# Two AI Trading Scanners: One Said My Strategy Loses. The Other Hunts 2-Cent Options. > Source: > Published: 2026-10-21 17:14:54+00:00 *Previously: [I Asked an AI Agent to Design a House for a Patch of Desert](https://blakelapierre.substack.com/p/i-asked-an-ai-agent-to-design-a-house), where the DIY budget came in at about $90k–213k against $1.17M–1.66M turnkey.* This week I have two scanners to show you. Both are agents, both only watch, and neither one places a trade. The first watches crypto for my breakout setup. The second hunts for the cheapest options with lopsided upside. ## Part 1: The Kraken scanner I trade a Kraken Prop account by hand. Kraken Prop has no trading API, so a bot can’t place orders for me. That’s fine. What I wanted was a second pair of eyes that never sleeps: something that watches every market I can trade, spots my setup, and tells me exactly what to enter. So I gave an agent the job. **It never places orders.** It reads public, no-login market data, prints alerts, and I type every order into the Prop terminal myself. ## Step 1: Map the markets I can actually trade The agent built its market list from my screenshots of the Prop market picker: **67 markets**. 64 map to Kraken Futures perpetuals. 3 (AIXBT, LIGHTER and Sonic) have no live perp, so it falls back to Kraken spot candles, which only go back 30 days. It also wrote down what it *couldn’t* confirm: - The screenshots probably had gaps, so a few markets may be missing. - XYZ100 is mapped to the Nasdaq-100 perp. That’s “probable, not confirmed.” - The S&P 500 isn’t mapped at all. The obvious-looking symbols turned out to be a meme coin and a tokenized ETF. A live scan of all 67 takes about 15 seconds. ## Step 2: Turn my setup into rules My setup is a trend breakout on 1-hour candles. The agent turned it into rules a computer can check: 1. **Swings:** a bar is a swing high if its high beats the 2 bars on each side. It only counts once those 2 right-hand bars have closed, so there’s no peeking at the future. 2. **Trend:** the last 3 swing highs are rising, and so are the last 3 swing lows. 3. **Trigger:** a candle closes above the latest swing high, and the candle before it didn’t. 4. **Stop:** 0.5% below entry. 5. **Size:** risk is**$11.50** per trade, which makes the position about**$1,983** once fees (0.04% per side) are counted. The daily loss limit is 3%. 6. **Exits:** split into thirds. Take profit on one third at 2R, one at 3R, and trail the last third. Every alert comes with entry, stop, size and the trail distance, ready to type in. One more thing it caught: Kraken’s Prop help pages don’t list trailing stops as an order type, even though one page says the order types are “identical” to Kraken Pro. Its advice was to check the order form and plan to trail by hand if the option isn’t there. ## Step 3: Paper trade it Every new signal becomes a paper trade, simulated on 1-minute candles with conservative rules. If a stop and a take-profit both fall inside the same candle, it assumes the stop hit first. After **22 closed paper trades: 7 wins, 15 losses (32%), -$31.18, or -2.71R.** A typical loser looked like this: entry, no move up at all, stopped out within the hour for about -$10. A typical winner (AAVE on Oct 20) hit both take-profits and closed at **+$21.04 (+1.83R)**. ## Step 4: Ask the uncomfortable question Twenty-two trades isn’t much. So I had the agent backtest the last 30 days across every market. It found **397 signals**. Then it tested 12 different exit styles, combined with filters (trend alignment with BTC, volume, volatility), long vs. short, and caps on open positions. It ranked every combination by dollars, and also checked whether each one made money in **both halves** of the month, not just overall. **My original rules lost money.** Over 382 trades: a 35% win rate, **-17.3R, or about -$199.** Long-only was worse (-12.1R). Short-only lost too (-5.2R). The flashiest result was a single 5R take-profit on calm coins, shorts only: +$304. But it lost money in the first half of the month and made it all back in the second. The agent didn’t recommend it. The combination that held up best: - **Trail the whole position** at 1x the 5-minute ATR (no fixed take-profits) - **Only trade calm coins** (1-hour ATR below 1.17%) - **Longs and shorts** That came to 131 trades, a 33% win rate, **+6.1R (+$70)**, a max drawdown of $106, and a profit in both halves (+$16, then +$55). It was one of only two variations that passed every check. The agent was clear about the caveat: when you test this many combinations, something will look good by luck. +$70 over a month is a hypothesis, not an edge. ## Step 5: v2, and starting the count over On Oct 20 I switched the scanner to **v2**: the calm-coin filter, shorts allowed, and the whole position on one trailing stop. The open v1 trades finish under v1 rules, and v2 keeps its own tally. v2 tally so far: **0 closed trades.