The $624K "two-sided" crypto-candle bot: real profit, mostly rebates, and a recipe that would not build it
An X article holds up one Polymarket bot as proof you can build your own with an AI backtesting tool. We pulled all 1.6 million of the bot's public transactions. It made about $402K in cash, and more than half of that was fee rebates paid by Polymarket.
The claimx-MoneyForWhiskas built a two-sided BTC/ETH "Up or Down" bot that made +$624,300 in four months — and you can build one like it with Grok Bot and Horizon. — X (@RetroValix)
The short version
The article, titled "How to build a trading bot on Polymarket", opens with a profile screenshot of the account x-MoneyForWhiskas showing +$624,300.88 of profit since June. It describes the bot as buying both outcomes of five-minute BTC and ETH "Up or Down" candles, so that a matched Up and Down pair settles for $1 while costing a little less, and then holding a leftover directional position guided by its own probability estimate. The second half shows how to develop your own bot: Grok to analyze trade exports, Horizon to turn a momentum idea into backtested rules, and a separate execution module for Polymarket. Every call to action links to Horizon's sign-up page and the close recommends its paid Builder plan. The article discloses no relationship with Horizon.
The account is real, and its trades are public, so we did not rely on the screenshot. We downloaded the wallet's full activity history from Polymarket's public data API: 1,609,614 records from June 10 to October 26, 2026, covering 1,537,753 buy fills (median $7.60), 71,567 redemptions and 294 rebate and reward payments. Adding up every dollar out and every dollar in, plus the roughly $700 still held, gives the bot's actual cash profit: about $401,700. Polymarket's own public leaderboard shows $164,660 all-time for the same wallet, and the profile page now shows $629.4K. Neither matches the cash.
Two things explain the gaps. First, the bot paid about $468,000 in taker fees: three-quarters of its buying crossed the spread, and crypto-candle markets charge takers on every such fill. Its trading profit before fees, about $634,500, is close to the profile's headline, so the headline appears to leave fees out. Second, Polymarket paid the wallet $235,240: $202,735 as "taker rebates", $30,054 as maker rebates and $2,450 as liquidity rewards. The leaderboard figure matches the trading profit after fees but before those payments ($166,440). Counted in full, 59% of the cash the bot made came from those rebates and rewards, not from the trades themselves.
The mechanism the article describes is accurate. 99.99% of the money went into BTC and ETH candles, and across about 35,700 markets where it bought both sides and held to the end, a matched pair cost it about 97¢ before fees, a gap of roughly 3¢ (the article says 2.07¢). But fees took most of that: after fees a pair cost 99.19¢, leaving 0.81¢ per pair, which over 18.7 million pairs is about $151,000. The directional leftover, the part the article says to build with your own signal, made about $31,000 on 36,575 positions and won 51.7% of the time, little better than a coin flip. The bot's margin on trading alone also fell as it grew, from 1.75% of volume in June to 0.33% in September, while taker rebates rose from 4% of the taker fees it paid in June to 53% in October.
Would building this work for you? The evidence says no, not as described. The money in this bot comes from scale, execution and a fee-rebate program the article never mentions; the signal the article teaches you to build is the smallest part of it. A newcomer would start at small volume, as this bot did in June, when rebates covered 4% of its fees. Our own live tests on these same candle markets found no route in either: buying both sides below $1 at once was available in 0.2% of 21,512 markets, a resting two-sided bid grid lost 31.6% over 15,366 candles, and over 118,000 candles short moves reversed slightly more often than they continued, which is the opposite of the momentum signal the Horizon walkthrough tests. Verdict: a real and impressive bot used as the hook for a product pitch whose recipe would not reproduce it.
Technical appendix
Exhibit A Three numbers for one wallet
Wallet 0x3048d65321be3497164cdfc2996f94f98a2e7537 (x-MoneyForWhiskas, pseudonym Flashy-Gold), checked October 26, 2026. The bot was still trading.
| Source | All-time P/L | What it appears to count |
|---|---|---|
| Profile screenshot in the article | $624,300.88 | — |
| Profile page, Oct 26 | $629.4K | close to trading profit before fees ($634.5K) |
| Polymarket public leaderboard (data-api v2), Oct 26 | $164,660 | close to trading profit after fees, before rebates ($166.4K) |
| Cash flow, all 1.6M public records | $401,680 | every dollar in and out, plus positions still held |
The "appears to count" column is our reconciliation, not Polymarket's documentation: the figures line up to within about 1%, but Polymarket does not publish how each figure is calculated.
