AI Agent Pipeline for Selecting Research EAs from the MQL5 Code Base

Research
The MQL5 Code Base publishes many Expert Advisors (EAs). Some EAs include source code and implementation examples, which can make them useful material for learning MT5 and MQL5. At the same time, it is not appropriate to turn listed EAs directly into a profitability ranking or judge them from a short backtest alone. EA behavior depends heavily on the broker environment, symbol names, spreads, commissions, historical data, and parameter settings. This article treats EAs as research material for breaking down trading logic and risk management, not as tools for live operation.MQL5 Code Base MT5 Expert Advisors category through a small-scale review and presents a reader-friendly reconstruction of an experiment that used an AI agent to narrow research candidates step by step.
This is a technical and research article that treats EAs listed in the MQL5 Code Base as research material. It describes an experiment for organizing research candidates with an AI agent. It is not investment advice, a recommendation to run any EA live, or a guarantee of profitability. A detailed disclaimer appears later in the article.
目次

What You Will Learn

  • How to screen EAs listed in the MQL5 Code Base as research material
  • How to use list data, detailed metadata, static analysis, compile diagnostics, and smoke tests
  • How to separate research candidates, baselines, comparison candidates, and implementation examples
  • Why this workflow evaluates research suitability rather than profitability
  • What to watch for when handling statistics derived from backtest logs

Conclusion

This review was a small-scale probe, not a full crawl. It started with latest pages 1 to 3 and best pages 1 to 3, leaving 158 candidates after deduplication. From there, list scoring, detailed metadata collection, static analysis, compile diagnostics, and limited Strategy Tester smoke tests were used to organize roles that are easier to discuss in a public article.
  • 73958 ADX Trend Pullback EA: main research case for ADX trend pullback analysis
  • 59303 RSI Ea MT5: research candidate for an RSI baseline comparison
  • 68704 Price Action Intraday Trading – Expert for MT5: price-action comparison candidate
  • 59001 Risk Management EA Based on ATR Volatility: implementation example candidate for ATR-based risk management
This is not a profitability assessment. It organizes what should be read in detail next for a research article and which EAs may be useful as baselines or comparison targets.

Overall Experiment Workflow

The workflow can be divided into many detailed phases, but for readers it is easiest to understand it in five stages.
  1. List indexing
  2. Research-value scoring
  3. Detailed metadata and static analysis
  4. Compile Diagnostics and Smoke Tests
  5. Organizing candidates for additional validation
Research pipeline diagram showing the step-by-step workflow for organizing EAs listed in the MQL5 Code Base as research material
The important point is not to backtest every EA immediately. First, the candidates were logged from list and metadata information, and only EAs with visible research themes or structures were moved gradually into heavier validation.

List indexing

First, latest pages 1 to 3 and best pages 1 to 3 were collected from the MT5 Expert Advisors category in the MQL5 Code Base. This was a small-scale probe for research design, not high-frequency crawling or a full crawl.
  • Candidates after deduplication: 158
  • Appearances on list pages: 240
  • latest only: 38
  • best only: 38
  • both latest and best: 82
These numbers show that using only popularity order or only newest order can bias the candidate set. Some EAs appear only in latest, some only in best, and some in both, so logging the candidates and separating duplicates has value.

Research-value scoring

Next, research-value scoring was performed using only list information. The A/B/C/D/X labels used here are not ratings of EA results. They indicate priority within the research workflow, not profitability rank.
  • A: 13
  • B: 15
  • C: 43
  • D: 49
  • X: 38
Topics that received positive weighting included ATR, ADX, RSI, breakout, trend, risk management, and position sizing because they are often suitable for studying trading logic and risk management. Candidates with strong grid, martingale, or recovery wording were handled carefully. This was not a blanket rejection. The purpose was to decide whether they fit the research goal of this comparison article.

Detailed Metadata and Static Analysis

After list scoring, only 16 items were selected for detailed metadata collection. The detail pages were checked for information that list pages do not show, such as descriptions, authors, publication dates, and whether file lists were available. Then 7 items were selected for code retrieval and static analysis. The retrieved source files were local research copies, not material for reposting or redistribution. The workflow assumes respect for the MQL5 Code Base listing source, EA authors, copyright notices, and author attribution. Static analysis checked items such as OnTick, CTrade, OrderSend, CopyBuffer, iATR, iADX, iRSI, MagicNumber, StopLoss, and RiskPercent. External DLL use, WebRequest use, and notification-related use were not detected in this target group. At this stage, 6 items became targets for compile diagnostics.

