POLYSTRIKE

Research · · markets from 2026-01-30 to 2026-11-07

How Polymarket tweet-count markets settle: what 296 resolved markets show

Most of these markets are decided by a few posts, and the number on the Polymarket page is often several posts behind. Here is what 296 resolved markets show.

The rule

Each market splits the possible post counts into numbered ranges called buckets. When the window closes, Polymarket takes the final count from XTracker, its settlement source. The one bucket that contains that count resolves YES and every other bucket resolves NO.

What counts: main-feed posts, quote posts, reposts and main-feed replies. What doesn't: replies inside a thread and community reposts. A post that existed during the window can still count after it is deleted. The full rules and the three counters are on How it works.

The markets

AccountMarketsKindsBucket width (posts)
Elon Musk (@elonmusk, X)20278 weekly, 115 shorter windows, 9 monthly20 weekly · 25 or 40 windows · 40 monthly
Ted Cruz (@tedcruz, X)6565 weekly20
Donald Trump (Truth Social)2929 weekly20

Every market here is listed with its final count and winning bucket on Resolved markets. 280 of the 296 have a winning bucket we could grade; the other 16 (14 Ted Cruz weeks from mid-March to April, and three Elon Musk markets) are left out of the numbers below.

Most markets are decided by a few posts

We measured how far the final count sat from the nearest bucket boundary. In the median market it was 6 posts. In one market in four it was 3 posts or fewer.

In 57 of 261 markets (22%), three posts more or fewer would have changed the winning bucket. At that distance the details decide the market: one deleted post, one reply classified as a thread reply, one post a few seconds either side of the window close.

Measured on the 261 markets where our count fell in the winning bucket, so the distance is to the boundary of the bucket that actually won.

The Polymarket page number runs behind

The count printed on a Polymarket market page is a display value, and it refreshes irregularly. For each market we took the largest gap between that number and our own count at any moment during the window.

The median largest gap was 7 posts: about a third of a 20-post bucket. In one market in ten it reached 65 posts or more. Elon Musk and Ted Cruz markets both had a median of 7; Donald Trump's Truth Social markets had a median of 22.

The page number doesn't settle anything. A reader who judges a market from it can be looking at a count that is already a bucket out of date.

281 markets had page readings.

Where an independent count disagrees

Our own count, taken at the window close, fell in the winning bucket in 261 of 280 markets (93%). The 19 misses fall into three groups:

  • Six were 1 to 3 posts away from XTracker's count, next to a bucket boundary.
  • Nine were 8 to 16 posts apart. Six of these are Elon Musk markets in February and March 2026, when our count ran 12 to 14 posts above XTracker's: the two counters classified some posts differently.
  • Four were outages on our side, with gaps of 66 to 176 posts: Ted Cruz and Elon Musk in mid-May, Donald Trump in September.

The market pays on XTracker's classification of each post. An independent count is a check on it, not a substitute, and it is most useful where it shows how close the count is to a boundary.

What to take from this

  • Read the count the market settles on, XTracker's, not the number on the market page.
  • Check how far the count is from the nearest bucket boundary. Within a few posts, the counting rules matter more than the posting pace.
  • Deleted posts can still count if they existed during the window.

This is a description of how the markets settle, not trading advice.

Method

Winning buckets are Polymarket's resolutions. Our count is Polystrike's own count of the account at the window close (X polled every second for @elonmusk and every 5 seconds for @tedcruz, Truth Social every 15 seconds). We don't compare against XTracker's final numbers for closed markets, because XTracker revises them after the fact; the page gap uses the page value we recorded during the window.