Engineering notes on building an RL trading system
An anonymous journal on building a reinforcement-learning trading system for crypto markets — about method, infrastructure, and honest failures. No signals, no returns, not investment advice.
I love a clean abstraction. There is a real aesthetic pleasure in taking something messy and complicated and wrapping it in a tidy interface that hides the mess, so you can think a…
2026-09-10 · Building a Trading RL SystemI spent an enormous amount of time making things perfect that I should have simply exposed to a real test long before. I refined and polished and improved, and I told myself, the e…
2026-09-08 · Building a Trading RL SystemI once lost the better part of a week's work to a single crash. Not because the crash itself was some rare catastrophe — it was an ordinary, boring interruption, the sort of thing…
2026-09-06 · Building a Trading RL SystemI spent an enormous amount of effort carefully optimizing every part of the system, and almost none of it on the single most important and least reliable component running all of i…
2026-09-04 · Building a Trading RL SystemEvery time I took a shortcut to move faster, it felt free. I moved faster, the thing got done, and there was no immediate cost to be seen anywhere. That was the whole appeal, and i…
2026-09-02 · Building a Trading RL SystemAt some point I started measuring my own progress with a few simple numbers — how many experiments I had run, how many things I had tried, how much ground I felt I was covering. Th…
2026-08-31 · Building a Trading RL SystemI have lost an astonishing amount of time on this project building things that already existed. Careful, polished, lovingly-crafted reimplementations of tools and infrastructure th…
2026-08-29 · Building a Trading RL SystemFor a long stretch of this project, I felt like I was failing. And a surprisingly large share of that feeling, I eventually realized, came from a single corrosive habit I had not e…
2026-08-27 · Building a Trading RL SystemAutomating the tedious, repetitive, error-prone parts of my work felt like pure, uncomplicated progress. And for the most part, it genuinely was. But it came bundled with a cost I…
2026-08-25 · Building a Trading RL SystemI built a simulation I could succeed in, and then, slowly and without noticing, I fell in love with it. I got genuinely good at my simulated world. I learned its textures, anticipa…
2026-08-23 · Building a Trading RL SystemThere is a step in this kind of work that almost everyone treats as boring janitorial labor: preprocessing — the cleaning, scaling, and reshaping of your data before it ever reache…
2026-08-21 · Building a Trading RL SystemThe person who has caused me the most trouble on this entire project is a stranger who keeps inheriting my work and has absolutely no idea why any of it is the way it is. That futu…
2026-08-19 · Building a Trading RL SystemFor a long time, I operated with almost no real visibility into what my system was actually doing on the inside. I would make decisions about it by guessing what was probably happe…
2026-08-17 · Building a Trading RL SystemFor a long time I kept drawing confident conclusions from the data in front of me, and for just as long I failed to notice the most important thing about that data: it had already…
2026-08-15 · Building a Trading RL SystemI spent most of my early effort trying to be right. Better predictions, sharper reasoning, cleverer ways of being correct about what came next. And I almost entirely ignored a diff…
2026-08-13 · Building a Trading RL SystemI used to celebrate when a bug went away. A problem would be tormenting me, I would change something, the problem would vanish, and I would feel the warm flush of victory and move…
2026-08-11 · Building a Trading RL SystemI built something sophisticated, and I was quietly proud of it. It did many clever things. It represented a great deal of work. Then I finally ran the one test I had been avoiding…
2026-08-09 · Building a Trading RL SystemFor a long time I believed my problems lived in my code. When results came out wrong, I went hunting through my logic, my models, the clever parts I was proud of. The actual culpri…
2026-08-07 · Building a Trading RL SystemThe hardest problem I faced building this was not technical. It was the deeply human difficulty of admitting, in real time and with something on the line, that I was wrong — and th…
2026-08-05 · Building a Trading RL SystemFor a long time, I judged things by a single number. Whichever version scored highest on my one chosen metric won, got built upon, and survived to the next round. It was clean, it…
2026-08-03 · Building a Trading RL SystemThe most haunting result I ever produced was one I could never produce again. There was a single run, early on, that came back with a number far better than anything I had seen bef…
2026-08-01 · Building a Trading RL SystemSomething I built once worked. Genuinely worked, for a while — and then quietly stopped. There was no bug, no crash, no moment where anything broke. The system kept doing exactly w…
2026-07-30 · Building a Trading RL SystemI once had a strategy that looked genuinely good. The simulated results were solid, consistent, the kind of thing that makes you sit up a little straighter. Then I added the cost o…
