The Optimizer — walk-forward validated, not a chatbot.
Backtest → optimize → validate on data the optimizer never saw. A quant engine, not a prompt.
Optimizer requires PRO · $19/mo · browsing validated strategies is free
The pipeline
Three steps. The third one is the point.
Backtest
Your strategy is replayed over real historical price data — every entry, every exit, every bad day included. You get the full metrics: total return, maximum drawdown, win rate, Sharpe ratio, trade count. Not a summary — the evidence.
Optimize
A parameter search finds the settings that actually worked across that history. The honest caveat: finding the past’s best settings is the easy part — and the dangerous one. Search enough combinations and something will look brilliant by pure luck. On its own, an optimized backtest is a curve-fit story.
Walk-forward validation
So the chosen settings are then tested strictly out-of-sample — on a later slice of data the optimizer never saw. If the edge survives data it wasn’t fitted to, it holds up; if it degrades, we report by how much. This is the step that separates a validated strategy from a curve-fit story.
The same rigor — out-of-sample validation, regime breakdowns, no hidden results — is documented on the track record page and in the Anny Line methodology.
Real data, not a demo
What the pipeline has produced so far
These numbers come from published strategy ideas — AI-discovered strategies backtested on real market data and validated out-of-sample. They update automatically as new strategies are published.
Backtest performance, not a promise of returns. Every card on the strategies page shows the full record — return, Sharpe, win rate, maximum drawdown, trade count, regime breakdown, and out-of-sample degradation.
The honest architecture question
“Couldn’t I get the same from a news feed and the OpenAI API?”
Fair question — a lot of “AI trading” products are exactly that. Here is the specific difference.
An LLM predicts plausible text
Ask a chatbot whether a strategy works and it produces a fluent, confident answer — without ever running the strategy. It has no price history in front of it, simulates no trades, and computes no drawdown. It is an articulate guess, and it sounds equally confident when it’s wrong.
The Optimizer executes deterministic math
It replays your rules over years of real candles, computes every metric from actual simulated trades, and then re-tests the chosen parameters on data it never saw. Same inputs, same outputs, every time — auditable, repeatable, falsifiable. A prompt cannot change the numbers.
Where Anny does use LLMs
Ask Anny — reasoning about your portfolio, explaining backtest results, and answering questions in plain language. Language is what LLMs are good at, so that’s where they run.
Where it never does
The CFO Anny Line regime engine and the Optimizer’s math. Indicators, backtests, parameter search, and walk-forward validation are deterministic code — no LLM anywhere in that loop.
Frequently asked questions
What is walk-forward validation?
The strategy’s parameters are optimized on a training window of historical data, then tested on a separate, later holdout window the optimizer never saw. Anny reports the degradation — how much worse the strategy performed on unseen data. It’s the standard quant defense against overfitting: a real edge survives data it wasn’t fitted to; a curve-fit one doesn’t.
Does a good backtest guarantee future returns?
No. Past backtest performance does not predict future results — crypto is volatile and you can lose your entire investment. Walk-forward validation reduces the risk that a backtest is pure curve-fitting; it cannot remove market risk. That’s why every published strategy shows its drawdowns and out-of-sample degradation, not just the good numbers.
What data does the Optimizer use?
Public Binance daily candles and standard indicators — RSI(14), MACD(12,26,9), ADX(14), EMAs, volume vs the 20-day average. No proprietary feeds: open the same candles on TradingView or Binance and you’ll get the same numbers. That’s deliberate — results you can verify beat results you have to trust.
How is this different from a signal group or a GPT chatbot?
A signal group asks you to trust unverifiable calls. A GPT chatbot generates plausible-sounding text without running any test — it has no price history in front of it and computes no drawdown. The Optimizer executes deterministic backtests over real price history and then validates the result on data it never saw. You don’t have to trust the output — you can audit it.
What does the Optimizer cost?
The Optimizer requires the PRO plan ($19/mo). Browsing already-validated published strategies — full backtests, drawdowns and out-of-sample degradation included — is free, as is the live CFO Anny Line on BTC.
Stop trusting. Start validating.
Run your strategy through backtest, optimization, and walk-forward validation — or start by browsing strategies that already passed all three.