Exhibit B Where the cash came from
| Flow | USD |
|---|---|
| Buys (including taker fees) | −20,831,584 |
| Redemptions of winning shares | +20,997,304 |
| Sells | +18 |
| Positions still held (marked) | +703 |
| Trading profit after fees | +166,440 |
| Taker rebates | +202,735 |
| Maker rebates | +30,054 |
| Liquidity rewards | +2,450 |
| Cash profit | +401,680 |
Taker fees are measured per fill as the amount paid above price × shares; fills with no such excess are counted as maker fills. 56% of buy fills and 75% of buy dollars paid a taker fee, about 3% of each taker dollar.
Exhibit C By month: the trading margin shrank as rebates grew
| Month | Bought | Trading profit after fees | Margin | Taker fees | Rebates + rewards | Taker rebate ÷ taker fees |
|---|---|---|---|---|---|---|
| Jun (from 10th) | $0.77M | $13,491 | 1.75% | $15,795 | $1,661 | 4% |
| Jul | $3.22M | $50,077 | 1.56% | $71,066 | $26,481 | 31% |
| Aug | $5.46M | $55,054 | 1.01% | $123,409 | $57,318 | 40% |
| Sep | $7.84M | $25,563 | 0.33% | $173,545 | $100,827 | 50% |
| Oct (to date) | $3.55M | $21,552 | 0.61% | $84,287 | $48,952 | 53% |
Months are UTC. In September the bot made about $25,600 from trading and about $100,800 from rebates and rewards. We do not know the rebate program's terms, so we cannot say whether the rising refund share came from volume tiers or from changes to the program; the article does not mention rebates at all.
Exhibit D The complete-set mechanism, measured
Across the 35,645 markets where the bot bought both Up and Down and held to resolution with no sells:
| Quantity | Value |
|---|---|
| Average cost of an Up+Down pair before fees (weighted by pairs) | ≈97¢ |
| Same, after taker fees | 99.19¢ |
| Profit per matched pair after fees | 0.81¢ |
| Matched pairs | 18.7 million |
| Profit from matched pairs | ≈$151,000 |
| Directional leftover positions | 36,575 |
| Leftover win rate | 51.7% |
| Profit from leftovers | ≈$31,000 |
Per-side average costs are computed per market from the bot's own fills; leftovers are valued at that side's average cost. This decomposition covers the cleanly held markets, so its parts do not sum exactly to the wallet-wide trading profit. The fee arithmetic is the point: on these markets a taker pays about 0.07 × p × (1 − p) per share, about 1.75¢ at a 50¢ price, so a pair bought entirely by crossing the spread costs about 3.5¢ in fees, more than the ~3¢ gap.
Exhibit E Our own tests on the same markets
We have tested every version of this idea we could reach, live against executable order books, on the same five-minute crypto candles:
- Buying both sides below $1 at the same moment: 0 hits in 4,195 scans; even buying each side at its cheapest price of the market's whole life cleared fees in 46 of 21,512 markets (0.2%).
- Resting cheap bids on both sides (the maker version): −31.6% over 15,366 settled candles. The fills that arrive are the ones where the price is running against you.
- Direction: across about 118,000 candles, a short streak reversed 51–54% of the time. Momentum, the signal the Horizon walkthrough backtests, is the measurably wrong side at this horizon, and acting on the reversal lost money too once real fills were counted.
See the Lab entries on crypto-candle arbitrage, the passive dust-bid grid and mean reversion at maker prices. None of this means the showcased bot is fake. It means its profit comes from things the article does not teach: scale, execution and a rebate program.
Exhibit F Conflict of interest
The article's two links both go to Horizon's sign-up page (join.horizon.trade/build), it closes by recommending Horizon's paid Builder plan, and it shows a Horizon screenshot as step two of the recipe. It describes the showcased bot as a "reconstruction" whose "exact indicators remain unknown", and states the author has studied over 1,000 bots and tested over 250 strategies without showing any of those results. The article itself notes that a BTC price backtest does not tell you how a Polymarket contract position performs. Those Polymarket costs are what decided this bot's result.