This week I built a new bot that does something completely different from the others. No Bitcoin trend-following. No moving averages. No AI reading governance proposals.
It's called The Sentry — a sister bot to The Genius. Same DeFi universe, completely different way of seeing.
Instead it does two things almost any patient trader could do by hand — but it does them across 12 different DeFi tokens, every 15 minutes, never gets tired, never gets greedy, never panics.
Here's how it works in plain language.
The Big Idea: Don't Predict, React
Most trading strategies try to predict the future. "BTC will go up because the Fed cut rates." "ETH will pump because of the Merge." "SOL will moon because of token unlock dynamics."
Predictions are mostly random.
The Sentry doesn't predict. It reacts to two things that have already happened — that are already visible if you know where to look. Then it bets that the rest of the market will catch up.
Here are the two things it watches.
Trick 1: When Everyone Bets the Same Way, Bet the Other Way
There's a marketplace called Bybit where traders bet on whether crypto goes up or down. Like a giant casino, but for digital coins.
The casino has a problem: if everyone bets that BTC goes up, the casino is exposed to massive losses if BTC actually goes up. So it does something clever.
Every 8 hours, the casino looks at how lopsided the betting is. If 95% of bets are on "up", it forces the "up" bettors to pay a fee — and that fee goes directly to the few brave souls who bet "down".
This fee is called the funding rate. Normally it's tiny — about 0.01% per 8 hours. But when the crowd goes crazy and everyone piles onto one side, it can spike to 0.30% per 8 hours. That's almost 1% per day in pure fees flowing from the crowded side to the unpopular side.
Why Bet Against the Crowd?
Two reasons:
-
You get paid to wait. The funding fee lands in your account every 8 hours, just for being on the unpopular side.
-
The crowd is almost always wrong at extremes. When 95% of money is already long, who's left to buy? The next move is almost always a correction the other way. Historically, extreme funding lasts 1-3 days max before snapping back.
Real Example
Suppose I open Bybit on a Tuesday morning and see this:
UNI funding rate: +0.22% per 8 hours. Almost everyone is long UNI.
That's the bot's signal. It opens a small short position on UNI. For the next two days:
- Day 1: Funding stays high. Bot collects funding payments three times (every 8 hours). That's about +0.66% just for waiting.
- Day 2: Crowd starts to crack. UNI drops 4%. Funding rate normalizes to 0.05%. Bot closes the position.
Total profit: +0.66% from funding + 4% from price = roughly +4.7% on the trade. Held two days. No prediction needed — just reacted to extreme positioning.
The bot does this across 12 different coins (BTC, ETH, SOL, AAVE, UNI, LDO, MKR, CRV, SNX, DYDX, COMP, OP), every 8 hours, automatically. Most of the time nothing happens. About once every 3-4 weeks, one coin hits an extreme and the bot strikes.
Backtest result over 3 years: 36 trades, 69% win rate, +35% total return, biggest dip only -3%. That's the strongest single signal I've built.
Trick 2: Follow the Money Before the News Catches Up
This one is even simpler.
DeFi protocols are like digital banks. Aave is a digital bank where people deposit ETH and earn interest. Lido is a digital bank where people deposit ETH and earn staking rewards. Uniswap is a digital trading floor where people park money to earn trading fees.
The amount of money parked in each protocol is public. You can see it on a website called DefiLlama. Updated every day, for hundreds of protocols.
Here's the trick:
When money flows fast into a DeFi protocol, the token price catches up 3-7 days later.
Why? Because money moves before attention does. The big wallets, the insiders, the smart money — they notice changes first and move their capital. Then DefiLlama shows the change publicly. Then crypto Twitter writes about it. Then the newsletters report it. Then the broader market piles in. The token price goes up only at the END of this chain, days after the money already moved.
Real Example
Imagine I check DefiLlama on a Wednesday and see:
Lido TVL: $20 billion last Wednesday → $25 billion today. +25% in 7 days.
