How these numbers are measured — and what they cannot say

The fill rate

The headline number is the share of listings that reached five applications — the point at which a listing stops accepting more. A listing counts only if it was observed from the moment it had zero applications and at least 48 hours before its start date, and it is then followed for 48 hours.

The choice rate

The owner-side reading of the same experiment: the share of watched listings that received at least two applications inside the same 48-hour window — the point at which the owner is choosing a sitter rather than hoping for one. Same entry rules, same floors, same window as the fill rate; the two are different thresholds on one measurement, not two studies.

Every rate here is a floor. Two things bias it downward, and both bias it hardest in the busiest markets: a listing that fills between two observations is never caught at zero and so never enters the sample, and a listing that is withdrawn and reposted has its outcome discarded. Fast-filling listings are exactly the ones most likely to be lost to both. So a contested city's true figure is higher than the one shown, and the gap between the top and the bottom of the table is wider than it looks. We do not know by how much, and we would rather say so than publish a correction we cannot support.

What the 48-hour window measures — and what it doesn't

Both rates stop counting 48 hours after a listing is first seen at zero applications. They measure how fast a market moves, not whether sits eventually fill: a listing that draws all of its applications on day four or day ten counts as a zero in these rates, and many do. A low 48-hour rate means a slow market, not a dead one — which is why the completed-sits record below is published alongside it.

Provably completed sits — the long record

Separately from the 48-hour experiment, we count sits that provably happened: an assignment marked confirmed, or one carrying a written sitter review. Either proves a sitter was chosen — so every counted sit had at least one real application. Listing histories reach back years, so this record long predates the measurement window above.

These are counts and floors, never rates. A past sit that went unfilled leaves no trace on any listing page, so there is no denominator — dividing completed sits by anything produces survivorship bias, and we don't. Counts also depend on how thoroughly each city was swept, so they are not comparable across cities; "at least N sits completed here" is the entire claim.

Rates are measured over a window beginning 1 May 2026, and recomputed as it grows. That start is not arbitrary: three changes to how listings were discovered, plus a two-week collection outage, make anything earlier non-comparable. Each row carries its own n and window in the CSV.

What is counted

One row per date window, not per home. A home offering three separate windows is three separately winnable opportunities; collapsing them to one loses about a sixth of the bookable inventory. "Homes" counts distinct properties; "n" counts windows.

Gapless days

The longest run of consecutive housed days that can actually be booked: no two stays overlap, and each begins exactly when the previous ends, starting on any day within the next 90. This is deliberately stricter than "days on which at least one listing is open", which is a supply ceiling rather than an itinerary — those windows overlap and cannot all be taken.

Suppression

• No city with fewer than 10 open listings is published.
• No rate is shown below n=100. At n=30 a proportion still carries a ±18 percentage-point confidence interval. Below the floor the page says so, in place, rather than printing noise.
• No individual listing, photo, owner name or listing text is republished.

What these numbers cannot say

• Rates are comparable across cities. Counts are not. The corpus is discovered from a set of search centres with very different radii, so a city's listing count partly measures how thoroughly it was swept. Rankings here use rates for that reason.
• This ranks within the cities observed, not all markets worldwide.
• Applicant counts are a snapshot as reported on each open listing at the last refresh — not a final tally.
• Nothing here records who won a sit. No claim about what makes an application succeed can be supported by this data, and none is made.
• Seasonality is not published. It needs two comparable years; the continuous record began mid-2025 and has a documented gap between December 2025 and April 2026.

Citing this data

Aggregates are open under CC BY 4.0: reuse freely with attribution to "SitFacts (sitfacts.pages.dev)". The per-city CSV and column definitions live on the data page.

Generated 2026-07-27. 84 cities published.