How Prediction Pantheon sources market data

Overview

This page describes, as precisely as we can, where our market data comes from, what we do to it, and what its limits are. If something here stops being accurate, treat it as a bug and tell us.

Where the data comes from

Market data is ingested from the public APIs of the venues we cover — currently Polymarket and Kalshi. We use publicly documented endpoints and do not use private, paid or scraped feeds. Platform facts in our reviews (fees, funding methods, regional availability, regulation) come from each venue's own published documentation and from hands-on testing.

What we ingest and display

Market and contract identifiers, question text, outcome labels and category.
Current YES pricing, and where the venue publishes it, volume and liquidity.
Deadlines, expiry timestamps and the venue's settlement or rule summary where available.
The canonical link back to the contract on the venue.

Question wording shown on the feed may be a shortened or normalised version of the venue's own title, so that two venues' phrasings can be compared side by side. Market detail pages show the original venue question, and every card links back to the source contract.

Refreshes, delays and outages

Prices are refreshed by a scheduled pipeline rather than a live socket, so what you see is a recent snapshot, not a tick-by-tick quote. Pricing refresh is scheduled every three minutes. Deeper ingest, matching, scoring, and rebuild work runs separately and is not part of the three-minute pricing freshness path.

If an upstream venue is unavailable, we keep the last successful values and mark the data as delayed rather than inventing a price.
If a field cannot be established truthfully, it is shown as unavailable rather than filled with a placeholder number.
Expired or resolved contracts are filtered out of the feed rather than shown with a frozen price.

The current status and last successful refresh are published on our data status page and in the public feed response.

The venue is always authoritative

Prices shown here are venue-derived market-implied probabilities, not Prediction Pantheon forecasts. We do not run a proprietary forecasting model, and if we ever add one it will be separately labelled and documented. The venue remains authoritative for its own contract terms, availability, fees and resolution rules — always confirm there before acting.

How our reviews and articles are produced

Our platform reviews are written by a person, not generated or summarised by an automated system. Every review is based on first-hand experience: we open a real account on the platform where our region allows it, place or observe real trades, fund and withdraw through the actual methods each venue offers, and read the venue's own fee schedule, terms and contract rules. Where we could not complete part of that journey — for example a funding method unavailable in our region — we say so plainly rather than guessing.

Our learning and trading articles are grounded in real trades and real account use, not hypothetical scenarios. Examples, figures and step descriptions are drawn from actions we actually took on the platforms we cover, and any screenshot or number we cite can be traced back to a specific session. Where a detail has changed since we wrote it, we update the article and note the change rather than leaving stale information in place.

Reviews are authored by a named human reviewer, credited on the page.
Ratings reflect our own testing experience and are editorial judgements, not measurements.
Articles cite real trades and real account sessions; hypotheticals are labelled as such.
When we could not verify something first-hand, we describe it qualitatively or omit it rather than invent a number.

This human-first, first-hand approach is the reason our reviews can be slow to publish and why our coverage list grows gradually rather than overnight. We would rather have one review built on real use than ten assembled from other people's summaries.

Frequently asked questions

Is the data real time?

No. Pricing refresh is scheduled every three minutes, so it can lag the venue. Treat it as a recent snapshot.

Do you publish your own probability estimates?

No. Every price shown is derived from a venue's own market. We do not produce proprietary forecasts.

Can I access the data programmatically?

Yes. We publish an unauthenticated JSON feed for AI agents and researchers, documented in /llms.txt and /.well-known/ai-feed.json.