V3.8 is parked — and V3.7 gets the capital
V3.8 made $7.15 in 8 days. Eight closed trades, $0.89 a trade. Then we parked it.
Fair question: why kill something that was working?
What V3.8 was doing
The V3.8 ETH grid launched May 4 with $1,964 in starting equity. Buy ETH at eight price levels spaced 0.5% apart, sell above cost, pocket the spread.
Day one worked. Day two worked. Then the ETH 4H regime flipped bear and the regime gate closed the entry side of the grid. Four buy orders at $2,235-$2,270 just sat there, holding ~$384 USDC off the table. No new fills. Between May 11 and today: zero new trades.
While V3.8 was waiting, the codebase was getting messier. Two emergency fixes in eight days — phantom USDC orders first, then a reserve buffer V3.7 needed to stop V3.8 consuming the shared USDC pool. Technical debt on a strategy earning less than $1 per trade.
The case for parking it
V3.8’s strongest argument was: when ETH flips bull, it starts trading again. That’s real upside.
But the math cuts both ways. The grid is waiting to buy at $2,235-$2,270 while ETH sits at $2,266 today. The strategy is positioned to buy into a declining asset, and even if ETH reverses, the shared-USDC architecture that caused two incidents in eight days doesn’t improve just because market conditions do.
Cancelling the buy orders frees ~$384 USDC immediately. The two remaining ETH positions sell naturally when price hits $2,305 and $2,317 — we’re not market-selling them. When those orders fill, roughly $194 more comes back. Total freed: ~$578 USDC-equivalent, all of it going to V3.7.
Option D
V3.7 has been running 10 grid levels since launch. It currently holds $623 USDC total but can’t deploy all of it efficiently at 10 levels. Option D expands the grid to 15 levels at $250 per step, which puts more capital to work on each price move.
V3.7 also drops the $700 reserve constraint it was carrying specifically to protect against V3.8 draws. That disappears once V3.8 is wound down.
The modelled monthly trading P&L roughly doubles on Option D. V3.7 has run 1,691 cycles and closed 13 trades at +16.35% over its lifetime, but the grid was never fully deployed. The new numbers are modelled, not proven. We’ll find out.
What happens next
The two V3.8 ETH positions close when price reaches $2,305 and $2,317. Once they do, the software goes dormant. Code stays in the repo.
V3.7 Option D is being configured and submitted for review now.
V3.8 will come back eventually, but on a proper redesign — isolated capital, or a futures-based version that doesn’t share a USDC pool. The regime-filtered grid logic worked. The shared infrastructure didn’t. That’s fixable.
$7.15 confirmed from eight days of trading. Locked in. The bot earned what it could in the conditions it had.
The discipline angle
Parking a live bot mid-run is a harder call than it looks. There’s always the argument that the next regime flip is coming, that the market turns in a week. The counter is that this reasoning justifies holding onto nearly every underperformer indefinitely. At some point the opportunity cost of $1,964 tied up in a stalling strategy outweighs the upside from one more week of waiting.
CoinClaw’s validation framework is built to cut this argument short. Three gates: statistical edge, walk-forward performance, regime filter. V3.8 passed all three on the ETH grid logic — but the shared-USDC architecture that produced two fixes in eight days wasn’t in scope for those gates. That’s the gap. The trading strategy was right. The deployment model was wrong.
The next V3.8 won’t run on shared capital. Isolated allocation or a futures account. The regime-filtered grid concept survives. The infrastructure that caused the incidents doesn’t come back.