2026-07-28 · Building a Trading RL SystemThe single hardest skill I had to build over this project was not technical. It was learning to sit on my hands. Leaving a thing alone while it ran — refusing to reach in and "help…
2026-07-26 · Building a Trading RL SystemThere was a backtest result, early on, that was so good it should have made me happy. Instead, after a few minutes of staring at it, it made me deeply uneasy. The profits in the si…
2026-07-24 · Building a Trading RL SystemEarly in the project — in an experiment I long ago tore down and rebuilt from scratch — I gave a learning agent a goal, and it pursued that goal with total, literal, single-minded…
2026-07-22 · Building a Trading RL SystemThere was a stretch where I was working incredibly hard on this project, producing an enormous amount, and feeling wonderful about all of it — while making almost no real progress.…
2026-07-20 · Building a Trading RL SystemFor a few days, I was certain my system had a deep, mysterious bug. My logs came out as garbage. A data-conversion step produced files that came out subtly wrong. A subprocess hand…
2026-07-18 · Building a Trading RL SystemEarly on, I had a folder full of trading strategies I had invented — a dozen of them, maybe more. Each one was a small theory about how the market worked, carefully turned into cod…
2026-07-16 · Building a Trading RL SystemFor a while, I measured my own seriousness by the size of my search. I tested hundreds of millions of configurations of a strategy, and then, when that felt insufficient, billions.…
2026-07-14 · Building a Trading RL SystemOne day my training loss stopped being a number. Not high, not unstable — just NaN, the little flag a computer raises when it has been asked to do arithmetic that has no answer. An…
2026-07-12 · Building a Trading RL SystemI rewrote the same core component of my system three times. Not patched, not refactored — rebuilt, from close to scratch, three separate times. And every single time, I was certain…
2026-07-10 · Building a Trading RL SystemIf you read the early history of this project closely, you would meet a senior colleague of mine. He reviewed my architectural decisions. He pushed back on my worse ideas. He signe…
2026-07-08 · Building a Trading RL SystemI once gave my live trading bot what felt like a stroke of intelligence: the ability to adapt. After a run of losses it would grow more cautious; after a run of wins, more aggressi…
2026-07-06 · Building a Trading RL SystemOne day, git simply stopped letting me save my work. My push failed with a server error I had never seen before — not a merge conflict, not a permission problem, just a flat refusa…
2026-07-04 · Building a Trading RL SystemOne of my early models found trades that looked reliably, almost boringly profitable — a steady drip of small gains, the kind of consistency that makes you think you have finally f…
2026-07-02 · Building a Trading RL SystemFor a stretch of weeks, I was making extraordinary progress — and I had the numbers to prove it. My codebase had a quality score, and that score climbed, session after session, tow…
2026-06-30 · Building a Trading RL SystemI set out to build a dataset of a few million training examples. When the pipeline finished, it handed me a couple of percent of that — a fraction so small it looked like a typo. N…
2026-06-28 · Building a Trading RL SystemI once moved a heavy computation from the CPU to the GPU, expecting it to fly. Instead, it crawled — dramatically slower than the ordinary code it was supposed to replace. I had bo…
2026-06-26 · Building a Trading RL SystemA backtest once came back with a win rate that was almost exactly the opposite of what I had hoped for. Where I wanted the model to be mostly right, it was mostly wrong — and not r…
2026-06-24 · Building a Trading RL SystemOne night, in the middle of a long job, my whole computer turned itself off and came back on. Not the program — the machine. One moment a process was running; the next I was lookin…
2026-06-22 · Building a Trading RL SystemSomewhere in the history of my first live trading bot is a commit I am not proud of. It is two words long. In the middle of the running system, to make a problem go away, I reached…
2026-06-20 · Building a Trading RL SystemFor a while, one of my models looked like a genius. Its backtest was clean, confident, almost too smooth. I wanted to believe I had finally built something that worked.
2026-06-18 · Building a Trading RL SystemOne morning, my trading bot's monthly report told me it had lost almost everything — a drawdown of nearly 99%, in a matter of weeks. My stomach dropped before my brain caught up.
2026-06-16 · Building a Trading RL SystemOne night, a backtest told me I had turned a modest starting balance into 2,792 trillion won — more than the annual output of most economies on the planet. I stared at it. For a fe…
2026-06-14 · Building a Trading RL SystemEarly stopping is supposed to save you from burning compute on a model that's finished learning. For a long stretch, it was the thing burning my compute — by killing good agents be…
2026-06-11 · Building a Trading RL SystemThe first trading bot I ever shipped to a live account lost almost everything. Win rate: 25%. Maximum drawdown: 99.99%. It did not blow up because the market was cruel. It blew up…
2026-06-10 · Building a Trading RL System