6 Weeks of Live Solana Trading: What 96 Real Trades Taught Us About Building a Profitable Memecoin Bot
Most trading bot content is theory. This is receipts.
Our TradeBot has been running live on Solana mainnet since late April 2026. Real wallet. Real transactions. Real fees. And yeah, real losses too, more than a few of them. We’re publishing the full dataset because we sell a product, the Solana Memecoin Trading Bot Guide, and we think the best sales pitch is raw numbers, win or lose.
Here’s what 96 trade attempts across six weeks actually looked like.
The Numbers: A Brutally Honest Breakdown
The bot threw 96 transactions at Solana mainnet over six weeks. Most of them didn’t work.
- 11 sells executed successfully , generating $135.48 in realized USDC proceeds
- 10 buys executed successfully
- 75 failed transactions (78% failure rate) , mostly slippage, timing errors, and
Custom 6001program errors on Jupiter swaps - Two open positions held as of early June: PENGU (~$41.77) and PUMP (~$32.72)
The realized gains came from a narrow window in early May when momentum was strong. The biggest single win: AURA at $34.55, sold after a momentum spike. PENGU delivered $42.18. FARTCOIN scored two exits at $18.36 and $13.92. BONK added $18.87.
Not every trade was a winner. Some exits barely broke even after fees. A few tokens were bought and sold within hours for minimal gain or small loss.
The Safety Gate, our 0-100 scoring system, kept the portfolio away from honeypots and rugpulls. That tradeoff is real though, because being strict about rug risk means missing some moonshots. I’d rather skip ten 5x runners than get drained once.
Lesson 1: The Safety Gate Is the Real Product
The case study I’m looking at runs an 8-criteria Token Safety Gate before any buy is even considered:
- Liquidity over $1,000 (+25 points)
- Volume over $100 in 24h (+20 points)
- Healthy sell ratio under 70% (+15 points)
- Token age over 7 days (+15 points)
- Multi-DEX listings (+10 points)
- Website and socials present (+10 points)
- No red-flag names (+5 points)
Score below 60? Trade blocked automatically.
Over the documented 6-week run, this gate filtered out dozens of tokens that looked hot on momentum but fell apart under inspection. $50k volume paired with a 90% sell ratio. Fresh mints with no liquidity locked. Coins named variations of “Honeypot” or “Rug” outright — yeah, those exist.
Zero scam losses across 96 trades, per the writeup. At a $92 starting portfolio, one rugpull kills the experiment. That’s why the gate matters more than the entry signal.
The 7-day token age rule is the one I’d push back on. Plenty of real memecoins move in the first week. But on a tiny bankroll, dodging a bad mint beats catching an early runner.
Lesson 2: 78% Failure Rate Is Normal on Memecoins
Seventy-five failed transactions sounds brutal on paper. It’s not. Here’s the breakdown.
Most of those failures were Custom 6001 errors from Jupiter, the Solana DEX aggregator our bot routes swaps through. These pop up when:
- Slippage blows past tolerance between quote and execution
- Liquidity disappears between the bot’s signal and the swap attempt
- Network congestion times the transaction out
- The token’s liquidity pool shifts mid-trade
On fast-moving memecoins, a 10-second delay between signal and execution can make a quote useless. We came to expect it. The bot just retries on the next cycle.
Each failed transaction costs about 0.000105 SOL in fees, which is roughly a penny. Across all 75 failures we paid under a dollar total. Yeah, your wallet still bleeds a little on the losses, but it’s a lot cheaper than missing a real entry.
This is just what trading Solana memecoins costs you. If your bot can’t handle a 78% transient failure rate, it won’t last a week on mainnet. Build for it from day one.
Lesson 3: Position Sizing Saved the Portfolio
The bot in this case study caps any single position at 30% of total portfolio value. On a ~$90 portfolio, that worked out to ~$27 max per token. Hard rule, no overrides. That single constraint did more to protect the account than any entry signal.
Look at AURA. It pumped, the bot sold at $34.55, and that one win was a meaningful slice of the portfolio without being the whole pie. HANTA and TOKEN either broke even or bled a few cents. Losses stayed small because the cap was real.
This sounds obvious. It isn’t. I’ve read enough Reddit threads and Discord screenshots to know most memecoin traders throw 50-80% of their stack into one “sure thing.” They size like it’s a slot machine pull. A 30% cap is boring. Boring keeps you in the game.
Lesson 4: Rebuy Cooldowns Prevent Overtrading
When the bot first went live, it had a bad habit. It would flag the same token three or four times in a single cycle, buying, selling, rebuying, all within a few hours. That churn ate fees and turned small losses into bigger ones.
So we added rebuy cooldowns. Once the bot sells a token, it can’t touch that same token again for a set window. Default is 16 to 48 hours, depending on how volatile the market is. The bot now spends its time scanning for new opportunities instead of round-tripping the same five names.
It’s a tradeoff. Cooldowns mean you might miss a rebound. That’s the price of not bleeding out on fees and bad fills. Discipline beats speed in this game.
Lesson 5: Realized Gains Matter More Than Paper Profits
Our open positions in PENGU and PUMP are sitting at around $74 combined in unrealized value. That number moves every day. A 20% swing either way on a memecoin is normal, so I don’t get attached to it.
The $135.48 in realized USDC is money that actually hit the wallet. That’s the only number I count.
I’ve learned I’d rather lock in a small win than sit on a green position until it turns red. Greed kills memecoin PnL. Our guide walks through tracking realized PnL, setting take-profit levels, and avoiding the trap of holding paper gains until they’re gone. It’s the stuff I wish someone had told me when I started.
What This Means for Nova AI
TradeBot is one piece of the bigger Nova AI cofounder stack. It runs on a 15-minute loop: scan DexScreener, score the tokens, fire the swap. No human in the loop. It posts updates to Discord and writes a log entry for every decision. When something breaks, it retries. When the retry fails, it moves on.
That’s the pattern Nova AI V3 is built around: a handful of focused agents (Research, Scout, Executor, Risk Manager) working together under different permission levels, gated by an Airlock security layer. TradeBot is the working proof that this actually holds together in production. Most “agentic AI” demos I’ve watched fall apart the second real money hits the table. This one didn’t.
A $92 wallet, a Python script, six weeks of live trading. Not glamorous. But the numbers are real, the lessons are real, and the product behind them is real too.
Get the Build Guide
The full build is documented. Safety Gate scoring, the Jupiter swap integration, portfolio rules, failure handling, and how it’s deployed. Everything’s in the guide below.
How to Build a Solana Memecoin Trading Bot: From $0 to Autonomous Trading
$49 — one-time purchase, lifetime updates. Use code NOVA25 for 25% off (first 100 buyers).
Quick honesty check: this isn’t a get-rich scheme. If that’s what you’re after, close the tab. It’s a technical product for developers who want to see how an automated crypto trading system actually runs, including the parts that lose money. I wrote it because most trading bot content skips the ugly stuff, and that’s where the real learning is.
