The thread and Kalshi's reply
In late September 2026, a public data dispute landed on Kalshi's crypto perpetual markets. CoinDesk reported that an analyst posting as “Beni,” co-founder of Stealth Neolab, alleged that Kalshi's ETH-PERP contract showed high 24-hour volume against a much smaller open interest, and that repetitive fixed-dollar trades made up a large share of ETH perp volume on four separate days (CoinDesk, Sept 21, 2026).
Kalshi responded through its crypto lead, posting under the handle “IcoBeast.eth.” CoinDesk reported his response that volume is reported as “maximum potential payout, not the upfront cash spent,” with the example that a trader buying 100,000 contracts priced at 30 cents spends $30,000 in cash while the system records $100,000 in volume (same CoinDesk report). Unchained reported a similar framing from the company: Kalshi “counts volume the way Polymarket does, as the maximum payout of contracts traded,” rather than cash spent (Unchained, Sept 22, 2026).
A follow-up CoinDesk analysis reported that a single recurring trade size near a fixed dollar value was a majority of sampled ETH-PERP volume across several days, and that the recurring dollar target shifted over time. CoinDesk also reported that after publication, Kalshi said the trades came from one market maker posting fixed-size orders under a flat-monthly-fee market-making program, that hundreds of distinct traders took the other side, and that it found no evidence of collusion or wash trading (CoinDesk follow-up, Sept 22, 2026). Commentary by Rajiv Sethi argued the pattern could arise from “volume inflation without wash trading” through exchange incentive design and adverse selection against quoting market makers (Substack, Sept 23, 2026). Separately, Matthew Snider argued the volume is real but structurally subsidized and urged judging the business on revenue, not volume (Substack, Sept 23, 2026).
ZenHodl does not capture Kalshi's crypto perpetual markets. This article makes no claims about crypto perp volume. It applies the same volume-definition questions — notional versus cash spent, trade-size clustering, and same-second bursts — to ZenHodl's own sports trade-print data and reports what that shows.
What headline volume measures on Kalshi
Kalshi's own price-mechanics page states: “The combined investment of both users must equal $1” (Kalshi help center). For every contract, a YES side priced at p cents and a NO side priced at 100-p cents together post p + (100-p) = $1 in collateral. Each contract pays $1 to the winning side at settlement and $0 to the other.
That is why headline volume — contracts times $1 of maximum payout — and the combined premium of both sides are the same quantity described two ways. Headline volume is not the cash a buyer transferred at execution, not the collateral posted by the aggressor alone, and not revenue or fees. The taker's cash on a given print is the contract count times the taker's own execution price: a trader buying 100,000 contracts at 30 cents spends $30,000 even though the headline records $100,000 of notional, as in the example Kalshi's crypto lead gave. The resting side's share (the other 70 cents per contract) was posted when that order was placed. Trade prints carry execution prices but no fees, so every cash figure in this article is premium before Kalshi's trading fees. Kalshi separately publishes a volume incentive program in which eligible volume uses contracts priced $0.03-$0.97 on the central limit order book during active reward periods, with an explicit note that the price band “does not apply to perpetual futures” (Kalshi help center).
One mechanical consequence matters for any venue comparison: dividing headline notional by taker cash is arithmetically equal to 100 divided by the volume-weighted average taker price in cents. A cheap contract produces a much higher ratio than an expensive contract. A high notional-to-cash ratio can therefore rise automatically for low-priced, longshot-style contracts, without implying anything else.
What ZenHodl's sports data shows
The 2026-09-06 vintage of ZenHodl's Kalshi Microstructure Tape holds 15,808,966 executed trade prints, each with a distinct trade_id, across 15 sports families: game-winner markets for MLB, WNBA, NFL and college football; spread and total markets for MLB, NFL and college football; and game markets for five European soccer leagues. Our recorder began polling on Jun 21, 2026 and retrieved its first print on Jun 22; the vintage ends Sep 6. Print timestamps start on May 15 because Kalshi's trade-history endpoint also returned earlier prints on our first polls, but only 3,484 prints (0.02%) predate our first retrieval. Those prints total 3.12 billion contracts — 3,122,611,883.85 contracts at exact precision — which at $1 of maximum payout per contract is $3.12 billion of headline notional. The taker cash — each print's contracts times the price the aggressor paid, before fees — is $1.48 billion. The ratio of headline notional to taker cash is 2.10x. These totals cover only the prints we retrieved for these 15 families, so they are a floor on those families' traded volume and not Kalshi's sports total (see Methodology).
That 2.10x is mechanical, not a verdict: it equals 100 divided by the volume-weighted average taker price of 47.6 cents. For cheap sports contracts, the ratio rises automatically. Prints where the taker bought YES were 71.5% of contracts, and prints where the taker bought NO were 28.5%. By print count, MLB game-winner markets (KXMLBGAME) accounted for 68.3% and WNBA game markets 13.7% — a reminder that the aggregate is heavily shaped by a small number of families, and that GAME, SPREAD, and TOTAL markets within the same sport are structurally different line formats rather than directly comparable baselines.
Prints that share the same market, second, execution price, taker side and exact size with at least one other print made up 7.5% of trades and 4.2% of contracts: 1,191,536 prints in 476,628 such groups, the largest holding 273 prints. These are pattern counts only. Because the prints are anonymous, they say nothing about whether one account or many accounts produced any group.
