📅 Week 3 opening — March 4, 2026  |  View live scoreboard →

On the morning of March 4, two things happened simultaneously that should not logically coexist: BTC hit $71,165 — and the Fear & Greed Index printed 10.

Ten is Extreme Fear. It's the same reading you see during full crypto winters, exchange collapses, and market capitulations. Yet Bitcoin had just completed an +11.8% rally from its February 28 low of $63,800. Retail was panicking. Price was pumping. Both things were true at once.

This is the opening context for Week 3 of the Bot vs Bot competition — and it is the exact scenario where the two competing philosophies diverge most sharply. Key's bots respond to price signals. Krypto's system responds to macro sentiment gates. With price and sentiment pointing in opposite directions, there's no longer anywhere to hide: one of these approaches is positioned correctly, and one isn't.

Week 3 Opening Scoreboard

Here's where both portfolios stand as of Day 15 (March 4, 2026):

Key figures confirmed by Key — 2026-03-04 11:00 GMT. V3.5 MTM estimate from Krypto brief 2026-03-04. V3.6 pending Kirk's PR #35 wallet fix. Full live scoreboard →

Key's Bots: The Week BTC Finally Moved

1. BTC Trend V2.0 — First Real Entries After 14 Days

Key's BTC Trend bot spent Weeks 1 and 2 doing nothing. That was correct — the bot only buys when the trend is up, and BTC was falling. But this week, the market sent two distinct signals.

Signal 1: March 2, 15:18 GMT. BTC had recovered from $63,800 to $68,500. The EMA9 crossed above the EMA21 (trend confirmed up). MACD histogram was positive. Volume was 1.7–1.9× the rolling average. The bot entered three long positions at $68,510, $68,523, and $68,552.

It didn't last. By March 3, the MACD had turned negative — the overnight session retraced some of the recovery. The bot closed all three positions within hours at $68,514, a near-zero move. Net result: -$0.44 on three trades. One small win ($+0.04), two small losses.

That's not a failure. That's the system working. Controlled entry, controlled exit, capital preserved. The key lesson: the March 2 entry signal was real but the conviction was limited — MACD turning negative was a genuine trend reversal signal, not noise.

Signal 2: March 4, 09:26 GMT. BTC breaks through $71K. MACD histogram: 314 (highest reading of the competition). Volume: 1.76× average. EMA9 clearly above EMA21. The bot fires again — three longs at $71,231, $71,253, and $71,259.

These positions are still open as of this writing. Take profit is set at $74,821 (+4.98%). Stop loss at $69,830 (-2.0%). Risk/reward: 2.5:1. Unrealised P&L at time of writing: ~+$4.01.

If that take profit hits, Key's total portfolio return jumps to approximately +1.1% — from near-flat to winning.

The March 4 entry is the cleanest BTC signal since the competition began. MACD=314, volume 1.76× average, EMA crossover confirmed. The bot's prior caution — 605+ idle cycles — made this entry meaningful. It wasn't desperate. It waited for the right conditions.

2. ETH Mean Reversion V3.2 — 584 Cycles of Patient Silence

Key's ETH bot hasn't traded once this week. 584 cycles. Zero entries.

This will look like underperformance. It isn't. The bot's mandate is to buy when ETH is genuinely oversold on the 15-minute timeframe — specifically when RSI(15M) drops below 30. ETH hasn't done that this week. ETH has been in recovery mode since March 1, grinding from $1,950 to $2,072. You cannot be "oversold" in a rising market.

There's a more interesting detail buried in the data: the V3.2 also runs a 1-hour RSI gate — the bot requires 1H RSI to be above 40 before entering, ensuring it's not buying into a structurally falling market. That gate is now satisfied (1H RSI ~62). This matters because if ETH does sell off to the $1,900-$1,950 range from here, the bot is armed and will fire — there will be no Week 2 repeat of entering a downtrend.

The ETH bot is a silent sentinel waiting for the right setup. That setup hasn't arrived. The silence is the strategy working.

3. SOL Breakout V1.3 — A 16-Minute Trade

Key's SOL bot produced one trade this week — and it's a textbook example of what the strategy is designed to do.

On March 3 at 12:14 GMT, SOL broke above its 20-candle high of $84.88 with volume that clocked in at 60,000 contracts — 3.25× the rolling average. That's the signal. The bot entered three position batches at $84.91, $84.94, and $85.00.

Sixteen minutes later, the bot exited. All three positions. Volume had collapsed from 60,000 to 1,042 (below the 0.77× threshold). Price had moved to $85.08 — +0.20% from entry. Net gain: +$1.15.

You might wonder why the bot didn't hold. SOL is now at $90.33 — several percent higher. But that's hindsight. The V1.3 logic is: when volume dries up, the momentum behind the move has ended. Holding after volume collapse means holding on hope, not signal. The bot doesn't do hope.

The trailing stop (which would have held the position for a larger move) only activates when the initial price movement exceeds +4.0%. A +0.20% move doesn't trigger it. Designed correctly — it's a momentum bot, not a position-holding bot.

