Case study · Pantera Research Lab × Surf · by Ally Zach & Danning SuiCrypto on the Clock — every exhibit, with live provenance
Pantera Research Lab’s market-structure study of near-term crypto prediction markets, recreated here chart-for-chart on Surf’s prediction-market pipeline — with the data window, methodology, and API path behind each exhibit. Read the published report ↗
The Race
Two venues, two different mixes. Polymarket leads crypto dollars while Kalshi is a sports book with a fast-growing crypto wing — and their crypto shares are moving in opposite directions.
Two Venues, Two Different Mixes
Weekly traded dollars by market category across both venues (top categories, remainder bucketed as Other). Refreshed daily from cached public analytics calls.
Opposite Directions of Travel
Crypto as a share of each venue’s monthly dollar volume — Polymarket cooling while Kalshi climbs.
The Product Split
The same bet in two shapes. Almost all of the crypto tape is short-horizon — five- and fifteen-minute contracts — and both venues trade small, with Kalshi running a little larger.
Short-Term Is the Whole Game on Both Venues
Rolling 28-week crypto volume split by contract length, per venue, refreshed daily. The short up/down contracts carry the tape; toggle to the asset split.
Both Tapes Trade Small, But Kalshi Runs Larger
Rolling seven-day percentile ladder of individual trade sizes on a log-dollar scale, refreshed daily. Both venues are dominated by tiny tickets; Kalshi runs larger throughout.
The Economics
More volume did not mean more fees. Polymarket’s fee take is thin against its dollar throughput, concentrated in the shortest contracts; Kalshi’s crypto fees are modeled from its published schedule.
Polymarket Crypto Volume and Allocated Fees
The observed on-chain protocol fee total is deduplicated by day, then allocated to crypto using raw per-fill fee weights. This preserves continuity across the V1/V2 collection-method change without presenting the allocation as directly observed per-fill fees.
Maker Volume by Tier
Weekly maker volume bucketed by each wallet’s fixed pre-V2 tier — showing how maker activity contracted through the fee transition.
Crypto Fees by Contract Duration
Weekly crypto fees split by contract duration. Polymarket allocates the deduplicated observed protocol fee total using per-fill duration weights; Kalshi is modeled from its published schedule, 0.07 × contracts × p × (1−p). Each venue keeps its own bucket taxonomy.
Who Is Trading
The headline volume is broader than the real user base. A quarter of wallets — the bot-like layer — drives nearly nine-tenths of the flow, trades sub-second in micro tickets, and mostly trades other bots.
A Quarter of Wallets Drives Nearly Nine-Tenths of Flow
Share of wallets, trades and volume held by each behavioral class. A small bot-like cohort accounts for the overwhelming majority of trades and dollars.
Wallets, Trades and Dollars by Class
Absolute wallet, trade and volume counts per behavioral class. Bot-like: >100 trades/active day, or a near-constant ticket across 500+ trades, or >1,000 distinct markets. Sophisticated: median ticket $200–$100K. Retail: everything else.
Micro Tickets Carry The Tape
Distribution of the 5-minute taker population — share of trades and volume by wallet median-ticket band, then by cadence band and top-N concentration. Micro tickets carry the tape.
Machines Serving Machines
Who is on each side of the 5-minute tape: the counterparty split, then bot-both-sides vs no-bot share and the sub-second / tiny-ticket share across contract-duration horizons.
Settlement Integrity
A replication of Dai, Jia & Yu. In cycles that are still even seconds before the close, Binance spot volume erupts and price moves hardest — a behavioral footprint at the moment of settlement, then a snap back.
Based on Dai, Jia (Stanford) & Yu (SMU), When Prediction Markets Move Prices: Settlement Manipulation and Its Market-Design Remedy (Crypto Camp 2026). Independent replication on May 2026 Polymarket and Binance data. This is a behavioral footprint analysis, not a claim about any specific account, nor that any oracle was moved.
Inside the Cycle: Volume Erupts at the Close
Mean Binance BTC/USDT spot volume by second into the cycle, still-even cycles against already-decided ones. Reproduced from the study’s Binance 1-second tape.
Short Markets Resolve on Small Price Differences
Share of BTC up/down cycles by the walk-away move within each band, across 5-, 15-, 60-minute and 4-hour products. Short markets cluster in the smallest bands.
Over a Longer Window, Price Drifts Further
Typical (median) absolute BTC move from cycle open to close, by contract length, in basis points — longer horizons drift further.
Inside the Cycle: Price Moves Hardest at the Close
Mean absolute BTC price move per second across the cycle, still-even vs already-decided — the move concentrates in the final seconds before settlement.
The Price Moves, Then Snaps Back
Sign-aligned BTC spot return around the close for still-even cycles — price rises roughly 1.6bps into settlement, then gives back most of the move. 5-minute and 15-minute contracts shown separately.
Retention
The wallets leave, but the trading stays. Shorter contracts shed wallets fastest, yet the dollars and trades that remain are stickier than the head-count — the machine keeps trading after the humans go.
Shorter Contracts Lose Wallets Faster
Share of each cohort’s wallets still trading by weeks since first trade, split by product (5-minute / 1-hour / daily / weekly). Complete-week cohorts are recomputed daily; shorter contracts decay fastest.
The 5-Minute Funnel Is Shrinking
Weekly cohort size and week-1 / week-4 retention for each signup cohort — five duration funnels (5m / 15m / 1h / 4h / weekly), recomputed daily with incomplete future lags left blank.
The Wallets Leave, but the Trading Stays
The same 5-minute cohorts measured three ways — wallets, trades and dollars — as a percentage of their week-0 level. Dollars and trades outlast wallets.
From schema to shipped research, three calls
106 ClickHouse tables across 13 domains — trades, markets, wallets, candles, fees — with column-level types and comments. One call to see everything.
Raw ClickHouse SQL over 934M Polymarket trades and 55M Kalshi market snapshots. Up to 10K rows per call, 30-second ceiling, zero infrastructure.
The study's market, wallet and fee exhibits all trace back to queries on these tables. Same keys, same tables — the figures are reproducible.
Ship it on Surf's live API
The same tables and endpoints this page runs on — free tier, no card required.
API docs ↗Need a data feed, not a demo?
Talk to us about enterprise access, custom pipelines, and SLAs.
Talk to us