Strategy Builder vs. Strategy Discovery Modes
Understand the differences between Strategy Builder (Coding Mode) and Strategy Discovery (Research Mode).
Ascentrade features a dual-mode conversational AI assistant tailored for algorithmic and manual traders. By switching between modes, you change the operational focus and latent capabilities of the AI.
Strategy Builder (Coding Mode)
Best for: Writing, compiling, testing, and deploying custom automated trading scripts (Expert Advisors/EAs).
- Persistent Actions: The builder is authorized to create, save, and overwrite strategy versions in your permanent database.
- Sandboxed Execution: It writes raw Python/MT5 code and executes it in an isolated container to verify syntax, correctness, and logic.
- Deployment: Once the Builder compiles a stable version of your strategy, you can load and deploy it to a live AI Watcher.
Strategy Discovery (Research Mode)
Best for: Factual trade audits, behavioral reviews, literature searches, viability checks, and brainstorming.
[!NOTE] Read-Only Safeguard: Strategy Discovery operates in a strict read-only context. It cannot write or modify strategy files, compile code, or persist automated algorithms. If you ask it to write code, it will decline and instruct you to switch to Coding Mode.
Your Interactive Pair Trader (Sounding Board)
Beyond static research, Strategy Discovery is designed to be a conversational sounding board—a collaborative "pair trader" you can bounce ideas off of. Use this mode to:
- Bounce Ideas: Discuss hypothetical setups or risk parameters before writing code.
- Debate Concepts: Query market mechanics and indicators to refine entry/exit thresholds.
- Review Behavior: Gain external validation on whether a trade followed your strategy guidelines.
Latent Capabilities & Multi-Agent Crew
Discovery mode leverages a specialized team of autonomous sub-agents that work behind the scenes to audit your trading behavior:
- Factual Puller (
journal_facts_agent): Queries your database to retrieve historical trades, journals, and performance metrics across single or multiple accounts. - Market Data Retriever (
market_data_agent): Calls MT5 terminal bridges directly to retrieve live symbol quotes and OHLCV chart bars for context analysis. - Viability Skeptic (
viability_skeptic_agent): Critiques strategy viability and metrics (summarized metrics, skepticism). - Cognitive Bias Critic (
bias_critic_agent): Scans your trade logs and notes to spot emotional behaviors (e.g., revenge trading, loss aversion, over-trading, FOMO) and suggests mental adjustments. - Web Research Agent (
discovery_research_agent): Queries academic journals and online trading sources to match your strategy against industry standard methodologies (e.g., Mean Reversion, Momentum models).
Prompt Comparison Table (Simple to Complex)
| Complexity | Strategy Discovery (Research / Sounding Board) | Strategy Builder (Coding / Creation) |
|---|---|---|
| Simple | "What is the formula for calculating Average True Range (ATR)?" | "Create a simple script that opens a buy order when RSI is under 30." |
| Intermediate | "Analyze my recent trades on @Account1 and tell me if I'm closing wins too early." | "Build a strategy using a 50 EMA and 200 EMA crossover, including a 2% stop-loss." |
| Complex | "Let's brainstorm a pair trading strategy for Gold and Silver. How would we handle cointegration?" | "Write a multi-timeframe strategy that checks H4 trend direction before executing M15 RSI entries." |
| Image-Based | [Paste chart screenshot] "Analyze this EURUSD entry chart. Did I enter too late compared to the 20 EMA rule?" | [Paste setup screenshot] "Code an EA that matches this pattern. Buy on breakout of the resistance line shown in the image." |
Recovering from Abrupt Agent Stops
If the agent assistant (in either Builder/Coding or Discovery/Research mode) stops processing abruptly due to a temporary step timeout, connection loss, or token limit:
- Simply type and send the word "continue" (case-insensitive) in the chat window.
- The system will automatically resume execution and pick up exactly where it left off, maintaining full historical context and tool outputs.