How to read them
Read each venue as its own sample. Polymarket and Kalshi are collected through different pipelines with different identity models, and the whale tape on each has a different size floor, so the two columns describe two populations rather than one contest. A win rate is a win rate on buys only, because sells are never scored, and it says nothing on its own about return: the entry price decides the payoff, and the methodology shows why a 70 percent win rate at 80 cents loses money.
Watch the distinct-market count next to the event count. Ten recorded buys in one market are ten events and one outcome. A venue with many events per market is a venue where results are more correlated than the raw count suggests. And treat the unresolved share as a warning about recency: a window that ends yesterday will have many open positions, and the settled sample will lean toward markets that resolve quickly.
If you want the underlying rows rather than the aggregates, the same events are available through query_whale_events and backtest on the MCP server and REST API, where you can apply your own filters to a live copy of the history.
Frequently asked questions
What is a research snapshot?
A fixed export of recorded whale trades for a stated observation window, scored with the settlements observed at export time. It states its population filter, stake assumption, method version and export time, ships its aggregate table as JSON and CSV, and is never edited afterwards.
Why publish fixed snapshots instead of a live dashboard?
A live number cannot be cited. Settlements keep arriving, so the same window re-queried a month later gives different figures. A snapshot freezes one answer with its date so a reader can check it, quote it, or reproduce it.
Are these customer returns?
No. They are paper results on recorded trades: a fixed stake at the recorded price, held to settlement, gross of fees and slippage. They describe what the tape did, not what any account earned.
Can I download the data?
Yes. Every study links its JSON manifest and CSV table. The JSON carries a SHA-256 content hash so a copy can be verified against the published one.
How should I cite a study?
Give the study URL, the observation window, the export date, the venue, and the sizing assumption, and describe the figure as gross paper PnL on recorded whale trades. Each study page has a ready-made citation.