
/ng_price_drops.htmlCodex caught this one. A query that just asks "has the price changed" sweeps in the 1,642 listings that went up and labels them as cuts. It has to be written as a drop.
A portal shows an estimate. This shows the real closing price, and how far under the asking price it landed.
Every home in the zip that closed in the last eight months, with the gap between what was asked and what was paid.
/ng_closings.htmlThis is by zip code, not by street — the table holds city, zip and neighbourhood but no street address. The copy has to say zip. Also 148 rows carry a nonsense ratio above 2 and need filtering out before any average is shown.
For the buyer who cares about getting the most for the money rather than about a particular street.
The same homes, re-ordered by what you actually pay per square foot, so a big cheap house beats a small dear one.
/ng_best-value.htmlIt must rank on the server across all 82,798. Sorting the 24 cards already on screen is a lie dressed up as a ranking. The same warning the best-value page already carries applies here: a country-club condo looks cheap per foot because the buyer inherits the buy-in and the dues.
After four months on the market most sellers will listen to an offer they would have laughed at on day one.
The homes nobody has bought yet, oldest first, dated from when the listing really started rather than when it was last touched.
The trap is the relist trick: an agent pulls a listing and puts it back to reset the clock. We hold 2,224 dead-and-relisted records and 1,455 more candidates, so we can show the honest age. Without that cross-check, yesterday's relist looks like fresh inventory.
An unclosed permit can hold up a mortgage, an insurance binder and a closing. Knowing before you offer is worth real money.
Somebody pulled a permit for a roof, a pool or an addition and never closed it out. We can see it. The buyer's agent usually cannot.
/ng_permit_momentum.htmlThree counties only, and only 65.3% of listings reach a county record at all. So ‘no open permit’ must never be shown as a clean bill of health. The honest wording is ‘we found no open permit on file’.
This is the one that decides whether a tired house is a teardown or a bargain. We got it wrong once on a real appraisal.
The county zoning allows more units on this lot than the house standing on it. That is the difference between a house and a project.
/ng_upside.htmlZoning says what is allowed, not what will be approved. It has to read ‘the zoning allows’ and never ‘you can build’.
For duplexes and small buildings the only number that matters is what you pay for each rentable unit.
Rank small multifamily by price per unit instead of price per building, and print the unit count on the card so nobody has to trust us blind.
/ng_moneymakers.htmlCodex's flag: the promise as written says ‘for duplexes and small apartment buildings’ full stop. One in six of them has no unit count and would silently vanish from the sort. It has to say ‘where the unit count is known’.
Sometimes it is a two-hundred-dollar fence permit. Sometimes it is why the seller is selling. We state the record and let the buyer decide.
The county has an open code case or unpaid taxes recorded against the address. Shown as a fact on the card, never as a sort of its own.
/ng_lien_check.htmlGemini opened by calling this ‘Motivated Sellers’ and ranked it first. It folded when challenged and was right to. Never the words distressed, motivated or desperate about a named home. At 0.3% coverage unpaid tax is far too thin to headline a sort — it is a badge.
Most of the eight above are not new work. There is a finished page for them sitting in the same place as the listings page.
Every one of these answers HTTP 200 today. Nothing on the new listings page points at any of them.
Wiring the sorts is real work. Linking these is an afternoon. A visitor who signs up could land on a genuine tool the same day instead of on a promise.
Cheapest and safest first. Nothing here is built yet.
An afternoon of work, and a free account is worth something the same day.
Ten minutes. They disagree on the live page right now.
All three are pure listing data. No county join, so no gap to explain.
Cheapest per square foot is built and sitting unused.
The strongest hook here, and clean once the 148 junk rows are filtered.
Needs the county join, so it needs the careful wording agreed first.
A small list. Fill the missing sixth from the parcel roll first.
Written down rather than discovered later.
The page calls one endpoint and that endpoint orders by one thing only: most recently changed. There is no sorting code anywhere on the page. Every word in the amber box is a promise we have not kept yet.
The paragraph promises open permits. The bullets underneath drop open permits and put time-on-market in its place. Whichever four we ship, the paragraph and the bullets have to name the same four.
Anything needing county data — permits, code cases, zoning, unpaid tax — travels through the folio match, which reaches 54,081 of the 82,798 active listings. That is 65.3%. For the other third we do not know, and ‘we do not know’ must never be displayed as ‘it is clean’.
I told Gemini and Codex that ‘what homes on this street actually sold for’ was the strongest hook here. Gemini agreed enthusiastically. Codex said it could not verify it, and Codex was right to hedge: the closed sales carry city, zip and neighbourhood but no street address. Still rank 2 — but it is by zip, and the copy has to say so.
Codex caught two real bugs: that a changed price is not a cut, and that two of my own counts contradicted each other. Gemini named the county-data gap as the thing most likely to make us look dishonest, which was the right instinct, though it put the figure at 37% when the real number is 65.3%. It also opened by ranking distressed sellers first and folded cleanly when challenged on the legal risk. Every coverage figure on this page was then measured by running the query on the live server, not taken from either of them.
Write your notes above and I will build exactly what survives them — starting with the fifteen tools that already exist.