Market Guy
HomeOddsDaily RecapLearnPricing
Log inSign up
Market Guy

Evidence-based prediction market research — Polymarket & Kalshi, no hype.

Product

  • Live Markets
  • Odds
  • Daily Recap
  • Daily Challenge
  • Pricing

Learn

  • All guides
  • Research library
  • Learning paths
  • Glossary
  • How markets work
  • Reading the odds
  • Smart money

How it works

  • Methodology
  • Model card
  • Data sources
  • Track record

Company

  • Account
  • Insights
  • Privacy

Market Guy · Research tooling, not investment advice.

Data sources

Where every number on this site comes from, how often it refreshes, and what happens when a source fails.

Last updated Sep 1, 2026

Market data

Prices, volumes, rules and outcomes come from the venues themselves — Polymarket and Kalshi, through their public interfaces. Nothing on a page is fetched from a venue while you read it: pages read a copy that is refreshed in the background.

That is a deliberate trade, and it cuts both ways. A page never depends on a third party being reachable at the moment you open it, and a burst of traffic never turns into a burst of requests at the venue. In exchange, a price here can be a few minutes behind the venue. For anything time-critical, the venue is the authority — not this site.

How current a number is

Different data has different shelf life, and the refresh follows that rather than one blanket interval:

  • Prices refresh throughout the day, several times an hour. A quoted price is therefore current to within minutes, not seconds — for a market that moves on news, check the venue before acting on a number here.
  • Market metadata, rules and the set of open markets refresh continuously but less often than prices; a market that opened in the last few minutes may not be listed yet.
  • Cross-venue comparison data refreshes on the same order as prices.
  • Underlying and context data is cached per market and can be a few minutes old.
  • Resolved outcomes are picked up after a market settles, not at the moment it settles.

Underlying and context data

For markets with a tradeable underlying, context is computed from public endpoints that need no API key: Coinbase, Bybit and OKX for crypto spot, funding and open interest; Deribit for options-implied probabilities where the instrument exists; ESPN for sports schedules and standings.

The keyless constraint is the reason macro context is missing rather than approximated. It is a stated gap, not an oversight.

When a source is unavailable

What you see when something upstream is down matters for how much you trust the number in front of you:

  • A failed refresh keeps showing the last good data rather than an empty page. A stale number is more useful than a broken one — but it is still stale, and nothing on the page will look different.
  • When no matching contract is found on the other venue, the comparison says so instead of guessing at a near match.
  • When a market has too few qualified wallets, no signal is shown. An absent verdict means the evidence was not there, not that the wallets disagreed.
  • When context data for an underlying cannot be retrieved, the report omits it rather than falling back to an approximation.

What is stored

Market data, resolved outcomes, wallet statistics, the verdicts shown at a point in time and the generated analyses are kept so that later checks are possible. A probability that was not recorded when it was made cannot be scored after the fact, and once a market resolves, the price at the moment of a statement is no longer reconstructable — so the statement and that price are written down while they still hold.

This is what makes the forecast track record possible at all. Without it, any claim about past accuracy would be an assertion rather than a measurement.

More on how this works

  • How we research a market — The steps between a live contract and a probability estimate, and the limits of each one.
  • Model card — Which model writes an analysis, how it is configured, and where it fails.
  • Forecast track record — How the signals have actually performed — including where the sample is still too small to say.