Powering prediction market intelligence with real-time APIs, historical data, orderbooks, arbitrage and whale tracking.
Joined July 2026
- Tweets26
- Following21
- Followers32
- Likes4
Pinned Tweet
Introducing ProbSights — the data layer for prediction markets.
Access:
• Real-time APIs
• Historical market data
• Full orderbook history
• Arbitrage opportunities
• Whale tracking
• Market analytics
Built for developers, researchers and traders.
🌐 probsights.com
The interesting part isn’t spotting a 3–5¢ gap.
It’s knowing which gaps are worth chasing.
Some spreads close in seconds.
Some persist.
Some widen.
Some vanish once fees + depth are included.
That’s why the real edge in cross-venue arbitrage is not the scanner itself.
It’s the distribution behind the signal:
→ how often this setup appeared
→ how long it stayed open
→ what size was actually executable
→ which venue usually moved first
→ how much edge survived after costs
A screenshot shows an opportunity.
History tells you whether it’s a strategy.
That’s the part we care about most at ProbSights.
Most of prediction markets arbitrage dies in seconds
While you only start switching tabs, bots already farm the spread
You don't only have to find the opportunity, but have enough fast execution to actually make profits
But I found the best way to automatize your cross-platform arbitrage:
@ProbSights scanner:
> Automatically matches BTC markets pricings across Kalshi and Polymarket
> Calculates the net profit after fees to be paid
> 60 days of spreads history, so you can check how often a gap like that actually closes
Two different platforms, regulated and non-regulated, on-chain and off-chain
Different userbase sets similar but not the identic odds, just farm the pricing gap
There're also arbitrage webhooks and whale tracking on greater plans
Prediction markets are quietly expanding beyond politics + sports.
Kalshi’s entertainment volume reportedly went from $43M in 2024 → $813M in 2026.
Even Dancing with the Stars has already traded ~$5.7M. MarketWatch
That creates a market-structure question I think is underrated:
Are “fandom” markets less efficient than financial ones?
In an NFL market, sophisticated models arrive quickly.
In reality TV?
The edge might come from:
→ fan communities
→ social sentiment
→ leaked information
→ contestant screen time
→ niche domain knowledge
Different market.
Different information advantage.
As prediction markets expand into thousands of new categories, the opportunity may not be predicting everything better.
It may be figuring out which categories price information the slowest.
That’s a dataset I’d love to study across Kalshi + Polymarket with ProbSights.
Kalshi just launched an S&P 500 perpetual future.
Most people will look at it as a new product.
I think the more interesting angle is what it does to cross-market information flow.
A macro headline hits.
The S&P perpetual moves first.
Then:
→ Fed markets reprice
→ recession odds move
→ political markets adjust
→ Polymarket catches up
→ related Kalshi markets converge
That sequence is where things get interesting.
The trade isn’t necessarily “predict the news better.”
It’s measure which market reacts first, which one lags, and whether the gap is still executable after fees + spread + depth.
That’s the kind of same-timestamp cross-venue research we want ProbSights to make much easier.
Prediction markets are becoming less isolated.
The edge may come from understanding how information travels between them.
Everyone is talking about Polymarket V2.
I think the more interesting trade is what happens during the transition.
When market infrastructure changes, bots don’t all adapt at the same speed.
That can create temporary inefficiencies:
→ stale pricing
→ routing differences
→ wider spreads
→ slower cross-venue convergence
→ unusual depth around fast-moving markets
So the question I’d backtest isn’t:
Is V2 better?
“For how long after V2 launches do old assumptions stop working?”
That transition period could be where some of the best short-lived edges appear.
At ProbSights, we’re especially interested in comparing that behavior against Kalshi at the same timestamps.
Infrastructure changes.
Market participants adapt.
The gap between those two is often where alpha lives.
Kalshi just printed $52.98B in September volume.
Sounds bullish for prediction markets.
But here’s the more interesting part:
Kalshi is also ending its Volume Incentive Program no earlier than Oct. 13. The Block
That gives us a rare natural experiment.
Before incentives end vs. after:
→ Does quoted depth fall?
