The boring strategy that printed, decoded.

The post that made the wallet famous said it buys No on the extreme temperature buckets at 67–79c, never scales, never revenge-trades, and collects when the high lands in the middle. Two thousand nine hundred trades, eighty thousand dollars. The gross figure is right: the tracker's $86.7K reproduces from the fills. Almost everything else about the description is wrong, and the correct version is more interesting.

What the fills say it does

Polymarket's data API returns every fill of a wallet, taker and maker alike, with the side, the outcome token, the price, the size and the time. Gamma returns each market's resolution. Match buys to later sales of the same token first-in first-out, settle whatever is still held at resolution, and every buy carries its own profit or loss. Then the pattern is not subtle.

• It trades on the market's own day, at midday. 98.3% of its No dollars are bought on the calendar day the market is about, in the city's local time, and 77% of them between 10:00 and 16:00, the hours in which the day's high is being set.
• One bracket at a time, in the middle of the ladder. 3,269 of its 3,346 buying bursts touch a single bracket. Tails, the "or higher" and "or lower" buckets, are 5% of its shares.
• Below the day's high. Where the winning bracket can be identified, 282 of its No buys sit below the winner, 13 sit on it, 3 above. The 13 are its losses.
• Sold within minutes. The median time from a buy to the first sale of the same token is zero minutes; the 90th percentile is two. The sale price is 99c. 78% of its fills are taker fills; the 99.9c exits are resting orders.

When the station's running high passes a bracket, that bracket's No is worth a dollar. The bids still resting on its Yes side are stale. The wallet sweeps them.

That is the whole strategy. It watches the observations, not the forecast. The moment the day's running temperature makes a bracket impossible, it lifts every stale bid it can find on that bracket's Yes side, walking up the No book in one-minute bursts, then sells the position at 99c to whoever has re-quoted, and recycles the cash. One Taipei bracket on June 2 shows the shape in full: eleven fills between 04:30 and 04:31 UTC at No prices from 78c to 92c, then the whole lot sold at 99.9c ten minutes later. Its cash balance is $12.8K. It buys about $7,900 of contracts a day. The money comes from turning that balance over daily, not from holding anything.

Where the money is

The median entry is a No at 97c, which is a 3c Yes, and that is why the "collecting pennies" description stuck. But the pennies are not where the profit is. Map every buy onto the Yes side of the book, a No at 93c being a 7c Yes, and the profit sits in the 10–65c band: brackets the market still priced as one-in-ten to one-in-two, which the observation had already killed.

Two thirds of the profit, $58,000 of the $87,000, comes from the 602 buys in the 10–65c band: a sixth of the buys and a quarter of the dollars. The 1–5c band, half the dollars, makes $10,700. A trader reading the tracker's median entry would copy the wrong part of the strategy.

Where it loses

Thirteen of its 3,607 No buys settled Yes, costing $4,732. They are not spread around. They sit in Mexico City, Lucknow and Warsaw, cities whose stations report whole degrees Celsius. A reading of 36°C is 96.8°F; treat it as exact, round it to 97, and you rule out a bracket that can still settle at 96. We found the identical flaw in our own desk's observation feed this month, and it was the whole of that desk's loss in the same regime. The bot's edge is the observation; its risk is the observation's precision.

The same sweep, on Kalshi's tape

Kalshi lists the same kind of ladder for twenty US cities. We took every bracket that a station's confirmed running extreme had ruled out by a full degree over 47 days, 1,938 of them, and read Kalshi's public trade tape from that moment to the close. The market prices a dead bracket at half a cent by the time a half-hourly reader sees it. The wallet's trade, Yes sold into a resting bid at 3c or better on a dead bracket, printed 80,655 contracts in 47 days, on six days, and two of those days carry nearly all of it:

The median day is nothing. What runs thirty times a day on Polymarket, in cities without a weather-service forecast and with nobody else watching the observation, happens on Kalshi about twice a month, at size, hours after the fact, to whoever is watching every ladder. The cent that remains on a dead Kalshi bracket belongs to somebody else entirely, and that is the next piece.

Its cities