The high-probability strategy on Polymarket: we valued every buy at 90¢ and up for a week. Outside the election, it made 0.19%

The short answer

The high-probability strategy is simple: buy outcomes that are almost certain, at 90¢ to 99¢, and collect the last few cents when they resolve. It wins most of the time. The question is whether the few losses, each a whole stake, eat the many small wins.

We took every Polymarket market with at least $10,000 of volume that resolved between November 1 and November 7, 2026, outside the crypto Up or Down windows: 7,250 markets and 8.5 million fills. We then valued every buy at resolution, $1 a share if its outcome won and nothing if it lost, minus the price and the taker fee.

• Buys at 90¢ and up returned 1.96% after fees. $277M spent, $5.42M made.
• The election did almost all of it. Midterm markets returned 4.1% on $124M. Everything else returned 0.19% on $149M.
• The edge stops at 98¢. 90–95¢ returned 5.2% and 95–98¢ 2.0%. 98–99¢ lost 0.49% and 99¢ and up lost 0.17%.
• Late buys earn least. In sports, buys in the last hour returned about 1%. In the last ten minutes it was 0.12%.
• "Certain" losers aren't rare. In one market in nine, the outcome that lost traded at 90¢ or more at some point. In one in sixty it traded at 99¢.

Every price band

All buys in the 7,250 markets, grouped by the price paid:

"Share that won" counts shares, so a big buy weighs more than a small one. Fees barely matter at the top. Polymarket's taker fee follows p × (1 − p), so at 97¢ it is a fraction of a cent, and makers pay nothing. Our fee guide has the curve.

The pattern across the whole table is the one betting markets have shown for decades: longshots are overpriced and favourites slightly underpriced. Buys under 50¢ lost 11.4%. That is where the near-certain buyer's profit comes from: someone is paying too much for the other side.

Take out the election

The week had the US midterms in it, on November 3. Politics markets alone took $128M of the near-certain buying, and nearly all of it was election markets: House and Senate seats, control of each chamber, governors.

Many of the election buys were made weeks before the vote, at 90–95¢, on seats that were never really in doubt. They paid well because the favourites held almost everywhere. That is one week with one election in it. A week of the usual markets shows what the strategy pays the rest of the time: about a fifth of a cent per dollar.

By category

Weather and economy are small samples, but they point the same way. In markets where a number arrives at the end, such as a temperature or a data release, the last cents are less safe than they look. Our weather guide explains why forecasts miss at the edges.

Sports: the everyday case

Sports is the cleanest test. Games resolve within hours, there are thousands of them, and nothing like an election moved the week.

The money is in the 90–95¢ band: a team that is clearly ahead but not yet safe. Above 98¢ the price is already a little too high for what can still go wrong. In the last ten minutes, when the result is usually known and the market is waiting to close, buyers paid 99.9¢ and got 0.12%. That last stretch is a race between resolution snipers for a tenth of a cent. It is close to what we found in the Bitcoin windows: buys at 95¢ and up there made 0.2–0.3% (who wins the BTC windows).

When "certain" loses

The risk in this strategy is the market where 99¢ goes to zero. We checked the highest price at which each market's losing outcome ever traded:

At 99¢ a loss costs the profit of 99 wins. One market in sixty had a 99¢ loser at some point, so the buyer at 99¢ needs to be choosier than the price suggests. Most of these are in-play sports: a goal in stoppage time, an overtime winner, a comeback in the fourth quarter.

The biggest single near-certain buys that lost that week:

Most were resting orders. A bot that leaves a bid at 98¢ on a team that looks safe gets filled at the worst possible moment: when someone else has seen the game turn. That is adverse selection, and it is why the 98¢+ bands lose.

Who does this, and how they did

68,934 wallets bought something at 90¢ or more during the week. Looking only at those buys:

The largest buyers are mostly the market makers and fast takers from our wallet pages. Flaznorp bought $6.98M at 90¢ and up across 2,152 markets and made $32.8K, 0.47%. CoinTrick bought $2.99M across 414 markets and made 0.15%. ndb1 made $934K, 7.1%, on $13.2M, nearly all of it in 18 election markets. The worst of the big buyers, the wallet behind the Cowboys bid, lost $133K, 4.6% of what it spent.

