- Entry date
- 1 August 2021
- Category
- Tooling
- Access
- 🔓 Public
Four strategy rewrites, one dashboard that survived all of them, and a lesson that took four years to fully land: the edge decays, the system doesn't.
- v1–v4
- EMA strategy rewrites on Binance
- 2021–2022
- active development window
- 1
- React admin dashboard built alongside the bots
Honest evaluation
Four separate EMA rule-set rewrites were tested against live Binance data; none found a durable edge, and the project was abandoned, not shelved.
What would prove or disprove it further
What would prove or disprove it further: the fixed-rule hypothesis, as tested, is closed. The open question it hands off is whether a different decision-making approach — an LLM reasoning jointly over several indicators instead of one hard-coded rule — fares any better, and that's precisely the test TraderBot ran four years later. A rerun of the same four EMA rule-sets against current market data would be a cheap, direct re-check of whether "fixed rules decay" still holds, if anyone wanted to confirm it again.
The evidence — full reasoning behind the verdict
Verdict: disproven. The question this project actually asked — can a fixed EMA-crossover rule-set find a durable trading edge on live Binance data? — was tested across four separate rewrites, and the answer that came back, each time, was no.
That's not a hedge or a "not yet": the report says plainly that none of the four rule-sets found a durable edge, and the project was abandoned rather than shelved as a placeholder. The mechanism is understood, too — an EMA crossover tuned to one market regime quietly stops working once the regime shifts, and nothing in a fixed-rule bot can detect that the shift has happened.
Before agents, before LLM APIs even existed in a usable form, there was a much older question: could code make a trading decision on its own? CryptoTrader was the first real attempt to find out, and it answered a narrower question than the one we asked.
What it was
A set of Python trading bots trading EMA (exponential moving average) crossovers against live Binance price data, wrapped in a React admin dashboard for watching positions and tuning parameters. Four rule-set iterations, v1 through v4, each one a fresh attempt to fix what the last one got wrong.
What we built
ResearchEach version kept the same shape — an EMA-crossover signal generator feeding buy/sell decisions into the Binance API — and changed the rules: different EMA periods, different confirmation filters, different position sizing. The dashboard, an HTML/CSS/JS admin template, was the one piece that didn't get rewritten each time; it became the place strategies were watched, compared, and killed.
What we learned — including the honest negative(s)
None of the four rule-sets found a durable edge. That's the honest result: this was abandoned, not shelved as a placeholder for "later." Fixed technical rules decay because the market that made them profitable stops behaving that way — an EMA crossover that worked in one regime quietly stops working in the next, and nothing in the bot itself knew that had happened.
The more useful discovery was accidental. The actual value wasn't in any strategy — it was in the system architecture built to test strategies quickly: the dashboard, the iteration harness, the plumbing that let a new rule-set go from idea to live paper-testing fast. Four dead strategies later, that harness was still worth something. That's a distinction — system versus strategy — that shows up again and again in later work.
Where it went / status
Abandoned in 2022, honestly and without regret — the hard-coded EMA approach had run its course. It sat dormant for three years until 2025, when the same underlying idea was revisited with an LLM (Claude Sonnet 4) doing the reasoning instead of a fixed indicator rule, holding several signals — RSI, MACD, Bollinger Bands — in mind at once rather than one crossover rule. The founder's own description of the difference is blunt: night and day. Whether that difference amounts to a durable edge is a separate, still-open question — see where that thread goes next.
What is still open — kept visible
The honest edges, next to the wins. This is what turns 🔬 into 🟢 — honestly.
- Abandoned: none of the four EMA rule-sets found a durable edge against the market.
- The dashboard and strategy-iteration harness outlived every rule-set they ran — the system was the real asset, not any one strategy.
- Revisited four years later with an LLM reasoning about multiple indicators instead of one fixed rule; the founder describes the difference as night and day.
Where this connects
Sources
- vocabotics project audit — CryptoTrader (four EMA strategy iterations, Binance), Aug 2021vocabotics internal project history · as of August 2021