Mispricing Scanner
A Bayesian engine scans prediction markets daily for mispricings. Agent A finds the same event on the other platform (LLM-validated) and compares the price — when a validated twin exists. Agent B is Claude NLP estimating probability from the question. When the engine diverges from the market by 5pp+, the contract is flagged with Kelly sizing; 3–5pp near-misses go to a free watchlist band.
Runs every morning; results post by ~10 AM ET. How it works →
Quick Answer
the Mispricing Scanner publishes a Kalshi contract only when the same event is found on the other platform and a resolution-criteria check confirms the two prices are comparable. That check has not cleared across the full scan history, so the board is currently empty — the standalone model estimate that used to fill it did not beat the market price when reviewed against settled results, and is no longer graded or published. The tool is being rebuilt for the 2026 midterms as a markets-versus-pundits tracker.
If you're asking whether Kalshi/Polymarket mispricing flags actually pay out: the scanner flags markets where its Bayesian engine diverges from the platform price by 5pp or more. HIGH CONFIDENCE requires 8pp+ and a validated same-event twin on the other platform pulling the same direction — most contracts have no such twin, so many days produce no HIGH flags at all. That scarcity is the design. Everything else lands at MEDIUM or LOW, with a free 3–5pp watchlist band of forming edges below the flag bar. The Kelly fraction shown is quarter-Kelly, sized on a cost-adjusted entry price, not the headline edge.
Caveat: a flag is a signal to investigate, not a guaranteed trade. The engine now validates cross-platform matches with an LLM same-event gate, so the old “same name, different event” pairs are filtered out — but still read both contract pages and confirm the resolution criteria before sizing.
What Is the Mispricing Scanner?
The Mispricing Scanner is a daily automated scan of Kalshi and Polymarket binary markets. It publishes a contract only when the same event is listed on the other venue and a resolution-criteria check confirms both contracts settle on the same terms, so their prices can be compared at all. Without that confirmation there is no flag — two contracts with similar names and different settlement rules are not the same market.
Kalshi anchors the primary scan because it lists the broader set of contracts our other tools already cover, not because of any difference in legal access — both venues are open to US traders. A Polymarket price shown here comes from its international order book, a separate exchange from Polymarket US with its own liquidity and settlement, so the two are not interchangeable.
How the Mispricing Scanner works
For each platform: Agent A looks for the same event on the other platform — a title pre-filter, then an LLM gate confirming both contracts resolve on identical criteria before the prices are compared (no validated twin → no cross-reference). Agent B asks Claude for an independent probability estimate. Both feed into a Bayesian weighted average. Divergences of 5pp+ are flagged and 3–5pp near-misses go to the free watchlist band; HIGH CONFIDENCE needs 8pp+ and a validated twin agreeing, which is why it is rare.
When to Use This Tool
Check the scanner in the morning once the day's scan has posted — usually by 10 AM ET. HIGH CONFIDENCE flags are the priority when there are any; otherwise work the widest divergences down. Verify the cross-platform match is the same event, check for overnight news that could explain the gap, then size with the quarter-Kelly fraction shown. This is a signal to investigate, not a guaranteed trade.
Read the Methodology
The scanner's design decisions are written up in full, including the ones that make the board publish less rather than more: how the Polymarket Mispricing Scanner works is the place to start. From there, what HIGH CONFIDENCE actually requires explains why that tier is rare by construction, and why the scanner fails closed covers what happens when the model cannot reach a confident estimate — it publishes nothing instead of guessing. If you are weighing whether any of this is tradeable at all, why most prediction market arbitrage isn't real is the honest version of the answer.
MARKET INTELLIGENCE
Find markets where the crowd got it wrong
What is this?
The crowd is often right. But not always. This scanner identifies prediction market contracts where the price seems significantly out of line with real-world data — news, historical base rates, or statistical models.
Think of it as a lie detector for market prices. When a market is way overpriced or underpriced relative to reality, this flags it. You still have to decide — but this gives you the shortlist of where to look.
