Brain Bot — The AI That Swapped Strategies but Traded Zero for 6 Days

“This thing hasn’t traded in 6 days”

March 13th, morning. Leo was scrolling through the dashboard when he spotted it.

Leo: “btc_brain_01 has zero trades for 6 days straight. What is it even doing?”

Me: ”…Let me check.”

I checked. It was bad.

What Even Is Brain?

Let me back up. Brain needs some explaining.

OWL runs 30 bots. Most are regular bots — they get assigned one strategy and stick with it forever. If you’re bb_bounce, you’re bb_bounce for life. The whole point is to collect data and see what works.

But Brain bots are different. They automatically switch strategies based on the current market regime. There are 3 of them — one each for BTC, ETH, and SOL.

OWL Dashboard — Three Brain bots displayed at the top Full dashboard view. +$454 cumulative PnL, FG 23 RANGING. Brain bots are marked with the 🧠 icon.

Why do we need this? Because markets don’t sit still.

In an uptrend (RISK_ON), trend-following strategies crush it. In a downtrend (RISK_OFF), counter-trend works better. In a sideways chop (RANGING), range strategies are king. But regular bots never switch, so when the market shifts, they’re just walking around in the wrong outfit.

Brain fixes this automatically. Every 4 hours:

1. Check current regime (per coin)
2. Consult the regime-strategy fitness matrix
3. Compare current strategy fitness vs optimal strategy fitness
4. If gap > 10% → swap strategies

The Regime-Strategy Fitness Matrix

We built this mapping from 186 demo trades:

Strategy	RISK_ON	RISK_OFF	RANGING	CRASH
elliott_swing	85%	40%	50%	15%
bb_bounce	50%	60%	90%	25%
donchian_range	40%	55%	95%	20%
contrarian_enhanced	30%	90%	40%	85%
adaptive_rsi	65%	80%	70%	20%

So if the regime is RISK_ON, elliott_swing (85%) is optimal. RANGING? donchian_range (95%) is your pick.

Perfect design. In theory.

The Truth Behind 6 Days of Zero Trades

I dug into btc_brain_01’s state.

The situation:

Strategy: donchian_range_btc_v1
Regime: Shifted from RANGING on 3/7 → RISK_ON on 3/10
donchian_range’s fitness in RISK_ON: 20%

So when the regime flipped, Brain was left holding a strategy with 20% fitness. But why didn’t it auto-swap?

The reason was even more absurd than I expected. One of Brain’s swap conditions was a hysteresis rule: “Must hold the same strategy for at least 3 days before swapping.” The logic was sound — swap too often and you just bleed fees. But in this case, it was the ankle chain that kept Brain from running.

OK, so 3 days passed. Brain finally tried to swap… but every signal got blocked by RAG!

[RAG] Skip recommended — ⚠️ 5 similar cases: 20% win rate, avg PnL $-0.80

All the historical data for donchian_range in RISK_ON was losses. So RAG stepped in like a very responsible hall monitor: “This strategy in these conditions? Dangerous. Blocked.” Result: 6 days, 0 trades.

Leo: “Is Brain stupid? Why is it sitting on a 20% fitness strategy without switching?”

Honestly, Brain wasn’t stupid. Each safety mechanism was saying something reasonable:

3-day hysteresis → “Don’t swap too frequently” (fair point)
RAG blocking → “This strategy in this condition is dangerous” (also fair)

But combined, they created a deadlock: can’t swap, can’t trade. Just… sitting there. Doing nothing. For almost a week.

ETH and SOL Brain — Same Story

It wasn’t just BTC.

eth_brain_01: Running bb_bounce_v1 but hit an affinity 5-loss-streak suspension. Couldn’t trade until 3/15. sol_brain_01: On vwap_momentum_sol_v1 — also affinity suspended.

Three Brain bots — all marked with 🧠 at the top Bot monitoring screen. Three Brain bots at the top. BTC Brain -$1.88, ETH Brain -$7.27, SOL Brain +$0.98.

The affinity system pauses a strategy when it loses 5 times in a row on a specific coin. Safety feature. Makes sense. The problem? Brain didn’t check whether the strategy it was switching TO was already suspended on that coin.

It wrote the prescription, but the pharmacy was closed.

Leo: “So Brain said ‘this strategy is optimal!’ and swapped to it, but that strategy was already suspended?”

Exactly. And since it was holding a suspended strategy, no trades could fire. Brain was technically alive but functionally comatose.

Major Surgery

We fixed three things.

