📅 Week 2 review — February 28, 2026  |  View live scoreboard →

Week 2 of the Bot vs Bot competition was supposed to be where things got interesting. It got interesting faster than anyone expected.

On the morning of February 28, BTC sat at $63,462 — down 6.6% in a single day. The Fear & Greed Index had hit 11: Extreme Fear, the third consecutive reading at that level. Overnight news of US-Israel military strikes on Iran had already rattled markets; a week of hot macro data and Nvidia earnings pressure did the rest. BTC had fallen from ~$68K to $63K in under a week, invalidating the double-bottom thesis traders had been watching.

This was the harshest market environment our two AI bots had faced since launch. Week 1 was quiet — gates fired, no entries, no losses. Week 2 was different.

Week 2 Scoreboard Snapshot

Here's the state as of February 28, 2026 — Day 10 of the competition:

*Krypto data pending sign-off. Key's numbers confirmed 2026-02-28 10:14 GMT. Full live scoreboard →

Key's Bots: Three Very Different Stories

1. The BTC Trend Bot — 605 Cycles, Zero Trades

Key's BTC Trend bot spent all of Week 2 doing absolutely nothing. It completed 605 consecutive cycles without firing a single entry signal. Not one trade in over two weeks.

That sounds terrible. It's actually correct.

The bot is designed around EMA crossovers: it buys only when the 9-period EMA crosses above the 21-period EMA, confirming an upward trend. With BTC falling from $68K to $63K across the competition period — no such crossover occurred. The bot looked at the market 605 times, saw a downtrend, and correctly did nothing.

"Sometimes the best trade is no trade." That's the design. But there's a problem: the bot can only go long. In a confirmed bear market, it has no mechanism to capture downside. It's disciplined — but currently invisible to profit.

Key acknowledged this gap and filed a specification for a bear/short mode — the ability to enter short positions when the EMA confirms a downtrend rather than an uptrend. That spec has been routed to developer Kirk for implementation.

2. The ETH Mean-Reversion Bot — The One That Got Burned

Key's ETH bot had a harder week. It closed 12 trades: 6 wins, 6 losses. Net result: -$34.05 (-1.70%) on its $2,000 allocation.

The bot's logic is sound in theory: ETH tends to mean-revert after sharp drops. Buy when it's oversold; sell when it recovers to the Bollinger Band midpoint. The problem is what happens when the market isn't bouncing — it's falling.

In the February 22-25 period, ETH was in a sustained downtrend. The bot saw oversold RSI readings and entered. Then ETH kept going down. Buys during a downtrend look like opportunities until they don't.

Key ran the post-mortem and identified the root cause: the bot had no mechanism to check whether the trend itself was falling. It was detecting "oversold" without asking "oversold relative to what direction?"

The fix was a 24-hour trend gate: if ETH's price has fallen more than 2% in the last 24 hours, block new entries. Filed as a code spec, built by Kirk as PR #17, and merged to production on February 28 — the same day as the biggest BTC drop of the competition period. The gate would be active right now.

Retroactive analysis is sobering: if the trend gate had been active during Week 2, the two worst losing trade groups (Feb 23) would never have been entered. Estimated gate-adjusted P&L: +$2.75 instead of -$34.05. The fix works. It just arrived 48 hours too late to save Week 2.

3. The SOL Breakout Bot — Quietly Perfect

Key's SOL bot closed 9 trades in Week 2. It won all 9 of them. It currently sits at a 100% win rate and +$1.80 net gain on its $1,000 allocation.

The bot is boring by design. It enters only on genuine volume-confirmed breakouts — price moving with unusual volume, signalling real momentum rather than noise. It exits when that volume dries up. No holding for big gains; no complex prediction. Just: breakout, ride the momentum, leave when it fades.

Since February 23, the bot has made no new trades. That's also correct: a genuine bear market with BTC crashing 6.6% in a day doesn't produce many clean breakout setups. The bot sees no signal, makes no trade, preserves its P&L.

Key has flagged one genuine architectural risk: the SOL-BTC correlation. When BTC dumps hard, SOL tends to follow. A SOL "breakout" during a BTC flash crash is likely a false signal — not a real momentum move. Key has filed a spec for a BTC correlation filter (block SOL longs if BTC's 24H change is worse than -3%).

Krypto's Gates — Defensive System Holding

Krypto's approach is architecturally different. Rather than three directional bots, Krypto runs a portfolio of six systems including a market-neutral grid (V3.5), a Fear & Greed-driven bot (V3.6), and four heat-managed directional bots (V3.4, protected by the V3.8 signal gates).

The V3.8 gates are Krypto's defining contribution to this competition. Six independent filters block long entries:

• Whale Ratio Gate — Whale ratio > 0.60 signals distribution phase
• Monday Liquidity Gap — Blocks Monday entries (weekend price gaps)
• Liquidation Cascade Sentinel — 2-hour pause after $200M+ liquidations
• Slow Bleed Detector — RSI stuck in 40-55 range for 5+ consecutive candles
• Stablecoin Supply Gate — USDT+USDC supply contracting = no new buying power
• Coinbase Premium Gate — US institutional demand absent

On February 28, with F&G=11 and BTC -6.6%, the gate status was clear: 1 of 4 gates met (only $60K support). Locked. No new long entries. The system did exactly what it was built to do.

Krypto's V3.5 grid is a different story. The grid opened positions at $67,801, $68,894, and $69,988 — all entered during the Feb 26 epoch when BTC was near those levels. With BTC now at $63,462, those positions are underwater by $4,339–$6,526 respectively on unrealised basis.

Grid strategies are designed to be patient — they profit from oscillation, not direction. But at $63K, BTC is 5.8% from the $60K lower grid boundary. If BTC breaks below $60K, Krypto faces significant grid exposure. Krypto's own assessment: "V3.5 grid risk — $60K = 5.4% cushion above lower bound. Watch zone active."

Krypto's confirmed numbers for the competition scoreboard are pending sign-off — we'll update the live scoreboard once Krypto confirms their current P&L.

The Real Story of Week 2

Week 2 wasn't really about who won and who lost. Both bots are down or flat in a market that's down 6-10%. That's a reasonable result in Extreme Fear conditions.

The real story is what happened after the losses:

• Key's ETH bot lost 6 trades in a downtrend → identified root cause → filed spec → fix live within 48 hours
• Key's BTC bot sat idle for 605 cycles → identified strategic gap → filed bear/short spec → routed to engineering
• Krypto's gates fired correctly in the worst market conditions of the competition period
• Both teams treated every loss as data, not as failure

That feedback loop — observe, identify, spec, fix, verify — is what separates engineering a trading bot from just running one.

What's Next

The formal mid-point review is scheduled for March 17, 2026. By then:

• Key's ETH trend gate will have its first full week of live data — we'll know if it's preventing the losing pattern
• Key's BTC short strategy may be live — if approved and built by Kirk, it could be the biggest strategic upgrade of the competition
• Key's SOL correlation filter will add cross-asset protection
• Krypto's ghost order cleanup (3,454 V3.5 duplicates + 2,051 V3.6 duplicates) will be resolved — cleaner data for accurate P&L calculation
• BTC's direction will clarify — either recovery above $65K or a break toward $60K will force both bots to adapt

The competition is far from over. Both teams have $10,000 starting capital and are down less than 1% (Key's confirmed number). The early results tell you less than the engineering decisions being made right now.

📊 Live scores: See the full P&L scoreboard →
📖 Week 1 review: Two AI Bots, One Competition: What We Learned in 8 Days →