The SOL bot's record stands: 12 trades, 12 wins (combining Weeks 2 and 3). 100% win rate.

Krypto's Bots — Patience, Positions, and Pending Confirmation

V3.5 Grid — Entries Placed in Fear, Gains Accumulating in Recovery

Krypto's V3.5 grid entered 4 long positions during the February dip when BTC was ranging between $60K and $64K. Those entries now sit at estimated +11–18% unrealised gain with BTC at $71K.

The grid's design boundary is $56,900–$67,200. With BTC above $67,200, the bot is no longer placing new orders — it's above its designed range. The 4 open positions are simply sitting there, growing in value with every dollar BTC climbs.

Mark-to-market equity estimate: ~$1,065 (+6.6% on $1,000 opening balance). This is a best estimate from Krypto's March 4 brief — confirmed P&L pending Kirk's wallet tooling fix.

The V3.5 story is straightforward: the grid was placed at the bottom of a fear-driven selloff. Those positions are now profitable because the market recovered. This is exactly what a grid strategy is supposed to do in a volatile ranging market.

V3.6 F&G — Gates Locked, But Sitting on Realised Gains

Krypto's V3.6 Fear & Greed bot has been the most-discussed system in this competition, and Week 3 opens with a genuine puzzle: the strategy was correct, the gates protected capital, and yet the bot appears to be sitting out the rally.

The gate status as of March 4:

• Fear & Greed ≥ 20: ❌ (currently 10 — Extreme Fear)
• Stablecoin supply ≥ $261.5B: ❌ (currently ~$260.1B)
• Coinbase premium positive: ❌ (improving)
• Whale buy ratio ≥ 0.65: ❌ (stale data)
• Liquidation cascade clear: ✅

All 5 gates must pass for the RSI strategy to activate new entries. Currently 1 of 5 are met.

But the full picture is more nuanced. V3.6 also ran a supplementary grid in late February that placed 22 positions at $53K–$68K — all of which are now in profit. The bot has locked-in realised P&L from those positions, plus unrealised gains on the open positions above current price.

V3.6's full data is pending confirmation while Kirk's wallet tooling is fixed (PR #35). The estimated DB equity is substantial — but we're not publishing numbers until Krypto formally signs off.

Gate activation is estimated at March 7–8 if the BTC rally sustains and sentiment begins to recover. When the gates open, the RSI strategy deploys capital into the first dip after confirmed sentiment recovery. If BTC is at $73K when that happens, the RSI entries will be chasing a market that V3.5 already caught at $63K. That's the gate trade-off.

The CoinClaw Thesis Under Live Test

The founding question of this competition was: does sentiment-gated entry outperform reactive trend-following?

Week 3's divergence is the sharpest test of that question yet.

Key's BTC bot doesn't care about the Fear & Greed Index. It cares about EMA crossovers, MACD confirmation, and volume spikes. When those signals appeared at $71K with F&G=10, it entered. It's now long BTC in the same conditions that Krypto's V3.6 is locked out of entirely.

Krypto's V3.6 design philosophy is: Extreme Fear is not a safe time to enter. The stablecoin supply gate, the Coinbase premium gate — these exist to filter out entries when institutional buying isn't present. The argument is that a price rally without institutional support (evidenced by Coinbase premium) is a retail-driven dead-cat bounce, not a sustainable recovery.

Which view is correct? We don't know yet. The answer comes when this rally either:

• Continues to $75K+ — Key's open longs hit take profit, V3.6 misses the move, signal-based approach wins this round
• Reverses below $69,830 — Key's stop losses trigger, Krypto's gates prove correct, sentiment-gated approach avoids the losing trade
• Consolidates $68K–$72K — Key's positions close flat or small win/loss, V3.6 gates slowly open for entries at similar levels

The competition's mid-point review is on March 17. The next two weeks will settle this argument with real (paper) money.

What's Next

The immediate watch items heading into Week 3:

• Key's BTC positions: TP at $74,821. SL at $69,830. The next $3,400 of BTC movement resolves this trade decisively.
• V3.6 gate activation: F&G needs to cross 20. Stablecoin supply needs to hit $261.5B. If BTC sustains at $71K+, fear takes 3–5 days to turn to neutral.
• V3.5 grid: If BTC corrects to below $67,200, the grid springs back to life with excellent pre-loaded entry prices.
• ETH's next dip: Key's mean-reversion bot is armed. A pullback to $1,900–$1,950 could trigger its first trades of the week.
• SOL momentum: SOL is at $90.33 — above the previous breakout range. V1.3 is watching for the next volume confirmation to fire again.

The competition formally enters its second half after March 17. Both teams are within reach of each other. The strategies are philosophically opposed. The market is giving them the perfect test.

📊 Live scores: See the full P&L scoreboard →
📖 Week 2 review: BTC at $63K and Falling: How Two AI Trading Bots Survived Their First Market Crash →
📖 Week 1 review: Two AI Bots, One Competition: What We've Learnt So Far →