→ Do spreads widen?
→ Does slippage increase?
→ Do Kalshi ↔ Polymarket gaps persist longer?
→ Does “volume” drop without actual market quality changing?
That distinction matters.
A market doing $50B+ in volume sounds incredibly liquid.
But traders don’t execute against headline volume.
They execute against the book.
At ProbSights, this is exactly the kind of cross-venue market-structure change we want to measure.
My guess: the most interesting number after Oct. 13 won’t be volume. It’ll be executable depth.
Kalshi is reportedly raising ~$1B at a $40B valuation.
A few months ago: $22B.
Everyone will talk about the valuation.
The more interesting part for traders:
Prediction markets are turning into an execution + data arms race.
As more capital enters:
→ obvious mispricings close faster
→ spreads get tighter
→ bots compete on milliseconds, not minutes
→ cross-venue price discovery matters more
→ historical market structure becomes more valuable
The easy edge was:
“Find a badly priced market.”
The next edge is:
know where the price is going before the other venue catches up.
That’s the research we’re increasingly interested in at ProbSights.
Everyone is backtesting prediction market strategies now.
But there’s a problem.
Most backtests assume:
“I saw YES at 42¢, so I bought at 42¢.”
That’s not how the trade happened.
Maybe there was:
$90 available at 42¢
$240 at 43¢
$1,100 at 45¢
Your $2k “42¢ entry” never existed.
And once bots start competing for the same edges, this gets even worse.
The backtests I trust need to answer:
→ what was actually on the book?
→ how much could you fill?
→ how much did the spread cost?
→ what happened after your trade size hit the book?
→ did real trades confirm the quoted liquidity?
Prediction market backtesting is moving from:
“Was the prediction correct?”
to:
“Could this strategy actually have been executed?”
Very different question.
Most people backtest prediction market strategies completely wrong.
They pull the final result.
Maybe they look at last traded price.
Maybe they use a clean chart.
Then they convince themselves they found an edge.
But live trading does not happen on clean charts.
It happens inside the orderbook.
That is why historical orderbook data matters.
We tested a simple ETH 5-minute market strategy using historical orderbook snapshots instead of just market outcomes.
The idea was basic:
When ETH 5m Up/Down markets started showing short-lived orderbook imbalance, enter only when the ask was still tradable, avoid stale markets, and exit before the final chaos near settlement.
Nothing exotic.
No magic indicator.
No “AI alpha.”
Just replaying what the book actually looked like at the time.
And the result was the part most traders miss:
The strategy looked amazing when tested on simple price history.
But when we replayed the historical orderbook, a lot of those “profitable” entries disappeared.
Why?
Because the price you think you could trade at was not always available.
The size was not always there.
The spread widened right when the signal appeared.
Some entries only worked on paper because the market moved before you could realistically hit it.
That is the difference between toy backtesting and real backtesting.
ProbSights gives you historical orderbook data for Kalshi and Polymarket markets so you can test against the actual market structure:
Best bid.
Best ask.
Orderbook snapshots.
Venue.
Market ID.
Timestamp.
Liquidity.
Spread.
Execution context.
For ETH 5m markets, this changes everything.
You can stop asking:
“Would this strategy have predicted the right outcome?”
And start asking the question that actually matters:
“Could I have entered at this price, at this time, with this liquidity, before the edge vanished?”
That is where serious traders separate from screenshot traders.
Because the edge is not just being right.
The edge is knowing whether your trade was executable.
Most people will discover their strategy is fake only after they lose money live.
ProbSights lets you find out before that.
Backtest ETH, BTC, and SOL Up/Down strategies.
Replay Kalshi and Polymarket historical orderbooks.
Compare venues.
Measure slippage assumptions.
See where your signal survives contact with the book.
The uncomfortable truth:
If your backtest does not use historical orderbook data, you may not be testing a trading strategy.
You may just be testing a fantasy fill.
ProbSights is for traders who want to know before they risk capital.
The next edge will not belong to the person with the prettiest chart.
It will belong to the person who knows exactly what the book looked like when the opportunity appeared.
Subscribe before you find out your “profitable” strategy was only profitable in hindsight.
