Broad public-source intake before deduplication and classification.
MSR-RPT-002COMPLETED RESEARCH PROGRAMcompleted
MSRLab public research edition · 28 Aug 2026
Mining MQL5 for trading edges
What five research phases searched, what survived scrutiny, and why a large negative-result program was closed.
Every observed record received a final research or safety route.
Public-source-derived and independently expressed cases combined.
No candidate passed the complete evidence chain.
01
The evidence-led verdict
The MQL5 program produced no new strategy that completed MSRLab's evidence chain from public idea to realistic execution and portfolio usefulness. It did produce a durable intelligence and validation system, a permanent negative-evidence record, and a small watchlist that remains explicitly non-deployable.
02
What the work established.
Each section states the public conclusion while keeping reconstructive details out of the publication pipeline.
Phase 1 · Build the map
The first phase treated MQL5 as a noisy research corpus rather than a catalogue of ready-made systems.
- Separated trading ideas from utilities, panels, calculators, and operational tools.
- Classified duplicated mechanism families and quarantined recovery, averaging, and uncontrolled-tail designs.
- Introduced cached, rate-limited acquisition so evidence collection remained reproducible.
Phase 2 · Convert descriptions into hypotheses
Research moved from collecting files to independently expressing falsifiable market ideas across asset and horizon families.
- 5,252 labelled concept-family entries, 436 findings reports, and 604 research runners were preserved.
- Chronology, realistic costs, neighboring variants, market transfer, and portfolio contribution were tested before promotion.
- A provisional cross-market watchlist remained useful research knowledge but did not earn capital authority.
Phase 3 · Audit downloadable public source
The program narrowed to a complete observed listing of free MT5 Expert Advisors with legitimately accessible source.
- Forty listing pages produced 1,590 records, 578 archived source packages, and 199 compilable program entrypoints.
- Eighty-five frozen source-derived runtime cases and 142 independent native cases produced no validated survivor.
- Operationally dangerous behavior and very high trading activity were preserved as negative evidence rather than optimized away.
Phase 4 · Test commercial popularity as information
Commercial listings, public descriptions, review histories, and linked public records were treated as research leads—not proof.
- The program mapped 210 popular products, parsed 37 detailed pages, and reviewed nine linked public performance records.
- Popularity, ratings, screenshots, and young performance histories did not substitute for chronology, risk quality, or executable economics.
- Some lower-level observations were retained, but none became a reproducible trading edge.
Phase 5 · Start from attributable track records
The final phase scored public traders on longevity, observability, and the strength of attributable methodological traces.
- Ten public records were scored; two supplied enough public detail for bounded hypothesis work.
- The visible mechanisms failed chronological, stability, or realistic-execution tests in their reconstructed forms.
- The conclusion is deliberately narrow: public traces were insufficient to reproduce the observed edge.
What the program changed
The lasting output is a stronger laboratory process and an explicit stopping discipline.
- Discovery provenance, deduplication, static safety review, isolated compilation, and serialized broker testing are reusable infrastructure.
- A signal may contain information without supporting a trade after scheduling, spreads, rollover, sizing, and portfolio interaction.
- Negative evidence prevents future agents from repeatedly mining the same closed families under new names.
Progress is a decision,
not just a curve.
A large public library can be information-rich without yielding durable alpha. Repeated mechanisms, incomplete decision rules, hidden tail risk, weak chronology, and execution frictions eliminated the apparent opportunities. Closing the broad program is an evidence-based decision, not an absence of work.
Next evidence gate
Keep the program closed. Reopen only for a genuinely new and attributable mechanism, a new immutable public snapshot, mature supporting evidence, or fresh broker-quality data for one predeclared watchlist question.