Why the list refreshes every nine minutes

The Fulton Yard Loft went on our list at 09:14 on a Tuesday. By 09:41 it had eleven enquiries and by Thursday evening it was gone. Three days on market. If our list had synced overnight like most of them do, the first person to see it through us would have seen it on Wednesday morning, twenty-two hours late, behind eleven people who were already ahead of them.

That is the whole argument. Everything below is the arithmetic behind it and the parts we got wrong on the way.

What a nightly sync actually means

Most listing portals pull from the multiple listing service once a night, usually between two and four in the morning. It is a sensible engineering decision. The feed is large, the API has rate limits, and running it once while nobody is awake is cheap and predictable.

The cost lands entirely on the person reading the site. A house that comes to market at ten on Saturday morning does not appear until Sunday at three. A house that goes under contract at four on Friday afternoon stays on the page, looking available, all weekend. You spend Saturday planning your day around a property that has been spoken for since before you woke up, and you find out when the agent does not call back.

We ran that way for the first fourteen months. It is not a scandal. It is just wrong often enough to be embarrassing, and the embarrassment is always yours rather than ours, which is the part we could not live with.

Nine minutes, and why not one

The obvious answer is to poll constantly. We tried a sixty-second cycle in a staging environment and it fell over for three reasons.

• Rate limits. Our data agreements permit a fixed number of calls per hour per region. At sixty seconds across 34 metros we exhausted the hourly allowance before the hour was half done, and the feed shut us out — which produces a stale list, the exact thing we were trying to fix.
• The source is not that fast. An agent marks a property under contract when they get round to it. The median gap between the accepted offer and the status change in the feed is a little under four hours. Polling every minute to catch a change that arrives four hours late is theatre.
• Cache invalidation. Every refresh that changes anything rebuilds the affected pages and purges the edge cache. At one minute, the cache never warms up, so every visitor waits on an origin response. The site got measurably slower in exchange for freshness nobody could perceive.

Nine minutes is where those three curves cross. It fits inside every rate limit we have with room for a retry, it is well under the four-hour reporting lag so we are never the bottleneck, and the cache stays warm because most cycles change nothing at all. On a normal Tuesday, seven of every eight refreshes across the whole network find no change whatsoever.

It is also honest to say the number is not magic. Seven minutes would work. Twelve would work. Nine is the middle of the band where the constraints are all satisfied, and having picked it we would rather print it on the page than describe the list as "live", which is what everyone says and nobody means.

What it costs to run

About $310 a month in compute and roughly forty hours of one engineer's year in maintenance, mostly spent on feed formats changing without notice. It is not a large number and we are not asking for credit for spending it. We mention it because "why doesn't everyone do this" is a fair question and the answer is not that it is hard — it is that nobody gets sued for a stale listing, so nothing forces the spend.

Where nine minutes still is not enough

Two places, and pretending otherwise would make this post the same marketing noise it is complaining about.

Fast rentals. In Brooklyn the median rental goes from listed to leased in under five days, and the good ones go in under two. Our refresh gets you the listing within nine minutes of it appearing. It does not get you to the front of a queue of forty people. That is a viewing-slots problem, not a data problem, and the honest answer is to have your documents ready before you start looking rather than to refresh faster.

Off-market and pre-market. A meaningful share of houses in tight metros never reach the feed at all — they are sold inside a brokerage before anyone publishes them. No refresh interval touches that. It is why we still make phone calls, and why the enquiry form asks what you are looking for rather than only which listing you clicked.

How to tell if a site is lying to you about this

Three checks, none of which require any inside knowledge.

1. Find a property you know sold last week. If it is still listed as available, the site is either syncing slowly or deliberately keeping dead listings up to inflate its inventory count. Both are worth knowing.
2. Look for a timestamp. Not "updated daily" — an actual time. If a site is proud of its freshness it will show you the clock; if it says "real time" with no clock, it is a word rather than a claim.
3. Count the listings against the local board's published figure. A site showing four times more homes than the market contains is counting expired, withdrawn and off-market records as live.

Our number is on the listings page, next to the results, and it says how many minutes ago rather than "live". Six homes this week. That is genuinely all of them, which is a smaller and more useful claim than most.

The part that matters on a Saturday

If you are looking at rentals, the arithmetic of whether to keep renting at all is a separate question and a bigger one — the rent-versus-buy arithmetic will tell you how many years you would need to stay before buying wins, and in Brooklyn the answer surprises people. If you are looking at a specific place, look at the timestamp before you plan the day around it.

And if a listing on this site turns out to be gone when you call, tell us. That is a bug, not a fact of life, and we would like to see the log line.