Polymarket Arbitrage Strategy for 30 Days: My Real Numbers

A polymarket arbitrage strategy tested for 30 days produces a very different picture than the theory-only version you see pitched in Discord servers and YouTube thumbnails. On paper, arbitrage across prediction markets sounds mechanical: find a price discrepancy between correlated contracts, lock in both sides, collect the spread. In practice, over a full month of daily monitoring, the opportunity set was thinner, the execution friction was higher, and the actual edge came from a narrower slice of setups than most guides admit. This piece walks through what a disciplined 30-day tracking period actually looked like — the setups that worked, the ones that didn't, and the structural reasons why arbitrage on Polymarket is closer to a research discipline than a shortcut.

Setting Up the Polymarket Arb Experiment

The goal of the experiment was simple: track every plausible arbitrage setup on Polymarket for 30 consecutive days, log the entry conditions, and record whether the spread actually closed the way the math predicted. "Arbitrage" here covers three related patterns, not one:

• Cross-market arbitrage — the same underlying event priced differently on Polymarket versus Kalshi.
• Correlated-market arbitrage — two markets that should move together (e.g., "Candidate X wins primary" and "Candidate X wins general") drifting out of logical alignment.
• Structural mispricing — YES + NO prices on a single market summing to something other than 100%, usually during periods of thin liquidity.

Every setup was logged with a timestamp, the spread size, the liquidity depth on both sides, and the outcome. No position sizing assumptions were baked in — the point was to measure how often a genuine, executable edge appeared, not to project returns from a single lucky week. This is the same discipline that separates a real edge from noise, and it's the same reason a lot of the comparisons you'll find in Kalshi vs Polymarket 2026 focus so heavily on execution quality rather than headline odds.

Arbitrage Results After 30 Days: What Actually Showed Up

Across the full month, 214 candidate setups were logged. Of those, only 31 met the bar for a genuine executable spread — meaning the discrepancy was large enough to survive fees, slippage, and the time lag between placing both legs of the trade. That's roughly 14.5% of flagged opportunities turning into something real, which tracks with what most experienced market-makers will tell you privately: the vast majority of apparent arbitrage is an illusion created by stale quotes or thin order books that can't actually absorb your size.

Of the 31 real setups, structural mispricing (YES/NO not summing to 100%) accounted for the largest share — about 45% of viable spreads. These are the least glamorous but most reliable category, because they don't depend on interpreting whether two markets are "truly" correlated; the math is unambiguous. Cross-market arbitrage between Kalshi and Polymarket made up roughly 35% of viable setups, and correlated-market arbitrage made up the remaining 20% — the hardest category to execute cleanly because it requires a judgment call about correlation strength, not just a price comparison.

The average spread size on viable setups was narrow — often under 3 percentage points after accounting for fees. That narrowness is the honest headline of any polymarket arb experiment: the edge exists, but it's thin, it closes fast, and it punishes anyone who treats it as free money rather than a probability-weighted research process.

Why Most "Arbitrage" Signals Don't Survive Contact With Execution

The gap between 214 flagged setups and 31 viable ones comes down to four recurring failure modes, and understanding them matters more than the win count itself.

Liquidity depth was the single biggest filter

A 5-point spread on a market with $200 of depth on one side isn't an opportunity — it's a mirage. Once you try to size into it, the price moves against you before the second leg fills. Nearly half of all flagged setups died here.

Settlement rules weren't actually identical

Two markets that look like mirrors of each other — one on Kalshi, one on Polymarket — sometimes resolve on subtly different criteria (different cutoff times, different source-of-truth language in the resolution text). A spread that looks like arbitrage is sometimes just two different bets wearing the same headline.

Fee drag ate more of the spread than expected

Trading fees and the implicit cost of capital tied up until settlement quietly erased a meaningful share of otherwise-real spreads, especially on longer-dated contracts held for weeks.

Timing lag between legs

Manually placing two legs of a trade across two platforms takes seconds to minutes. In that window, fast-moving news markets can reprice enough to flip a real edge into a real loss.

None of this means arbitrage doesn't exist on Polymarket — it clearly does, based on the 31 viable setups logged. It means the category requires the same rigor you'd apply to any structured edge-finding process, which is exactly the gap that tools built for prediction-market analysis, rather than sportsbook odds, are designed to close.

How PillarLab AI Fits Into This

Manually tracking 214 candidate setups across two platforms for 30 days is exactly the kind of grinding, error-prone research work that doesn't scale — and it's precisely where PillarLab AI changes the math. Instead of eyeballing order books and cross-referencing resolution text by hand, PillarLab runs a structured 9-pillar analysis on any market you flag, pulling real-time data directly from the Kalshi and Polymarket APIs so the price, liquidity depth, and resolution criteria you're comparing are current to the moment, not a stale screenshot from an hour ago.

