Research
Essential question: does the gap between the market's top-priced bucket and its runner-up correlate with how often the favorite actually wins? For each city-day, the single snapshot closest to the anchor time below is used to find the top two priced buckets; the gap between them is binned into 5-point bands, and within each band the actual win rate of betting the favorite is compared against the overall average win rate across every gap size. Built from every Polymarket snapshot captured since 2026-05-22, through 2026-11-06.
There's real spread in favorite win rate city to city (49.0% to 67.3% across 9 cities at the default anchor) - some cities look remarkably consistent, others swing much more band to band. But every individual city's data, split across the gap bands, averages only about 7.3 observations per band - nowhere near the 30-sample bar used elsewhere on this page. Read any single city's curve as suggestive at most, not confirmed, until that changes.
At the current per-city accumulation rate, that average would roughly double to ~14.6/band by 2027-04-23 - still short of 30, but a reasonable "worth a second look" checkpoint for whether the city-to-city differences are holding up or just noise.
X-axis is the gap between the top two priced buckets (band midpoint, 5-point bands). The dashed line is the overall average win rate of betting the favorite, regardless of gap size - if the solid line rises meaningfully above it as the gap widens, a bigger lead over the runner-up is a real signal, not just noise.
Same data and method as the chart above, just finer-grained: the top chart bins the gap between the market's top two priced buckets into 5-point bands, which could be smoothing over real structure that only shows up at finer resolution. This re-bins the same observations into 1-point bands, restricted to 30-75% - past the tight-race noise near 0% and short of the near-lock tail above 75% where there's little left to distinguish.
Splitting into 43 narrow bins instead of wide ones spreads the same pool thin - as of this data (2026-05-22 to 2026-11-06), that's roughly 432 qualifying observations across 43 bins, averaging about 10.0 per bin. The 30-sample minimum used elsewhere on this site isn't met by a single bin here - read this as exploratory, not confirmed, until that changes.
At the current accumulation rate (~2.57 qualifying observations/day across all cities), doubling this sample to a still-imperfect but meaningfully less noisy ~20.0/bin average lands around 2027-04-23. Clearing the full 30-per-bin bar on every bin would take much longer at this rate - 2027-04-23 is a "worth a second look" date, not a "fully resolved" one.
For each 1-point gap band: the actual PnL from buying 1 YES unit on the favorite at its own quoted price and holding to resolution, net of the entry taker fee - averaged per unit across every observation in that band. Same sample-size caveat as the chart above applies here too.
Gap and the favorite's own price are naturally correlated - a heavily-priced favorite almost has to have a big lead. This splits the same snapshots into three price tiers first (50-65%, 65-80%, 80-95% - below 50% isn't really "a favorite," above 95% the market's already decided), then checks whether a bigger gap still predicts more wins within each tier. One line per price tier; if the lines are flat, gap isn't adding anything price didn't already tell you.