Real-World Example
→ The Flag
Historical data shows government shutdowns resolve within 14 days 84% of the time. A market asks "Shutdown lasts more than 21 days" — priced at 55¢ (the crowd thinks 55% likely). That's a mispricing signal.
The scanner flags the gap: historical base rate says this should be worth about 16¢, not 55¢. You buy NO at 45¢ (inverse of YES at 55¢) and wait.
✅Action: Paused: the scanner publishes nothing while it is rebuilt for the 2026 midterms as a markets-versus-pundits tracker. The method above still works by hand — compare a price with a base rate before you take a position.
Bottom line: The crowd panics, overreacts, and misprices markets. The automated scan is paused; the base-rate check is not.
Full guide →Live engines, graded automatically at Kalshi settlement — winners and losers, never edited.
- Bitcoin Edge1,011 graded59% win+$56.24 netavg CLV +2.0pp
- Oil Edge618 graded46% win+$4.53 netavg CLV −5.9pp
- Gold Edge514 graded60% win+$34.09 net
- Weather Edge257 graded35% win−$8.97 net
- NHL Goal of the Day210 graded57% win+$4.08 netavg CLV −0.7pp
Loading today's scan
Checking Bayesian engine results...
Read the fine print
Half the edge in prediction markets is knowing exactly how a contract resolves. We read the official resolution rules of the markets traders care about and flag the traps — which data print counts, the exact metric, rounding, and settlement timing.
Today's trapWill the upper bound of the federal funds rate be above 4.75% following the Fed's Sep 15, 2027 meeting?Resolves YES if the federal funds rate upper bound published on the Fed's official website is strictly greater than 4.75% after the September 15, 2027 FOMC meeting.3 traps
- Strict 'greater than' threshold — 4.75% itself fails — A rate held exactly at 4.75% resolves NO, so a trader who thinks the Fed 'stays at 4.75%' is holding a losing long position, not a winning one.
- Sep 15, 2027 may not be an FOMC meeting date — FOMC meetings follow a fixed schedule and September 15, 2027 may not align with an actual meeting; if no meeting occurs on that exact date, resolution criteria may never be triggered or could be voided entirely.
- Official Fed website is the sole data source — Only the rate as published on the Federal Reserve's official website counts — discrepancies from other sources (CME, Bloomberg, etc.) are irrelevant to resolution.
Will the upper bound of the federal funds rate be above 3.50% following the Fed's Sep 15, 2027 meeting?Resolves YES if the upper bound of the federal funds rate target range, as published on the Federal Reserve's official website, is strictly greater than 3.50% after the Fed's September 15, 2027 meeting.3 traps
- Strictly greater than, not ≥ 3.50% — A rate of exactly 3.50% does NOT resolve YES — the upper bound must be above 3.50%, so a position holder needs the rate at 3.75% or higher.
- Sep 15, 2027 may not be an FOMC meeting date — If no Fed meeting occurs on that exact date, the resolution condition may never be triggered or could be voided/delayed, creating ambiguity about which meeting qualifies.
- Source is Fed's official website only — Only the rate published on federalreserve.gov counts — any other data vendor (Bloomberg, CME, etc.) is irrelevant, and a discrepancy could affect resolution.
WTI 15-minuteResolves YES if the WTI Oil close price on the Pyth PYTHOIL feed at the end of the 1-minute candle closing at the top of the settlement window is at least (≥) the close price of the 1-minute Pyth PYTHOIL candle at the start of the 15-minute window, measured in USD/Bbl.4 traps
- EST vs EDT timezone mismatch in rules — The rules cite 6:45 AM EST for the end candle but 6:30 AM EDT for the start candle — EST and EDT differ by one hour, so a literal reading puts the two reference candles 16 hours apart rather than 15 minutes; traders must confirm the intended timezone is consistent (almost certainly both EDT or both EST) to know the true comparison window.
