Binance correlation
Polymarket Binance Correlation Data
Polymarket crypto contracts settle against an oracle tracing the Binance-style market, so the venue's books and Binance's prints are coupled at the second. This page covers the correlation record: what the same-second repricing incident showed and where the data lives.
Figures measured as of 2026-09-24 on the published PolyOrderbooks archive.
The short answer
The venue follows the reference, measurably
Polymarket's crypto Up/Down resolves against a reference price sourced from Binance-consolidated markets, so a Binance print should appear in the venue's book within one or two 250ms frames. The measured ETH case in the same-second reprice (0.815 to 0.065 within one printed second) proves the coupling is tight.
Correlation here is mechanical, not incidental: the venue prices the reference, and the record shows the lag distribution between reference print and venue book reprint.
Measured
Measured linkage
The Aug-23 family event is the documented extreme: multiple coins collapsed together within 26 seconds on venue books as the reference moved — a cross-coin, cross-observation coupling that reads as a single correlated repricing cluster at 250ms.
For research, the reproducible measure is reprice-lag versus reference-print second on the shared UTC grid; the archive's latency and repricing series carry the inputs.
Honest
The honest framing
The linkage is real but bounded: the venue prices a reference, and arbitrage windows are what the lag distribution implies. A study should measure the lag and its variance (some minutes lag badly) rather than assert instant parity — the data supports the measurement, not the slogan.
FAQ
Is Polymarket correlated with Binance?
Mechanically yes — crypto contracts resolve against a Binance-consolidated reference, and books reprint within one or two 250ms frames of prints.
What proved the coupling?
The same-second reprice (ETH 0.815→0.065 in one printed second) and the 26-second multi-coin family repricing cluster.
Is parity instant?
No — the lag distribution varies by minute; studies should measure it rather than assume it.