Latency data
Polymarket Latency Data
Latency is the venue's clock quality: how fast a reprice event lands in the record. The archive captures at 250ms, timestamps to the UTC second, and stores the timing metadata that makes latency itself measurable.
Figures measured as of 2026-09-24 on the published PolyOrderbooks archive.
Coverage
The timing layer
Every archived row carries the capture timestamp at 250ms granularity on the UTC grid, so latency data is implicit in the series: reprice events (a reference print, a goal, a new quote) appear in the row stream with an exact arrival second.
The websocket layer (~24,000 events/second observed) feeds the live channel, and the archive stores the observed event timing alongside books and trades — the material for feed-latency studies.
Measured
Measured timing facts
Measured reprice events land inside one or two 250ms frames of the underlying print — e.g., the ETH same-second repricing documented in the replay family — and 250ms capture resolves event order that second-level data merges.
Latency's honest framing: 250ms is the record's cadence, not a promise that the venue itself quoted in 250ms; the archive separates observed arrival from implied venue latency.
How to use
Using latency data
Latency studies slice the row stream by second and measure the gap between a print's appearance in the tape and the corresponding book change — the venue's structural response — and whether the delay is constant or concentrates at event peaks.
For feeds and dashboards, the archive's timing metadata tells you the latest frame you can trust at a given second — the input to any real-time display.
Access
Getting latency data
Honest
The honest framing
Record latency and venue latency are different questions: the archive measures when events arrived in its grid, and honest studies name which one they are measuring rather than collapsing them.
FAQ
What is in latency data?
Capture timestamps at 250ms on the UTC grid, plus event arrival timing for reprints, tapes, and websocket throughput.
How fast do events arrive in the record?
Measured reprints land within one or two 250ms frames of the underlying print; the websocket carries ~24,000 events/s.
Is 250ms the venue's latency?
No — 250ms is the capture cadence of the record; the archive separates observed arrival timing from any claim about venue quoting latency.