How We Price, Call and Grade Markets

A Kalshi or Polymarket contract pays $1 if its event happens, so a YES price of 55¢ means the market is pricing the event at about 55%. Each engine produces its own probability for the same contract. Edge is the difference, in percentage points: a model at 62% against that 55¢ contract is +7 points on YES; a model at 48% is +7 points on NO.

Every call carries the price range where the edge still holds — “Buy YES up to 62¢” — so you know the point where the trade stops being worth taking. An edge is a disagreement with the market, not a promise. Markets are often right.

Each tool owns its own thresholds, calibrated on its own data; we never carry one tool's rule to another. The tool's page states its ladder. Two examples:

• Commodity and crypto edge tools run live edge calculations strike by strike and label a row BUY YES or BUY NO only when its edge clears that engine's actionable bar. Everything else is PASS. A PASS row appears on the public grid only when its edge is at least 3 points — the near-misses that could flip on the next tick — so noise stays off the page. The engine still records every strike for calibration.
• The Combo Edge Builder prices a same-game combo with correlation between legs and returns one of 7 verdicts, best to worst: SMASH → PLAY → LEAN → RISK → NO VALUE → PASS → RUN.

PASS is a real answer. Most markets, most days, are priced about right, and saying so is part of the job.

Prices are read from the exchanges' public market data — Kalshi first, Polymarket where Kalshi doesn't list the market, and both side by side wherever we compare venues. Commodity and crypto tools run real-time market analysis and update in real time; slower boards refresh on a schedule and show their as-of time.

When a read fails, the page says the price is unavailable. It never fills the gap with an estimate. How each contract settles — the source, the cutoff, the edge cases — is spelled out per series on How It Settles.

Kalshi charges a trading fee on each fill, not on winnings: 7% × contracts × price × (1 − price), rounded up to the cent, so it peaks at 1.75¢ per contract on a 50¢ contract and shrinks toward zero at the extremes. It is paid at entry whether the position wins or loses; there is no settlement fee. Resting (maker) orders pay no fee on most standard markets.

The fee matters most near 50¢, where it can eat a thin edge whole. Run any price through the Kalshi fee calculator before entering. Fee facts across the site come from one source, checked against Kalshi's published fee schedule as of 2026-09-23.

Every engine signal and every Play of the Day is written to our database with a timestamp before the market settles, then graded automatically against the exchange's own settlement. Nothing is edited after the outcome is known. Engines that lose money stay on the record.

Results are measured per single contract: a YES bought at P cents returns 100 − P on a win and loses P on a loss. No sizing, no compounding, no leverage — sizing is a separate decision, and the Kelly calculator is where you make it. Alongside win rate and P/L we publish the Brier score, because win rate alone flatters a tool that only takes heavy favorites.

• Track record — every settled engine signal and Play of the Day, winners and losers.
• Tool leaderboard — each predictive tool scored over 7, 30 and 90 days and all-time, with the math on leaderboard methodology (Brier, Wilson intervals, ROI, sample-size badges, voids).
• Calibration backtest — predicted versus realized hit rate, by decile, for the commodity engines.
• NFL calibration ledger — the football model is validated walk-forward and graded on its own card; how it's built is on the NFL model page.

Prediction Market Statistics collects citable numbers about the markets themselves — how accurate they are and what fees cost — updated monthly with a CSV download, free to reuse under CC BY 4.0: credit PredictionMarketsPicks and link the page. More feeds, in JSON and CSV, are in the open data catalog.

The same engines sit behind the 29 tools on the site and the 45-tool MCP server for AI agents, with a public API for the feeds. What we publish is derived: probabilities, edges, tiers and grades.

PredictionMarketsPicks is the website and data layer: the tools, the engines and the graded record. The 7 Oracles are the media and analysis layer, and every article carries its author's byline. Ownership, funding and how we make money are on About; the analysts and their credentials are on the team page. Referral relationships are listed in the affiliate disclosure — no platform pays for coverage, a rating or a call.

When we get a fact wrong, we correct it in place and set a visible “Updated” date on the piece. Graded results are never revised — a losing call stays a losing call. Send a correction to predictionmarketspicks@gmail.com. The policy also lives on About.

Prediction markets are event contracts and a contract can go to zero. Nothing here is financial advice: the tools show what the market is pricing and what our models read, and the decision is yours. Trade responsibly.