- Pyth PYTHOIL is the sole data source — Settlement uses only the Pyth Network's PYTHOIL oracle feed, not CME, ICE, or any traditional WTI benchmark — Pyth outages, stale prices, or deviations from spot could affect resolution independently of the broader WTI market.
- Exact candle-close print, not intra-candle price — Only the 1-minute candlestick close at each boundary timestamp is used; intra-candle highs/lows and the open are irrelevant, so a price that crosses the threshold during the minute but closes below it resolves NO.
- At-least (≥) comparison: hitting threshold exactly resolves YES — The rule uses 'at least', so a close price exactly equal to the opening reference candle's close is sufficient for YES — unlike strictly-greater-than contracts in the same series family.
Silver 15-minuteResolves YES if the Pyth Silver (USD/t.oz) 1-minute candlestick close at the end of the 15-minute window is greater than or equal to the Pyth Silver 1-minute candlestick close at the start of that window.4 traps
- Pyth is the sole data source — Settlement uses Pyth's SILVER feed exclusively — not CME, Bloomberg, or any other silver benchmark — so Pyth-specific outages, price anomalies, or feed latency directly affect resolution regardless of where spot silver actually trades.
- 1-minute close, not OHLC average or spot mid — Only the closing print of the 1-minute candle at each boundary is compared; the high, low, or average price within those minutes is irrelevant, so a candle that spikes but closes flat could swing resolution unexpectedly.
- Timezone mislabelling: start is EDT, end is EST — The rules label the opening candle 6:30 AM EDT and the closing candle 6:45 AM EST — EDT and EST differ by one hour, so if taken literally these are not 15 minutes apart; traders must confirm whether both timestamps are intended to be the same timezone offset to know the true window boundaries.
- At-threshold resolves YES (≥, not >) — The comparison is greater than or equal to, so an identical close price at both boundaries is sufficient for YES — a position expecting a strict price increase to win would be wrong.
Gold 15-minuteResolves YES if the Pyth GOLD (USD/t.oz) 1-minute candlestick close at the end of the 15-minute window is at or above the Pyth GOLD 1-minute candlestick close at the start of that same 15-minute window.4 traps
- Pyth is the sole data source — Settlement uses only the Pyth GOLD feed (USD/t.oz), not spot prices from any exchange or other benchmark; discrepancies between Pyth and other gold price sources are irrelevant.
- 1-minute candle close, not OHLC high/low — Only the closing print of each 1-minute candlestick is compared — intra-minute highs, lows, or opens at either boundary have no bearing on resolution.
- At-least (≥) comparison: exact tie resolves YES — The rule requires the close at the end of the window to be at least the close at the start, so if both prints are identical the contract resolves YES, not NO.
- EST vs EDT timezone discrepancy in rules — The rules cite the opening candle at 6:30 AM EDT and the closing candle at 6:45 AM EST — these are different UTC offsets (EDT = UTC−4, EST = UTC−5), so traders must confirm which UTC timestamps the Pyth feed actually uses to avoid referencing the wrong candles.
SOL 15 minResolves YES if the 60-second simple average of CF Benchmarks' SOLUSDRTI ending before the later 15-minute window's cutoff is at least the 60-second simple average ending before the earlier window's cutoff, on the settlement date.4 traps
- CF Benchmarks SOLUSDRTI is the sole source — Only CF Benchmarks' SOLUSDRTI prints are used — no other SOL/USD index, spot price, or exchange feed is a valid substitute, so traders must track that specific index.
- Strict 'at least' means tie resolves YES — The comparison is ≥, so if both 60-second averages are exactly equal, the contract resolves YES, not NO — a tie does not favour the short side.
- Each price is a 60-second simple average, not a spot print — Resolution uses the mean of sixty individual one-second SOLUSDRTI values before each cutoff, so a single-second spike or dip near the boundary does not alone determine the outcome.
- Two distinct windows define the comparison, not one settlement price — This is a 15-minute cadence series comparing the average before the later cutoff against the average before the earlier cutoff (15 minutes prior), so traders must monitor both windows rather than a single settlement moment.