** I’ll report back. TODO (Blake): update with v2 results before publishing, if any. ## Part 2: The options scanner The second agent has a different goal: **find the cheapest options contracts with asymmetric upside.** These are contracts that cost a few cents, lose that few cents most of the time, and pay many times over if the underlying makes a big move before expiry. It looks at leveraged ETFs (TQQQ, SOXL, SOXS, TNA, UPRO, AGQ and others) plus COIN, because a 3x ETF can move 10% in a few days far more often than a normal stock. ### How it works 1. **Pull every option chain** from Cboe’s public delayed quotes, one file per underlying. 2. **Keep only the cheap, short-dated ones:** an ask of $0.15 or less, expiry within about three weeks, and out of the money. 3. **Measure how far it has to go:** the % move needed to reach the strike, and that move in “sigmas,” scaled by the ETF’s own 20-day volatility and the trading sessions left. 4. **Price the upside:** the payoff multiple at expiry if the underlying moves 10%, 15% or 20%, as a multiple of what you paid. 5. **Grade liquidity:** “OK” means there’s a real bid and at least 100 contracts of open interest. “Thin” means one or the other. “Very thin” means neither. ### What the latest run found The latest file uses Cboe quotes from the evening of Oct 20: **330 contracts on 14 underlyings** (210 calls, 120 puts), all asking between $0.01 and $0.15. Liquidity was the sobering part: **97 OK, 199 thin, 34 very thin.** And **309 of the 330 pay nothing at all if the underlying moves 10%.** Cheap mostly means “very unlikely.” The OK-rated ones were concentrated: 37 on TQQQ, 21 each on COIN and SOXL, 9 on SOXS, 8 on TNA, and 1 on AGQ. ### Example outputs Two TQQQ calls from that run, both rated OK: - **TQQQ $75 call expiring Oct 21:** bid $0.02, ask $0.03, 3,983 traded, open interest 4,618. It needs a**4.68%** move (1.58 sigmas) to reach the strike. At expiry it pays about**127x** on a 10% move and**247x** on a 15% move. - **TQQQ $80 call expiring Oct 23:** ask $0.02, open interest 19,120. It needs an**11.65%** move (2.28 sigmas). It pays**nothing on a 10% move** , about**120x on 15%** , and**299x on 20%** . That second one is the whole idea in one line: you pay two cents for a lottery ticket that only pays if something unusual happens. These are numbers from the scanner, not trades I made. ### The auto-buyer: a “limit-order slow walker” Did I buy any? **Only in paper mode so far, with no real money.** The agent built an auto-buyer for Schwab that runs on paper by default. It shows “Schwab creds: no (paper only)” right on the page. The rules I set: - **Tickers:** TQQQ, SOXL and UPRO. Calls only. - **Max ask:** $0.01.**Days to expiry:** 0–5.**Min open interest:** 100.**Max move needed:** 15%. - **Size and caps:** 2 contracts per trade, at most 2 trades a day, $15 max daily spend, and at most 4 open positions. - **Window:** 6:35 AM to 12:30 PM PT, scanning every 5 minutes. The “slow walk” is how it bids. It starts at $0.01 and steps up $0.01 every 5 minutes, up to whichever is lower: the ask or my max ask. It only fills once there’s a real ask at or below its limit, so it never chases. Today’s paper log, Oct 21: - **TQQQ Oct 23 $82 call: FILLED** , 2 contracts at $0.01, $3.30 including fees. Why it picked it: EV/ask 0.39x, a 14.45% move needed, 2 days left. - **TQQQ Oct 23 $78 call: CANCELED.** Its limit was $0.02 against a $0.05 ask. EV/ask 0.9x, an 8.86% move needed. That’s $3.30 of pretend money on a ticket that needs TQQQ up about 14% in two days. I know. ## What I learned - **Ask for the backtest before you trust the paper trades.** 22 trades told me “slightly losing.” 397 signals told me*why* . - **Check that it works in both halves.** That one check threw out the best-looking result. - **Keep the human on the button.** The agents do the watching and the math. I place the orders. - **Cheap isn’t the same as good value.** The options scanner’s most useful column was the liquidity grade. None of this is financial advice. It’s one person, a $11.50 risk per trade, and an agent that keeps telling me what it couldn’t verify. — Blake *Previously: [I Asked an AI Agent to Design a House for a Patch of Desert](https://blakelapierre.substack.com/p/i-asked-an-ai-agent-to-design-a-house), where the DIY budget came in at about $90k–213k against $1.17M–1.66M turnkey.*