Compile Diagnostics and Smoke Tests

Compile diagnostics were performed with MetaEditor CLI. Backtesting and optimization were not performed at this stage.
  • success: 4
  • warning_only: 2
  • failed: 0
All 6 items were in a condition that allowed them to move to limited Strategy Tester smoke tests. The smoke tests checked whether each EA could run, whether initialization errors occurred, and whether trades appeared under fixed conditions. Candidates with zero trades were not rejected immediately; possible condition mismatches and role differences were checked. As a result, 59001 Risk Management EA Based on ATR Volatility was treated more naturally as an implementation example for ATR-based risk management rather than as a normal backtest comparison target. 62732 EMA_RSI_RISK-EA had zero trades under the current conditions, and 73638 Long-Only Trend Breakout with Dynamic Risk Management was excluded from this normal comparison because it assumes NAS100 H4 and differs from the test environment.

Research Backtest Lite and Statistical Limits

For the 3 items left as additional validation candidates, fixed-condition run checks were performed on three symbols, EURUSD, GBPUSD, and USDJPY, across two timeframes, M15 and H1.
  • 3 EAs x 3 symbols x 2 timeframes = 18 runs
  • All 18 runs completed
  • All 3 items showed trades in 6/6 conditions
The trade counts observed in the research logs were as follows. Because Price Action Intraday Trading produced many trades, it is treated as a comparison candidate that should receive additional sensitivity checks for spreads, commissions, and execution costs.

Important Note on Statistics

In this environment, official Strategy Tester HTML/XML reports could not be obtained. Therefore, the values ​​for PF, DD, win rate, total profit, and total loss presented here are not those found in the official Strategy Tester HTML/XML reports. They are research approximations reconstructed from deal price pairs in the run logs and scaled to the net profit in the final balance. They cannot be used for strict comparative evaluation. Net profit is also not treated as a measure of profitability ranking. It is treated as one operational check metric reconstructed from logs under the current conditions. The backtest results are for research screening only and cannot be used for investment decisions, trading decisions, or live-operation decisions. With these statistical limits in mind, the overall position of this article is as follows.

Disclaimer and Position of This Article

This article is for general information and technical research purposes. It is not investment advice. This is a media article separate from investment advisory and agency services provided by Trilogy, Inc. Actual investment advice is provided only when a separate formal agreement has been executed in accordance with applicable laws and regulations. This article does not recommend live operation, purchase, use, or trading decisions involving any EA. It also does not guarantee the profitability or future operating results of EAs listed in the MQL5 Code Base. Backtest and log-derived figures in this article are reference information for research screening. They cannot be used for investment decisions, trading decisions, or live-operation decisions. The PF, DD, win rate, gross profit, and gross loss values are not official Strategy Tester HTML/XML report values. Because HTML/XML reports could not be obtained, they are research approximations reconstructed from deal price pairs in the run logs. They are also scaled to the net profit in the final balance, so they cannot be used for strict comparative evaluation. This review did not perform optimization or parameter search. As a rule, it was a limited research probe centered on default parameters. Results vary depending on the broker environment, symbol names, historical data, spreads, commissions, and execution conditions. This article is not intended to reproduce or redistribute source code. It respects the MQL5 Code Base listing source, EA authors, copyright notices, and author attribution, and treats the material in a research and educational context. For this media site’s editorial policy and its relationship to investment advisory services, please see the company profile page.
Candidate screening funnel showing the step-by-step narrowing process from 158 EAs to research role assignment

Validation Process Summary Table

Stage Purpose Result
List indexing Log a small-scale latest/best probe 158 candidates after deduplication, 240 appearances on list pages
Research-value scoring Classify priority in the research workflow A=13, B=15, C=43, D=49, X=38
Detailed metadata and static analysis Check descriptions, file structure, and code structure 16 items reviewed in detail, 7 items moved to static analysis
Compile Diagnostics and Smoke Tests Check whether limited run confirmation is possible success=4, warning_only=2, failed=0
Organizing candidates for additional validation Separate research roles Classified into main research case, baseline, comparison candidate, and implementation example