ETH 15 minResolves YES if the 60-second simple average of CF Benchmarks' ETHUSDRTI ending before the later window's boundary is at least the 60-second simple average ending before the earlier window's boundary, both on the settlement date.4 traps
- CF Benchmarks ETHUSDRTI — not spot price — Settlement uses CF Benchmarks' ETHUSDRTI (a real-time index), not any exchange's spot price or BRTI; a trader using a different feed may see a divergent value.
- Exact 60-second average, not a single print — Each "price" is the simple mean of sixty individual one-second ETHUSDRTI observations, so a brief spike or dip within that minute window affects the result in a way a single timestamp snapshot would not.
- "At least" — equality resolves YES — The threshold is inclusive (≥), so if both 60-second averages are identical the contract resolves YES, not NO; traders pricing near-flat scenarios must account for this.
- 15-minute cadence — window boundaries are series-fixed — This is a 15-minute series; each contract's two windows are always 15 minutes apart and anchored to the top-of-the-hour cadence, so the relevant 60-second blocks are the minute immediately before each 15-minute mark, not some arbitrary clock time.
Will **real GDP** increase by more than 0.5% in Q4 2026?Resolves YES if the BEA's Advance Estimate of real GDP (seasonally adjusted and annualized) for Q4 2026 shows growth strictly greater than 0.5%.3 traps
- Advance Estimate only — revisions ignored — Only the first BEA print (Advance Estimate) counts; even if subsequent Second or Third Estimates revise GDP above or below 0.5%, they have no bearing on resolution.
- Annualized rate, not quarterly rate — The BEA's SAAR figure annualizes the quarterly change (~4x magnification), so a raw quarterly gain of ~0.125% would equal the 0.5% threshold — traders must not confuse the two measures.
- Strictly greater than 0.5 — exact threshold excluded — A print of exactly 0.5% resolves NO, since the rule requires growth *more than* 0.5, making the boundary unambiguous but easy to overlook.
Will the rate of CPI inflation be above 3.4% for the year ending in November 2026?Resolves YES if the BLS-reported CPI 12-month inflation rate for November 2026, rounded to one decimal place, is strictly greater than 3.4%.3 traps
- One-decimal place BLS print is binding — The contract uses the one-decimal place figure published by the BLS (e.g., 3.4% vs. 3.45% underlying), so a reading that rounds down to 3.4% resolves NO even if the unrounded figure is above 3.4%.
- Strictly above 3.4%, not at or above — The rules say 'more than 3.4%,' meaning an exact one-decimal print of 3.4% resolves NO — only 3.5% or higher resolves YES.
- Initial release vs. later BLS revisions — The rules do not specify whether the first release or a subsequent revision is used; if the BLS revises the figure after initial publication, it is unclear which print governs settlement.
XRP 15 minResolves YES if the 60-second simple average of CF Benchmarks' XRPUSDRTI ending before the later window's top-of-interval is greater than or equal to the 60-second simple average ending before the earlier window's top-of-interval, with both windows determined by the settlement date of the specific contract.4 traps
- Source is CF Benchmarks XRPUSDRTI specifically — Settlement uses CF Benchmarks' Real-Time Index (XRPUSDRTI), not spot prices from any exchange or other index provider — discrepancies between venues can affect whether the position resolves as expected.
- 60-second simple average, not a single print — Each window resolves on the arithmetic mean of sixty one-second XRPUSDRTI values leading up to the interval boundary, so a brief spike or dip may not move the average enough to flip resolution.
- Comparison is ≥ threshold, so ties resolve YES — The later window only needs to equal (not strictly exceed) the earlier window's average, meaning a flat market across the 15-minute gap still resolves YES — not a push or NO.
- Cadence is every 15 minutes; windows are series-fixed relative offsets — Each contract in this series compares the window ending at the top of one 15-minute interval to the window ending 15 minutes earlier; the specific clock times vary per contract, so traders must confirm the exact boundaries for their contract.
