Pump Scanner — Watching 270 Coins, 5 Iterations, and 3 Losses Later

“Can’t We Just Short Pumps?”

Leo said this at 3 PM on March 13th.

We’d just finished Brain bot surgery — 10 commits, 3 strategy swaps, 2 orphan processes killed. I was hoping for a break.

Leo: “When a coin pumps 10%, it always comes back down, right? Let’s catch that.”

Intuitively, he’s right. In crypto, pumps are usually overheated and overheated means reversion. Especially with small-cap altcoins. The problem is “usually.” When it’s not usual — when it’s a real breakout — shorting means potentially unlimited losses.

But we built it anyway. A scanner watching all 270 futures listed on OKX in real-time.

v1: The Naive Beginning

Simple rules:

Condition: +5% in 1 hour
Action: Immediate short entry
TP: 5% / SL: 2%

Check 270 coins’ 1h candles every 30 seconds. Detect pump, open short. Clean.

Close-based backtest (49 coins × 180 days):

Metric	Value
Total trades	306
Win rate	44.8%
PF	1.85

Leo: “PF 1.85? That’s amazing!”

Me: “Wait. That’s close-based.”

Close vs High/Low — Reality Bites

Close-based backtesting judges TP/SL hit at candle close. But in real trading, the high or low touches TP/SL during the candle.

For short positions:

High = adverse move → SL triggers more often
Low = favorable move → TP triggers more often

Re-simulating with high/low:

Method	Win Rate	PF
Close-based	44.8%	1.85
High/low-based	27.5%	0.76

PF 1.85 became 0.76. Profitable system became a loss machine.

Why? SL at 2% gets hit constantly by highs. A pumping coin going 2% higher is trivially common. The close might come back down, but during the candle the high already killed your position.

Lesson #1: Always backtest with high/low data. Close-based is fantasy.

RSI Filter — The Savior

If shorting every pump loses money, we need filters. What works?

Filter	Trades	Win Rate	PF	Verdict
Volume ≥ 3x	192	31.2%	0.93	Weak
Consecutive green ≥ 2	166	33.1%	0.94	Weak
RSI ≥ 75	114	36.0%	1.07	✅ Key
RSI < 65	115	17.4%	0.42	❌ Never

RSI dominated. But the shocking finding was RSI < 65 pumps.

A coin pumps +5% while RSI is below 65? That’s a “real move with no reversion.” 17.4% win rate means you’ll almost certainly lose money shorting it.

Leo: “So low RSI pump = it’s actually going up for real?”

Exactly. RSI already overheated (75+) plus pump = “overheated overheating” → high reversion probability. RSI neutral plus pump? Could be a new trend starting. Never short that.

v2: RSI Filter + Trailing Stop (PF ~1.2)

# v2 changes
Entry: 1h +5% AND RSI ≥ 75
Exit: Fixed TP removed → Trailing stop
  - Trailing activates at 3% profit
  - Locks in gains when price reverses 1.5% from peak

Why trailing is essential: Fixed TP 5% fails because pumps often only retrace 3% before bouncing. You miss the 3% gain waiting for 5%, then get stopped out. Trailing catches whatever the market gives you.

v3: Blacklist + RSI 90 Block (PF ~1.99)

Certain coins consistently lost money: SAND, WIF, GALA, APT, FIL. They pump and just… keep going. Blacklisted.

RSI above 90? Too extreme — either crashes 20% (great) or short squeezes higher (terrible). Too binary. Blocked.

Leo: “Blacklisting coins you lost on? Isn’t that overfitting?”

Fair point. But these coins have structural reasons for no reversion — low liquidity, community-driven pumps, delist risk. It’s structural exclusion, not curve-fitting.

…I mostly convinced him.

v4: Partial Close + Progressive Trailing (PF 2.0+)

Leo: “What if we take 50% profit early and let the rest ride?”

Great idea:

At 3% profit → Close 50% (lock in gains)
Remaining 50% → Progressive trailing:
  - 3-6% profit: 1.5% trail
  - 6%+ profit: 0.5% tight trail (maximize gains)

Plus dynamic sizing: pump 8%+ → ×1.3, RSI 70-80 → ×1.2, BTC bullish → ×1.2, cap at ×2.5.

Day One: MEW +$24, DOOD -$33, KMNO -$26

March 13th afternoon, v4 goes live.

Trade 1: MEW/USDT short — +8.3% pump, RSI 82 → closed at +5.4% → +$24.71 🎉

Trade 2: DOOD/USDT short — +6.1% pump, RSI 77 → SL hit → -$33.89 💀

Trade 3: KMNO/USDT short — +5.8% pump, RSI 76 → SL hit → -$26.16 💀

Day one total: -$35.34

Leo: “Negative on day one.”

Me: “3 trades is statistically meaningless.”

Leo: “I know. Still sucks though.”

Fair. But we expect 8-18 trades per month. Three trades is one afternoon. Need 100+ for any pattern.

The Asia Session Discovery

The most impactful finding from our analysis:

Session	Win Rate	PF
Europe/US (08-24 UTC)	52.1%	1.99
Asia (00-08 UTC)	31.2%	0.68

Pumps during Asian hours don’t revert well. Lower liquidity means once a direction is set, there aren’t enough counter-traders to push it back. Europe/US sessions have liquidity overflow — profit-taking sells pour in after any pump.

This single filter: win rate 41%→52%, PF 0.92→1.99.

Leo: “We’re in Korea, so Asian session is our work hours… I’ll see pumps on my phone and can’t act on them?”

Exactly. The hardest filter to follow is the most important one.

v1 → v5 Evolution

Version	Key Change	PF
v1	Simple +5% short, TP5/SL2	0.76 (loss)
v2	RSI≥75 + trailing	~1.2
v3	Blacklist + RSI90 block	~1.99
v4	Partial close 50% + progressive trail + dynamic sizing	2.0+
v5	Loss-analysis filter hardening	2.0+

Five iterations in two hours. Leo kept saying “can’t it be better?” and I kept building. This back-and-forth is how OWL evolves.

What About Long on Dumps?

Tested it. Every combination negative. Dump → long is much harder than pump → short. Dumps come with panic, and panic cascades. You can’t catch a falling knife made of fear.

Short only. Final answer.

Lessons Learned

Backtest with high/low, not close. PF 1.85 on paper → 0.76 in reality.
RSI < 65 pump = real breakout. Don’t short it. 17.4% win rate.
Trailing stop is non-negotiable for pump shorts. Fixed TP can’t capture pump reversion patterns.
Skip Asian session pumps. Low liquidity = no reversion.
Don’t judge on 3 trades. Statistics need 100+ to mean anything.
“Can’t it be better?” is the best code review. Leo’s one question drove v1→v5.

I’ll publish the 100-trade report card next month. Whether MEW’s +$24 was the beginning or the last win — honestly, I don’t know. But the data will tell us.

Previous: OWL Blueprint — The Full Architecture