Up/down polymarkets just got upgraded.
Trade 5m, 15m, & hourly crypto up/downs with time-weighted average pricing (TWAP).
To celebrate, we're giving away $1M in liquidity rewards across BTC, ETH, SOL, XRP, HYPE, BNB, & DOGE polymarkets.
I lost money on a market I was so sure about. Not because my thesis was wrong - it wasn't. I lost because by the time I put money in, the smart money was already out. I just didn't know it yet.
That's the story behind ProbSights.
Here's the thing nobody tells you about prediction markets: they're only "efficient" if you can actually see who's moving them. Whales - the wallets with real conviction, real capital, and often real information edges - move first and quiet. By the time a market swings and it's obvious to everyone, the whale that caused it has already locked in their price. You're not trading against the market at that point. You're trading against someone who already won.
ProbSights exists to close that gap.
What it actually does for you
It watches every meaningful position building across prediction markets in real time, and it doesn't just flag "big money moved." It flags smart money: wallets with a track record of being early and right, quietly accumulating in the same direction, before the odds have caught up.
So instead of finding out after the fact from a tweet or a chart that already moved, you see the position being built while it's still being built. That's the whole edge. Not predicting the future - just seeing the present a little sooner than everyone else.
What this actually changes for a trader
If you trade prediction markets seriously, you already know the real risk isn't "being wrong." It's being the last one in before a whale-driven repricing. ProbSights turns that risk into information you can act on:
You see accumulation patterns before the crowd does, so you're not buying the top of a whale-driven move.
You get context on which whales matter - a wallet with a strong track record moving into a position means something very different than a random large bet.
You stop mistaking noise for signal. A lot of "big moves" in these markets are nothing. The ones that matter look different once you know what to look for, and now you don't have to figure that out yourself.
Basically: you get to trade with the same information whales are quietly acting on, instead of finding out what they knew a day later when the market's already repriced around it.
Who this is actually for
If you've ever felt like prediction markets reward whoever has the best on-chain intel rather than whoever has the best read on the actual outcome - this is for you. You don't need to be a whale to trade like you're standing next to one.
We're just getting started, and the early signals we're catching already feel like the kind of thing that shouldn't have taken this long to build. If you trade these markets, you'll know exactly why this matters the first time you see a whale build a position you would've otherwise missed entirely.
Something interesting happened during an NBA game last week that most bettors never noticed.
Two sportsbooks had wildly different lines on the same matchup. One had the Lakers at -110. The other had the Warriors at +130. On their own, neither number looks unusual. Books disagree slightly all the time. But if you look at both lines together, they don't actually add up to a full 100% probability the way they're supposed to.
That gap is small, but it's real, and it means something specific: you can bet both sides of the same game, across the two books, and end up guaranteed to profit no matter who actually wins.
Here's roughly how it plays out. Put $546 on the Lakers at -110, and $454 on the Warriors at +130. That's $1,000 total, split across both outcomes. If the Lakers win, that $546 comes back as $1,043. If the Warriors win, the $454 comes back as about $1,044. Either way, you end up with roughly $1,043 on a $1,000 outlay, and it doesn't depend on the final score at all. It's decided the moment you place both bets.
That's arbitrage. Not a hot tip, not a lean, not "this team is undervalued." A mathematical gap between two books that briefly disagree with each other enough to create a locked-in return.
The reason it works is pretty simple once you think about it. No sportsbook is pricing games in a vacuum. Each one is adjusting odds based on its own customers, its own liability, its own internal models. Most of the time they land in roughly the same place. But every so often, two books drift far enough apart that a window opens, and for a little while, the math is just in your favor no matter what happens on the court.
The catch is that these windows don't stay open long. Once enough people spot the mismatch and bet into it, the odds move and the opportunity disappears. So the real skill isn't understanding that arbitrage exists - most people who've heard the term already get the concept. The hard part is actually catching it in time, across dozens of sportsbooks and thousands of games, faster than it takes to notice on your own.