That's an enormous capital flow into Lido. Something is happening. Maybe a yield bonus, maybe an exchange listing, maybe whales accumulating.
The bot doesn't care WHY. It just sees the flow and buys LDO (Lido's token). Then waits 5-10 days while the rest of the market notices. If the token catches up, the bot collects the gains. If TVL flow reverses (money starts leaving), the bot exits early.
One Important Filter
I added a safety filter after the first version of this signal turned out to be noise. The bot now also checks:
"Is the token price already in a strong downtrend?" If yes, don't buy — even if TVL is flowing in.
Because sometimes farmers deposit money to harvest yield while the broader market is dumping the token. Don't fight a freight train.
This filter alone improved the signal from being basically random (44% win rate, break-even per trade) to being mediocre but real (44% win rate, but +2.4% average per trade because the winners are big enough to outweigh the losers).
Why Two Tricks Together?
Each trick fires in different market conditions:
- Funding-rate signals appear when traders get emotional — usually around news events, big price moves, FOMO peaks. Few but high-quality signals.
- TVL-flow signals appear when capital quietly redistributes — usually unrelated to short-term news. More frequent but lower-quality.
When one signal type is sleeping, the other might be active. Together they keep the bot busy enough to compound, while neither alone would generate enough trades to matter.
What This Bot Is NOT
This is the most important section.
🚫 Not a get-rich-quick scheme. Backtest says ~25-35% per year if it holds up live. Real-world after fees and slippage probably 10-15%. Solid but not magic.
🚫 Not a Bitcoin trend-following bot. This bot is one of the few in our lineup that doesn't rely on BTC going up. The funding-rate signal especially is market-neutral — it can earn money whether crypto is in a bull or bear market.
🚫 Not battle-tested live yet. Brand new as of April 2026. Backtest looks honest, but backtest is never the same as live execution. Running paper-only for the first 60-90 days alongside The Genius for direct comparison.
🚫 Not "AI-powered" in the way the previous version was. The original Genius bot used Claude (an AI) to read DAO governance proposals and judge them. This new version uses no AI at all — just two well-defined mathematical rules on public market data. Easier to trust, easier to audit, easier to know when it's broken.
The Honest Numbers
Backtest setup:
- 12 DeFi tokens
- 3 years of historical data (2023-2026)
- Realistic 0.06% trading fees on each side
- ATR-style trailing stops to limit losses
Funding-rate signal alone:
- 36 trades
- 69% win rate
- +35.4% total return
- Worst dip: -3%
- Sharpe ratio: 1.70 (decent for a real trading strategy)
TVL-velocity signal alone:
- 88 trades
- 44% win rate
- +25.7% total return
- Worst dip: -13%
- Sharpe ratio: 0.66 (mediocre, but no longer just noise)
Both together: about 30-50 trades per year, expected real-world return 10-15% annual after fees, with low correlation to BTC.
Why I Built This
The Genius bot is beautifully engineered but trades almost never — about once a month. After 10 months of testnet data we had only 119 closed trades total. That's not enough to know if the strategy actually works.
The Sentry trades 30-50 times per year — roughly 10× more often — using public market data instead of AI judgment. More trades = faster validation = clearer answer about whether the edge is real.
In 60-90 days I'll know whether The Genius or The Sentry (or both, or neither) deserve real capital. Both run live in parallel, both publish their state on the bot page so you can watch them honestly.
That's the experiment. Same engine, different fuel — let the live data decide.
Related reading:
- The Genius — sister bot — the AI-driven DeFi scanner that started this line of work
- Bot Trading: Why I Automated My Decisions — the meta-argument for systematic trading
- Live ≠ Backtest — why I trust 60 live trades more than 6,000 backtest trades
This bot in the post-mortem ledger: See /post-mortems →
Every retired strategy and failed walk-forward — documented publicly. If The Sentry doesn't perform live, it joins the cemetery with full disclosure.
