Methodology

Implied probability

Both venues trade binary contracts that settle at $1 or $0. A price is therefore already a probability. Internally every price is normalized to a number between 0.0000 and 1.0000 and stored that way; the interface displays it as a percentage or in cents, but the two are never mixed in the same calculation.

The quoted probability is the midpoint of the best bid and best ask when both exist. When only one side is quoted, the last traded price is used and the spread is shown so you can see how much the midpoint is worth. A wide spread means the midpoint is an estimate, not a price anybody can trade at.

Movement

Movement is always reported in percentage points, never as a relative percentage. A market moving from 41% to 55% has moved +14.0 points. Describing that as "+34%" is arithmetically true and practically misleading, so this site does not do it.

Four windows are maintained — 1 hour, 6 hours, 24 hours and 7 days — comparing the current probability against the probability at the start of the window, taken from stored snapshots or from the venue's own reference price when one is published.

Movement significance

Raw movement rewards illiquid markets where a handful of dollars moves the price twenty points. The significance score combines the absolute move with four dampening components: the logarithm of 24-hour volume, liquidity relative to a reference market, the tightness of the spread and recent trade activity against what the market normally does. The result is normalized to 0–100.

Markets below a minimum 24-hour volume floor are excluded from mover rankings entirely, which is why a board can be shorter than you expect on a quiet day.

Trade size and trade impact

Notional value is execution price multiplied by quantity, with each venue's own reported values retained alongside. The largest trade and the most consequential trade are different questions, so they are ranked separately.

Impact combines the logarithm of notional with the realised price change from just before the trade to five minutes after it, adjusted for the market's liquidity. Trades are labelled Normal, Elevated, Large or Exceptional. A large trade that moved nothing scores low, which is usually the more interesting fact.

Matching contracts across venues

Two contracts are never matched on title similarity alone. Each market is reduced to structured features: primary and secondary entities, the action or event, any numeric threshold and its comparison operator, the unit of measurement, geography, the start, end and resolution times, outcome polarity and the resolution source.

Candidate pairs are scored deterministically:

Certain disagreements are hard conflicts that override the score no matter how similar the wording: a different election round, a different candidate, a different date cutoff, a different threshold, one contract asking "at least" where the other asks "more than", a materially different resolution source, a different geographic unit, or a settlement condition that cannot be normalized to the other.

Equivalence grades

Grade A — Equivalent. Score at or above 0.90, no hard conflict, aligned dates, outcome mapping and settlement rules, and a semantic verification pass that found no material discrepancy. Only Grade A pairs are used for theoretical edge.

Grade B — Near equivalent. The same event with one or more meaningful ambiguities remaining. The discrepancy is shown; executable arbitrage is not advertised.

Grade C — Related. The same topic or event family, but not equivalent enough for direct contract comparison. Useful for sentiment only.

A language model participates only as a verifier on candidates the deterministic scorer already considers plausible. It can downgrade a match or flag a difference. It cannot promote a pair over a hard conflict, and it never computes a price, a gap or a score.

Theoretical cross-market edge

Edge is never computed from displayed midpoints. Both directions are tested — Kalshi YES with Polymarket NO, and Kalshi NO with Polymarket YES — using executable ask ladders. Kalshi publishes only bids in its binary book, so the opposite side's asks are derived from reciprocal YES/NO pricing rather than assumed.

Because the top quote rarely covers a whole position, the order book is walked at $10, $50, $100, $250, $500, $1000 to show what each size could actually achieve. A depth level that cannot be filled is reported as unavailable rather than extrapolated.

Divergence and lead/lag

Two kinds of divergence are tracked: price divergence, where the venues currently disagree; and movement divergence, where one venue repriced materially before the other and the gap may or may not close afterwards.

A recorded lead is a statement about sequence, not cause. Markets and conversation can both respond to the same outside event, and often do.

Generated explanations

Explanations separate what is established from the data, what is a plausible interpretation, and what is speculation the evidence does not support. The application computes an evidence score from cross-venue confirmation, trade-flow signal, temporal alignment, source diversity and data completeness, and that score — not the model — determines whether evidence is labelled Weak, Moderate or Strong.

The model receives a fixed evidence package and may only write language around it. Its output is validated: every percentage, dollar amount, date and count must correspond to supplied evidence. A draft that fails is retried once with the violations attached, and if it fails again the deterministic text is published instead.

Data refresh

• Market state, trades, gaps and rankings: every 5 minutes.
• Full market universe, new listings and matching: hourly.
• Public API responses: cached at the edge for about 30 seconds.
• Morning briefing: once daily, published between 08:00 and 11:00 New York time.

All timestamps are stored in UTC and displayed in America/New_York using IANA timezone data, so the switch between EST and EDT is handled by the platform rather than a hardcoded offset.

Limitations

• Volume and liquidity are reported as each venue publishes them and are not perfectly comparable between venues.
• A matched pair is a judgement, not a legal equivalence. Settlement wording governs, and it can differ subtly.
• Order books change faster than any page can render. Depth simulations describe a moment that has already passed.
• Historical snapshots begin when this service began tracking a market, not when the market opened.
• Markets with very little volume are excluded from rankings and may have stale quotes.

What this is not

ProudGlobeHead Markets is an informational analytics service. Nothing on it is financial advice, a recommendation, a solicitation or an offer to trade, and none of it is personalized to your circumstances. Prediction markets can move against you and a position can lose its entire value.