That's basically the whole reason ProbSights exists. It watches odds across major sportsbooks at the same time, compares lines on the same events, and flags it the moment two books disagree enough to create a real opportunity. Instead of keeping ten tabs open and doing mental math during a game, you just get told when the gap shows up.
None of this is really gambling in the way people usually mean it. You're not picking a side and hoping. You're covering every outcome ahead of time, so the result of the game stops mattering to you financially. What matters instead is speed - whether you find the mismatch while it's still there.
Most people betting on sports have no idea this is happening under the surface, constantly, across games they're already watching. It's not a secret exactly. It's just easy to miss if you're only looking at one sportsbook at a time.
We are live now to provide arbitrage and whale tracking.
4 Structural Edges Top Weather Bots Exploit on Polymarket Every Day (Model Lag + Station Bias + Calibration + Behavioral)
a recent forensic study on 562 trades found something worth sitting with: 0.73% of wallets with a 97.8% win rate probably aren't better forecasters
they're faster. the edge is latency arbitrage, not superior modeling
here are the four structural edges top bots actually run on in 2026
[ 1. Model Update Lag ]
NWP models (GFS, ECMWF, HRRR) release on a strict schedule
a new run can shift the real probability significantly - but Polymarket doesn't reprice instantly, price only moves when someone actually trades
the lag has compressed, but it hasn't closed:
• major cities (NYC, London) - down to 5-15 min, bots got faster here
• secondary cities (Buenos Aires, Cape Town, Atlanta), can still run for hours
how to catch it:
• set alerts for key runs (ECMWF 12Z, HRRR)
• compare new forecast vs current price immediately
• enter if the shift clears your minimum edge (8%+ is a common threshold)
• wait for confirmation on 1-2 more runs to filter noise
[ 2. Station vs City Discrepancy ]
Polymarket resolves by the exact airport station - not the "city temperature" everyone sees in weather apps
the gap is often predictable:
• LaGuardia (KLGA) - frequently 1.5-3°C cooler than Midtown during sea breeze
• Paris Le Bourget (LFPB) - cooler than city center in anticyclonic conditions
• Hong Kong (VHHH), Burbank (KBUR) - each with their own known bias
how to exploit it:
• log historical deltas between station and city center, by season and synoptic pattern
• when the pattern confirms (sea breeze day, heat dome), adjust your probability
• the edge shows up when the market is still anchored to "city temp" instead of station temp
[ 3. Calibration Edge ]
same raw data, different traders can still produce completely different probabilities for the same bucket
this is the slowest-dying, most sustainable edge of the four - it doesn't require speed, just discipline
what real calibration includes:
• full ensemble distribution, not just the deterministic run - spread is your uncertainty
• station-specific bias correction based on historical error at that exact station
• correcting for ensemble underdispersion
• picking the right distribution shape (Normal works often, not always)
result: a meaningfully lower Brier score than someone trading raw model output
[ 4. Behavioral Edge ]
retail has predictable, exploitable flaws
• recency bias - anchoring tomorrow's bucket on today's temperature, even when models already show a shift incoming
• reverse favorite-longshot bias - cheap tail buckets ($0.10-0.15) often stay underpriced
• news-driven herding - after a heatwave or cold snap headline, retail piles into extremes the models already reverted from
exploit it by fading the crowd exactly when a notable weather event just hit the news
[ how to actually build this ]
• pick one edge and one time horizon - don't try to run all four at once
• define entry rules precisely (min edge %, model agreement, volume, lead time)
• backtest on historical Polymarket prices + real station resolutions, not city-average data
• track EV and Brier score, not just win rate
before you commit size to any of these, it's worth checking if the market already agrees with you under the hood
Market Strength in Parity shows how sharp traders (10k$+ PnL, 100k$+ volume) are positioned across Yes/No on that specific market
• stacked on your side → independent confirmation your edge is picking up something real
• stacked against you → worth pausing before you size up
and if you're trading the behavioral edge specifically - fading a retail-driven extreme - you want to know who you're actually trading against
Counter Party shows exactly who's on the other side of the trade, so you're not guessing whether it's real retail herding or a sharp wallet you're fading by mistake
Check who's actually positioned before you size in 👇