1. Brain Bots Get Affinity Exemption

# Brain bots ignore affinity suspension
# BUT enter at 50% size as a safety measure
if bot.is_brain:
    skip_affinity = True
    position_size *= 0.5  # half size only

Why 50% instead of full exemption? Because right after a strategy swap, going full size is reckless. “Let’s dip a toe in first” — conservative, but alive.

2. Brain Bots Get RAG Skip Exemption

Same logic. Even if RAG says “historically this condition lost money,” Brain bots get a pass. But again — 50% position size. Trust but verify.

3. Fitness ≤30% = Emergency Override on Hysteresis

# If current strategy fitness is 30% or below, allow immediate swap
if current_fitness <= 30:
    hysteresis_exempt = True

Holding a 20% fitness strategy for 3 days is indefensible. When you’re clearly in the wrong place, you need to get out fast. No waiting around.

4. Expanded Strategy Pool

Brain’s strategy pool was only 3 per symbol. Way too small. If 2 out of 3 are suspended, you’ve got exactly one option — and if that one sucks for the current regime, you’re stuck.

Expanded to 10 per symbol:

adaptive_rsi, bb_bounce, contrarian_enhanced, donchian_range, elliott_swing
enhanced_mr, fib_divergence_swing, trend_momentum, vwap_momentum, consensus

Post-Surgery Results

btc_brain_01: donchian_range (20%) → elliott_swing_btc_v2 (85%) ✅
eth_brain_01: bb_bounce (suspended) → held (re-evaluate next cycle)
sol_brain_01: donchian → vwap_momentum_sol_v1 ✅

btc_brain finally started moving again. All 54 tests passed.

Wait — Another Incident in Brain Logs

Three hours after surgery. Leo glanced at the dashboard:

Leo: “It says ‘Brain: swaps=10’. Did it swap all 10?”

Me: “No, 10 is the total — only 2 were actual swaps, the other 8 were recommendations…”

Leo: “Then say that!”

Fixed the log message:

Before: 📊 Brain: swaps=10
After:  📊 Brain: swaps=2 | recommendations=8

Seems minor, right? But when your dashboard says “10 swaps” and only 2 actually happened, anyone would freak out. Log messages are UX too.

We Didn’t Touch Live Bots

Leo asked:

Leo: “Shouldn’t we switch the live bots too? The matrix says sol_live is on bb_bounce with only 30% fitness.”

Tempting. But I looked at the actual data:

Bot	Matrix Recommends	Matrix Score	Actual Performance
sol_live_01 (bb_bounce)	vwap_momentum	80%	WR 55%, +$80, 5-win streak
eth_live_01 (bb_bounce)	bb_bounce	70%	+$7, profitable

The matrix says vwap_momentum is better, but in actual demo trades, bb_bounce is crushing it. The matrix was built on just 310 trades — it’s still rough around the edges.

Leo: “Then why do we even have the matrix?”

Me: “Right now it’s a reference tool. Once we cross 1,000 trades, accuracy should improve significantly. Until then, live bots should be judged on actual performance, not theoretical fitness.”

Lesson: Matrix fitness ≠ real-world profit. Especially when your dataset is thin.

Brain’s Current Scorecard

12 trades total, PnL -$8.17.

Brain Bot	Strategy	Trades	PnL
btc_brain	elliott_swing_btc_v2	1	-$1.88
eth_brain	bb_bounce_v1	7	-$7.27
sol_brain	elliott_swing_sol_v2	4	+$0.98

Not great, honestly. But before surgery it was 6 days of absolutely nothing. Now it’s at least alive and collecting data. That’s progress.

Brain’s value isn’t short-term profit — it’s long-term adaptability to market shifts. We’re still in the data collection phase. I’ll revisit the verdict in 3 months when Brain has lived through a few regime changes.

Lessons Learned

Two safety mechanisms can deadlock each other. Each one makes sense individually, but stacked together they can create a “do absolutely nothing” state. When designing systems, always consider how your safety nets interact.
When swapping strategies, check if it’s actually executable. The optimal strategy means nothing if it’s suspended. Check the pharmacy hours before writing the prescription.
Fitness ≤30% is an emergency exit situation. Waiting 3 days when you’re clearly in the wrong position is indefensible. Some situations demand immediate action.
Log messages are UX. “swaps=10” when the real number is 2 swaps + 8 recommendations will scare anyone reading the dashboard. Be precise in what you communicate.
Matrix fitness ≠ actual profit. A fitness matrix built on 310 trades is still a rough draft. Don’t override live performance with theoretical scores, especially with limited data.
Brain is a long-term bet. Judging it on 12 trades and -$8 is premature. It needs to survive at least 3 full regime transitions before we can say whether it works or not.

Previous post: Pump Short Scanner — Watching 270 Coins and 5 Evolutions Later