Related Tools
Go Pro · 26% off your first year
Every edge engine on the site, live — $150/yr
Pro opens every engine on the site — the full silver, gold, oil and bitcoin strike grids that price commodity contracts in real time, Thee Oracle, Theta Edge, and the full cross-venue price-gap board.
$111 for the first year, then $150/yr. Code applies automatically. Ends Dec 31. Monthly is $14.99/mo.
Every signal graded in public, losers left on the board. See the record →
Kalshi Mispricing Scanner — frequently asked
Which Kalshi binary markets are mispriced right now?
None that we are willing to publish. A contract only reaches this board when the same event is found on the other platform and an LLM gate confirms both contracts resolve on identical criteria before their prices are compared. Measured across the full scan history, that validated cross-platform check has not once completed — most Kalshi contracts simply have no Polymarket twin resolving on the same criterion. We used to fill the gap with a standalone language-model probability estimate and call the difference an edge. Reviewed against settled results, that estimate did not beat the market price, so it is no longer graded and no longer published. An empty board is the honest output of that rule, not an outage.
What is the Kalshi mispricing scanner?
A daily automated scan of Kalshi and Polymarket binary markets, looking for contracts trading away from a price that can be independently corroborated. The only corroboration it now accepts is a validated same-event twin on the other platform: a title pre-filter narrows candidates, then an LLM gate confirms both contracts settle on the same criterion before any price comparison happens. No validated twin means no flag. The tool is being rebuilt for the 2026 midterms as a markets-versus-pundits tracker, which compares the live market price against named forecasters and ratings outlets instead of against a model of our own.
Do Kalshi vs Polymarket mispricing flags actually pay out?
We stopped claiming they do. The scanner's original thesis was that our probability model could out-predict a liquid market. Checked against settled outcomes by price band, the claimed win rate exceeded the actual win rate in every band — the model was not beating the market, it was disagreeing with it. Rather than retune a losing signal, the tool now grades nothing that lacks a validated cross-platform price check, and publishes nothing it will not grade. The rebuilt version inverts the claim entirely: the market price becomes the benchmark and named pundits are the comparison, which is a scoreboard rather than a trade signal.
How does the scanner engine detect mispricings?
It compares a contract's price against a corroborating price for the same event on the other venue. A title pre-filter narrows candidates, then an LLM gate confirms both contracts resolve on the same event and criterion — without that confirmation there is no cross-reference and no flag, because two contracts with similar names and different settlement rules are not the same market and their prices are not comparable. The engine previously also weighted in a standalone language-model estimate; that input no longer produces a published flag on its own.
Why is the mispricing board empty?
Because the gate that would put something on it has not cleared. A flag requires a validated same-event twin on the other platform, and across the entire scan history no scanned Kalshi contract has had one that passed the resolution-criteria check. The alternative — publishing our model's disagreement with the market as though it were an edge — is exactly what the settled record showed did not work. Showing nothing is the correct output of an honest rule, and the board will refill only if the cross-platform check starts clearing.
Is the Mispricing Scanner free to use?
The board itself is a Pro tool at $14.99 per month, though it is currently publishing nothing while the tool is rebuilt for the midterms. The free tools — EV Calculator, Probability Converter, and Combo Edge Builder — cover the core math if you want to check a single contract yourself, and the Combo Edge Builder is fully free with no leg cap.
What is the difference between the Kalshi scan and the Polymarket scan?
They swap which venue anchors the price check: the Kalshi scan looks for the corroborating price on Polymarket, and the Polymarket scan looks for it on Kalshi. Kalshi is the primary scan because it lists a broader set of US-available contracts and its markets are the ones our other tools already cover — not because of any difference in legal access. Both venues are open to US traders. Note that a Polymarket price we quote comes from its international order book, which is a separate exchange from Polymarket US and does not share its liquidity or settlement, so the two prices are not interchangeable.