Overlap in the latest / best Small-Scale Probe

Category Count Meaning
latest only 38 Candidates that appeared only in latest
best only 38 Candidates that appeared only in best
both 82 Candidates that appeared in both latest and best

Trade Counts in Research Logs

EA Research Role Trades Confirmed Under Fixed Conditions Notes
RSI Ea MT5 Baseline 1256 For RSI baseline comparison
Price Action Intraday Trading – Expert for MT5 Comparison candidate 3896 High trade count; additional cost-sensitivity checks are needed
ADX Trend Pullback EA Main research case 616 For ADX trend pullback research
Note: This table shows trade counts confirmed under fixed conditions as a research-log metric. It does not indicate profitability ranking. In this environment, official Strategy Tester HTML/XML reports could not be obtained, so PF, DD, win rate, gross P/L, and similar values are not treated as official values. They cannot be used for strict comparative evaluation, investment decisions, trading decisions, or live-operation decisions.

Research Role Assignment

code_id EA Research Role How It Is Used in the Public Article
73958 ADX Trend Pullback EA Main research case Move to individual structure analysis
59303 RSI Ea MT5 Baseline Baseline for research comparison
68704 Price Action Intraday Trading – Expert for MT5 Comparison candidate Additional cost-sensitivity checks
59001 Risk Management EA Based on ATR Volatility Implementation example Implementation example candidate for ATR-based risk management
62732 EMA_RSI_RISK-EA Pending or excluded Zero trades under the current conditions
73638 Long-Only Trend Breakout with Dynamic Risk Management Pending or excluded Assumes NAS100 H4 and differs from the environment
Note: This role assignment is an organization method for a research article. It is not a recommendation to run any EA live or a profitability assessment.

What This Experiment Showed

Research candidates cannot be selected from popularity order or newest order alone. Instead of backtesting everything immediately, gradually narrowing the set with metadata and static analysis makes the research process easier to manage. Also, zero trades do not necessarily indicate a code-quality problem. The assumed symbol, timeframe, filter conditions, or EA role may not match the current conditions. Trading EAs, baselines, comparison candidates, and implementation examples should be considered separately. At least in the context of this article, EAs listed in the MQL5 Code Base are better treated as research material for breaking down trading logic and risk management, not as live-operation targets.

Limitations

  • This was a small-scale probe, not a full crawl
  • Official Strategy Tester statistics were not obtained
  • Results depend on the broker environment, symbol names, historical data, spreads, commissions, and execution conditions
  • The review was centered on default parameters
  • No optimization or parameter search was performed
  • This is not a live-operation evaluation
For these reasons, the results in this article should be treated only as research screening.

Next Steps

Future validation should naturally proceed by role, as follows.
  • Individual structure analysis of ADX Trend Pullback EA
  • Research comparison using RSI Ea MT5 as the baseline
  • Cost-sensitivity checks for Price Action Intraday Trading
  • Turning Risk Management EA Based on ATR Volatility into an implementation example
  • Expanding the same procedure to a full crawl
Again, these next steps are not meant to make profitability claims. They are additional work for reading EAs listed in the MQL5 Code Base as research material and improving the validation design. For this media site’s editorial policy and its relationship to investment advisory services, please see the company profile page.

FAQ

Is this article an EA recommendation ranking?

No. It is not a profitability ranking. It is a technical and research article that organizes a method for screening EAs listed in the MQL5 Code Base as research material.

Can I use this article’s results to decide whether to run an EA live?

No. This article is not investment advice and does not recommend live operation, purchase, use, or trading decisions involving any EA. The results should be treated only as research screening.

Are PF, DD, and win rate official Strategy Tester report values?

No. They are not official Strategy Tester HTML/XML report values. Because HTML/XML reports could not be obtained, they are research approximations reconstructed from deal price pairs in the run logs. Since they are scaled to the net profit in the final balance, they cannot be used for strict comparative evaluation.

Why not reject an EA immediately when it has zero trades?

Zero trades may mean that the assumed symbol, timeframe, filter conditions, or EA role did not match the current conditions. It cannot be used to make conclusions about code quality or profitability.

Does this article publish the source code?

No. This article is not intended to reproduce or redistribute source code, and it respects the MQL5 Code Base listing source, EA authors, copyright notices, and author attribution.

Where can I check this media site’s policy and its relationship to investment advisory services?

For this media site’s editorial policy and its relationship to investment advisory services, please see the company profile page.