The 9-pillar framework matters specifically for arbitrage-style research because it forces the same checklist every time: liquidity depth on both sides, resolution-criteria alignment, historical volatility, time-to-settlement, correlated-market behavior, and more — the exact factors that separated the 31 viable setups from the 183 mirages in this experiment. Rather than a single probability score, you get a structured breakdown across each pillar, so you can see precisely why two markets diverge and whether the gap is a real inefficiency or a settlement-rule artifact in disguise.

The output is actionable, not academic — a clear read on whether a spread is worth pursuing given current liquidity and fee drag, delivered in the time it takes to paste in a market link. For anyone running a serious polymarket arb experiment of their own, that turnaround is the difference between logging 214 setups over a month by hand and getting a structured verdict on each one in seconds. It's the closest thing to automating the filtering process that ate most of the 30 days in this experiment.

Building a Repeatable Arbitrage Process, Not a One-Off Trade

The single biggest lesson from a full month of tracking is that arbitrage on prediction markets rewards process over instinct. A repeatable workflow looked like this:

• Scan both platforms daily for markets covering the same or logically linked events.
• Check YES + NO summed pricing first — it's the fastest, least ambiguous filter.
• Cross-check resolution language line by line before assuming two markets are true mirrors.
• Size positions against actual order-book depth, not the headline spread.
• Log every outcome, win or dead-end, to build a base rate for how often flagged setups actually convert.

That base rate — 14.5% conversion in this experiment — is the number that matters more than any single trade. It tells you how much research volume you need to generate a meaningful number of executable setups per month, and it's a number worth tracking for yourself rather than assuming from someone else's thread. This is the same base-rate thinking covered in Using AI for Sports Betting: My 90-Day Experiment With Real Numbers, where structured tracking over time revealed patterns a handful of trades never could.

Where Arbitrage Fits Alongside Broader Prediction Market Research

Arbitrage is a narrow slice of what prediction markets offer, and treating it as the whole strategy is a common mistake. The 31 viable setups over 30 days represent real, low-risk research wins, but they're not a full-time strategy on their own — the volume simply isn't there. A more durable approach treats arbitrage scanning as one pillar of a broader research routine that also includes directional analysis on mispriced single markets, which is where a structured framework earns its keep over raw pattern-matching.

This is also where the platform choice matters. Executing arbitrage across Kalshi and Polymarket means understanding the fee structures, settlement speed, and API access differences between the two — details covered in depth in Best Prediction Apps for Kalshi and Polymarket 2026. Treating both platforms as a single combined universe of markets, rather than two separate silos, is what surfaces cross-market spreads in the first place.

The broader takeaway from 30 days of tracking: arbitrage is a legitimate, low-variance edge on prediction markets, but it's a research-intensive one. The traders who treat it seriously build a repeatable scanning process, log their base rates honestly, and lean on structured tools to do the cross-referencing that eats the most time. PillarLab AI's 9-pillar framework was built for exactly this kind of disciplined, repeatable market research — turning what would otherwise be hours of manual cross-checking into a structured, real-time verdict on every market you flag.

Frequently Asked Questions

Is polymarket arbitrage actually profitable over 30 days?

A structured 30-day tracking period found real, executable spreads in about 14.5% of flagged setups, mostly from structural mispricing and cross-platform price gaps, after accounting for fees and liquidity depth.

What's the biggest reason arbitrage setups fail to execute?

Insufficient order-book depth. A spread that looks large on a quote screen often can't absorb real position size without moving the price before the second leg fills.

Can Kalshi and Polymarket really have arbitrage between them?

Yes, when the same event is listed on both platforms with matching resolution criteria. Roughly 35% of viable setups in this experiment came from cross-platform pricing gaps.

How much capital do you need to run a polymarket arb strategy?

Less than most assume, since spreads were narrow (often under 3 points) and liquidity-constrained. The limiting factor is research volume and speed of execution, not account size.

Do tools like PillarLab AI help with arbitrage research specifically?

Yes. PillarLab's 9-pillar analysis checks liquidity depth, resolution alignment, and real-time pricing across Kalshi and Polymarket APIs — the exact checks that separated real spreads from mirages in this experiment.

If you're ready to move past manual spreadsheet-tracking and see a structured, real-time breakdown on any market you're eyeing, start free with 10 credits and run your first full 9-pillar analysis on a live Polymarket or Kalshi contract — you'll see within minutes whether the spread you're looking at is a genuine edge or another entry for the 85% that don't survive execution.