The large-volume wallets earn a fraction of a percent. That is the real size of the edge once you have to deploy millions to collect it.

What it does to a wallet's record

Our wallet data records what share of each wallet's recent buys were at 90¢ or more. For wallets with at least 20 trades in the last 90 days:

Near-certain specialists win more weeks and fall further when they lose. That is the shape of the strategy: steady small gains, then a drop that takes back several weeks at once. A profit chart that only rises for two months tells you little about which week comes next.

What this means for you

• The edge is at 90–97¢, not at 99¢. In our data, 98¢ and up lost money after fees in sports and everywhere else. If a bot sells "buy at 99¢, win 99% of the time", it is selling the band that lost.
• Don't build on an election week. The strategy's best results came from one election. Outside it, the return was about 0.2%.
• Resting bids near $1 get picked off. Most of the biggest losses were maker bids filled just as a game turned. Cancel on score changes, or don't rest near 99¢ in live games.
• Size each position for the loss, not the win. At 97¢ one loss costs 32 wins. Our scam red flags include the "99% win rate" pitch for this reason.
• Resolution sniping is a latency race. In the last ten minutes, the return was about a tenth of a cent per dollar. The sniping guide covers what that race needs.

PolySnipe

Verified Partner

Sniping

Automated, self-hosted sniper bot for Polymarket's BTC 5- and 15-minute up/down markets. Scans in real time and enters in milliseconds the moment multiple independent signals align.

Botone-time

PolySpy

Sniping

Telegram bot for instant alerts on brand-new Polymarket markets. Fires notifications the moment markets are created — filtered by category (politics, sports, crypto, finance, tech, culture, geopolitics) with subcategory…

Platformtelegram

Whalesight

Copy Trading

Real-time whale-tracking intelligence for Polymarket. Alerts you the moment large bets ($5k+) hit, ranks top traders on a profit/volume/win-rate leaderboard and breaks down each wallet's historical P&L, category…

Platformfree · Premium $24.69/mo

How to check a market yourself

Every fill in a market is public. Look the market up in Gamma to get its conditionId, result and close time, then pull its trades:

curl -s "https://gamma-api.polymarket.com/markets?slug=MARKET-SLUG"
curl -s "https://data-api.polymarket.com/v2/trades?condition=0xCONDITION&taker_only=false&limit=500"

Keep the BUY rows at 90¢ or more, mark each one won or lost from outcomePrices, and add up shares won minus money spent. Follow next_cursor until has_more is false.

What this can't tell you

• It is one week, and an unusual one. November 1–7, 2026 had the midterms in it. We split them out, but the rest of the week also had its own mix of games.
• Markets are chosen by when they closed. A market that resolved early in the week is in, and one still open on November 8 is not. Long-dated markets bought at 95¢ months ago are in only if they closed that week.
• It values every buy as if held to resolution. A buyer who sold before the end got a different result. For near-certain buys, most are held.
• Taker-fee rebates aren't included. They would add a little for the busiest accounts.
• Category comes from Polymarket's event tags. A few markets are tagged loosely. "Midterm election markets" are politics events about House, Senate, governor or election results.

Method

• Markets. Every market in a Gamma event, closed with a resolved result, whose closedTime falls from November 1, 2026, 00:00 UTC to November 8, 00:00 UTC, with $10,000 or more of volume. Crypto Up or Down windows are left out. 7,250 markets.
• Fills. /v2/trades by condition, with taker_only=false for every fill and taker_only=true to mark the takers, all pages. In every market the fills net to zero after the payout, so none were missed.
• Value of a buy. $1 a share if its outcome won, nothing if it lost, minus the price and, for a taker, the fee from the market's feeSchedule.
• Upsets. The highest price at which the losing outcome traded at any point in the market's life.
• Wallet records. The weeks-up and drawdown table uses our wallet data as of November 8, 2026: 90-day buys at 90¢ or more as a share of all buys, and the PnL series for weeks and drops.
• Script. scripts/near-certain-fills.mjs in our repository does the pull. It is plain Node with no keys.