The size-clustering test and why round lots are normal
Within each (family, date) cell with at least 30 trades, ZenHodl measured how much of the cell's contract volume sat at one exact trade size. The highest share, 66.1%, was KXNCAAFGAME (college football game markets) on Sep 1, 2026 — but it is not a repeated size. It is one print: a taker bought 49,650.45 NO contracts at 1 cent, about $496.50 of cash, in a 232-trade cell whose other 231 prints totalled 25,500.40 contracts. A single cheap trade can dominate a contract count, which is the notional-versus-cash point again. Counting only sizes that occur at least twice, the highest share was 53.6%: two 300-contract prints in KXNFLGAME on May 15, 2026, in a 32-trade cell. That is contract-count concentration, not a dollar-value pattern.
Round-number contract counts are ordinary and price-independent. Across all contracts, 21.1% were round sizes (multiples of 10, 25, 50, or 100), while 78.9% were non-round. Narrowing to sizes that actually repeat at least twice in a family-day cell, 62.8% of repeated-size contract volume was non-round. Those two non-round figures answer different questions and should not be conflated: the first covers all contracts including one-off sizes, the second covers only the population of repeated sizes.
A repeated number of contracts is not the same test as a repeated dollar value. The crypto allegation was about a fixed dollar notional recurring across changing prices over time, which implies the contract count would have to vary inversely with price. A round contract count does not imply that at all. The closer analogue in the sports tape is a repeated dollar amount, and ZenHodl measured that separately. The top dollar-cluster result was the same two KXNFLGAME prints on May 15, 2026: 300 contracts at 77 cents each, $231 of taker cash apiece, four seconds apart in the same game's two team markets, and the first two prints of that family in the file. Together they were 53.5% of that day's $863.52 of taker cash across 30 trades of at least $0.50. Among dollar amounts that recur at least three times in a cell, the highest share was 27.4% (twelve prints of about $24, KXNFLGAME, Jun 29, 2026). Even that comparison comes with a transparency caveat: 155 of 523 family-day cells (29.6%) had fewer than 30 trades and are excluded from the dollar-clustering ranking.
What anonymous prints cannot tell you
The ZenHodl trade schema contains no account or member identifier. That is a structural limit, not a missing feature to be worked around. What can be measured is pattern: same size, same price, same second, same family-day. What cannot be determined is whether two prints share a counterparty, whether one account or many accounts were involved, or whether any same-size or same-second cluster reflects self-matching or wash trading. For the same reason, a same-account or self-match test cannot be replicated from this tape. Every clustering and burst figure in this article is therefore a measurement of pattern only, never evidence of who traded. No wash-trading, self-matching, or bad-actor claim is made or implied here, and none is supported by the data.
What this means for anyone comparing venues
A few neutral, general takeaways follow for anyone reading “volume” headlines across prediction markets or exchanges. First, confirm the denomination: maximum payout and principal at risk are different numbers, and some venues report both. Second, remember the ratio mechanics: notional divided by cash is 100 over the average trade price, so low-priced contract mixes mechanically inflate the ratio. Third, match the clustering test to the claimed pattern — a repeated contract count and a repeated dollar value are different phenomena, and comparing them as if they were the same will mislead. Fourth, incentive design can generate volume without collusion, as the public commentary on the crypto dispute argued, so a concentration pattern alone is not a conclusion. Finally, an anonymous tape yields pattern counts, not counterparty identity; any wash-trading inference requires identifiers that may simply not exist in public prints.
Methodology
All figures are SPORTS trade prints only, from the 2026-09-06 vintage of the tape. ZenHodl does not capture Kalshi's crypto perpetual markets, and nothing in this analysis is a claim about them. The prints were retrieved from Kalshi's own trade-history endpoint and deduplicated on trade_id. Contract counts are preserved as genuinely fractional values from the raw API and were not rounded or cast to integers. Prints are anonymous: no account or member identification exists in the schema.
The file is a census of the prints our recorder retrieved, not a verified complete record of every print in these 15 families. On each poll the recorder pages back through a market's recent prints, which recovers prints executed while it was down, but only up to a cap per market per poll: one page of 100 prints until mid-July 2026 (we measured that a single page lost about 24% of prints on busy in-play markets), and up to five pages after that. Busy markets and long outages can exceed the cap, and the tape's depth polls show 351 global gaps of more than two minutes between Jun 21 and Sep 6 (listed in the tape's COVERAGE.md). Every total above is therefore a floor on those families' traded volume, and busy in-play periods are the most likely to be under-counted.
The headline numbers were independently recomputed from the full dataset by a separately written pipeline. Three of four headline metrics — total contracts, the notional-to-cash ratio, and the maximum single-size share — matched bit-for-bit. The fourth, the non-round repeated-size share, was deliberately computed under a narrower definition (repeated-size cells only) and is reported alongside the whole-data non-round share with both scopes stated, rather than forced to agree. A third, separately written check before publication reproduced every figure in this article from the delivered files. It found that 8 prints whose timestamps carry no fractional seconds had been dropped from the time-based counts; including them adds 2 prints to the matched-group count and removes 3 spurious family-day cells, and changes no percentage.
Check the prints yourself
To check these definitions against the prints yourself, start with the free schema and samples at zenhodl.net/samples, or the $9 two-day Kalshi tryout (same schema as the full file). The full Kalshi Microstructure Tape is a dated snapshot through Sep 6, 2026 that contains every print analysed here, with the schema and caveats described above. It is data for measuring volume, not a trading signal.