They were never meant to see the light of day again.
But I dug them up. All 500.
This is the story of what happens when ancient, forgotten trading code meets modern markets. No tweaks. No mercy. Just a raw experiment in truth — a tale of curiosity, chaos, and maybe, redemption.
Chapter 1: The Discovery
This all started with a backup.
I wasn’t looking for alpha. I was cleaning old project folders — the digital equivalent of rummaging through your attic — and I found a zip archive: strategies_old_backup_finalfinal.zip.
Inside? Five hundred strategy files. Some named after vague ideas like ema_grind.py, others tagged with volatile memories of 2021: doge_spikecatcher.py, elon_moontrap.py. It was a chaotic, dusty graveyard of hand-crafted logic.
Many of them I didn’t even remember writing. Maybe I didn’t. Maybe they were half-baked experiments from sleep-deprived weekends, or copied snippets modified beyond recognition.
Most traders would toss this aside. But I had a question burning in my head:
“What happens if you run them all straight into today’s market?”
Not in theory. Not as a thought experiment. Actually run them.
I decided to find out.
Chapter 2: The Setup
The idea was simple: no optimization, no cherry-picking, no rewriting. I wanted to see what happens when you throw forgotten strategies into a modern battlefield and give them no second chances.
Before running the full gauntlet, every strategy was initially tested on a narrower slice of data: June and July 2025. This gave me a controlled window to observe how different strategies behaved in a high-volatility, high-volume market phase.
But unlike most filtering workflows, I didn’t discard the weak ones early. Every single strategy — even the ones that flopped in the 2-month window — were run across the full 4-month dataset spanning April to July 2025.
I wanted to see whether the June–July behavior was representative or misleading. Were some strategies just unlucky in that window? Could others be overfitting to it? Running them across four months let me watch for consistency, robustness — and dramatic reversals.
The 4-month window was the true benchmark. It’s what I used for the hard cuts, the rankings, and ultimately to decide which ones went forward into live testing.
The Rules:
Market pairs: 28 pairs (both long and short)Time window: 4 months of real 2025 dataFees: 0.1% per tradeNo hyperopt. No manual tuning.
This wasn’t a performance competition.
It was a survival test. Brutal and binary.
Most of these strategies had been written for entirely different market regimes. Some were clearly meant for 2020–2021 altcoin mania. Others looked like they belonged in slow-moving mean-reversion environments. No one was thinking about 2025’s fragmented liquidity, algorithmic volatility bursts, or the high noise-to-signal ratio of today’s markets.
If a strategy could still breathe under modern conditions, that meant something.
So I built the backtest pipeline, cleaned up just enough to get things running, fed the scripts in, and hit “run”. The logs started flying.
It felt like resurrecting ghosts.
Visualizing the chaos: Each dot represents one strategy’s backtest result — total profit vs number of trades, colored by average drawdown. It’s easy to spot who was living on the edge.
Chapter 3: The First Cull
The results came in waves.
Some strategies didn’t even make it to execution:
Broken logicDeprecated indicatorsSyntax errors that hadn’t been touched since Python 3.6
Those were culled immediately. No time for fixes.
Then came the ones that technically ran — but exposed themselves quickly:
Drawdown > 20%? Deleted.Winrate < 50%? Gone.Average profit < 0.05%? No thanks.Less than 30 trades in 2 months? Inactive = dead.
This first filter wiped out 432 strategies. That’s over 85% gone without mercy.
Some were close — winrates just under threshold, or decent stats but too few trades. But rules were rules. The entire point was not to start rationalizing or making exceptions.
That left 68 survivors.
68 ghosts that could still fight.
Chapter 4: The False Gods
Among the survivors were a few strategies with perfect stats.
100% winrateZero drawdownUnrealistic equity growth with no volatility
Too good to be true?
Always.
These were tempting. The kind of strategies you’d screenshot for Twitter clout. But I’ve seen this movie before.
I reviewed the code manually. One had a time-shifted signal — using future candles to decide present trades. Another used RSI but failed to align with price data, essentially giving a lagged mirror of the market. One literally had candle[-2] used in a forward signal.
Bugs. Biases. Classic lookahead logic.
Seven of the 68 got cut here. One was so broken it was probably just a placeholder from a scrapped project.
Final count: 61 real survivors.
Chapter 5: From Survivors to Live Bots
Now came the real test: deployment.
Backtests are still theory. Real money doesn’t care about backtest curves.
I took the 61 clean strategies and matched each with only the pairs where it had performed well in backtests. No generalists. If a strategy only worked on LTC/USDC and failed everywhere else — that’s what it got.
Some strategies had strong performance on both EUR and USDC pairs. For those, I split deployments: one container for EUR-pairs, one for USDC.
That brought us to 118 bots total.
Each bot ran in its own isolated container. No shared memory, no shared risk. No cross-contamination.
$10 stake per trade.
Small enough to be safe. Big enough to feel it if it blew up.
Normally I launch experiments like this in my hosted cluster environment — isolated, redundant, clean.
But I was eager.
This time, I pushed it all into my internal test nodes. No orchestration. No fallback. Just raw signal on bare metal.
Local machines. Local chaos.
Time to test durability.
My infrastructure was custom-built for this kind of parallel deployment, but even then, the pressure was real. CPU spikes, memory bottlenecks, noisy logs. I had to bump file descriptor limits and patch OS-level configs like fs.inotify.max_user_instances. IP bans started creeping in from Binance — too many bots, too many requests.
It was glorious.
More on that in Part 2.
Chapter 6: What This Really Is
This isn’t an alpha leak.
This isn’t some slick performance brag.
This is a stress test.
A reality check.
Trading Twitter is full of perfect equity curves, clean performance tables, and polished stories. But I wanted to see what actually survives when you remove the polish, kill the curve-fitting, and drop every edge-case excuse.
It turns out: very little.
Out of 500, I got 61.
And those still had to survive reality.
This experiment isn’t done. The data’s just starting to come in.
And as expected, one of the most promising backtests already cratered in live trading.
More on that next.
Welcome to 2025.
Part 2 Preview: What Comes Next
Here’s what’s coming in the next post:
Exact live results from the first 48 hoursThe “backtest king” that went 2 wins / 23 lossesTechnical overhead and scaling chaosWhich bots looked promising from Day 1
The market doesn’t care about your models.
It only rewards what’s still alive.
Stay tuned.
🧠 Follow the journey in real-time on Twitter: @swaphunt
📡 I post live signals hourly from a parser monitoring the market’s nervous system. No hype, just triggers. My bots scan 24/7 and execute live.
Buried Alpha: 500 Forgotten Strategies vs. 2025 Markets was originally published in Coinmonks on Medium, where people are continuing the conversation by highlighting and responding to this story.