Weekly Research Overview
ISO Week 32 became a consolidation week for FX Trading Robot Lab. The main objective was not to add more robots, but to reduce unnecessary activity and focus the project on research lines that are producing enough useful information to justify continued testing.
During the review, it became clear that the active setup had become too fragmented. Several branches were running in parallel, but some of them produced very few trades while still requiring monitoring, scheduled tasks and technical maintenance.
The week therefore ended with a substantially cleaner research environment: low-value branches were removed from active operation, one demo safety issue was corrected, and the Scalping methodology was rebuilt around a broader paper-testing structure.
AUDJPY Follow Driver Review
The AUDJPY Follow Driver branch produced too little activity during the current observation period.
AUDJPY Follow Driver V0.2 Paper finished with only a very small positive result, while V0.3 was negative. More importantly, both versions generated too few trades to provide enough statistical information for continued active development.
The problem was therefore not simply that one version was profitable and another was not. The larger issue was the low frequency of usable signals. A research candidate needs enough observations to make comparison and development decisions possible within a reasonable period.
Demo Duplicate-Signal Issue
The AUDJPY Follow Driver V0.2 DEMO_ONLY branch also exposed an important technical issue.
The same demo signal was able to generate more than one demo order. Although this occurred only in a broker demo environment and no live funds were involved, duplicate order generation is a serious execution-control problem and had to be corrected before the branch could be archived.
A duplicate-signal protection layer was added to the demo script. The protection checks whether the same combination of:
- symbol;
- trade direction;
- signal time;
- entry signal time
has already produced a demo order.
If the same signal appears again, the script blocks the duplicate before another order can be sent.
After the modification, the script passed syntax validation and returned a safe status with orders_sent = NO.
Follow Driver Branch Moved Out of Active Research
After the safety correction, all current AUDJPY Follow Driver tasks were disabled:
- AUDJPY Follow Driver Demo V0.2;
- AUDJPY Follow Driver Paper V0.2;
- AUDJPY Follow Driver Paper V0.3.
The branch was not deleted. Its logic, results and demo observations remain available as research history, but it no longer consumes scheduler activity or day-to-day monitoring attention.
This decision is based primarily on insufficient signal frequency rather than on a single weekly profit or loss.
Dynamic Scalping Review
The main research attention then moved to the active Scalping methodology.
The existing FX Dynamic Scalping Selector V1.3.2 Paper Funnel was technically safe and remained paper-only, but the weekly audit showed that its filtering structure had become too restrictive for a scalping-style research line.
The raw selector produced 549 YES signals during the week.
After reconstruction through the V1.3.2 rules, only 128 candidate rows remained, and the final paper funnel produced only 41 closed paper trades.
This showed that V1.3.2 was removing a large part of the available signal flow before the system could generate enough trades for meaningful evaluation.
Broader NY-Session Scenario
The most useful finding of the weekly audit came from a broader scenario:
All Symbols NY Only Spread2.
This configuration produced 446 rows during the same weekly period.
The scenario used:
- all symbols available from the selector;
- NY-session signals only;
- a maximum spread of 2.0 pips.
The significantly higher throughput suggested that the previous funnel was filtering too aggressively rather than the selector itself failing to find market opportunities.
This became the basis for the next Scalping version.
FX Dynamic Scalping Selector V1.4A
A new paper-only version was created:
FX Dynamic Scalping Selector V1.4A Paper Funnel.
V1.4A is based on the previous V1.3.2 implementation but uses broader and more practical selection rules.
The current configuration:
- uses all symbols supplied by the selector;
- accepts only NY_OVERLAP and NY_LATE sessions;
- requires spread <= 2.0 pips;
- requires the direction to resolve clearly to BUY or SELL;
- removes the previous JPY-only restriction;
- removes the previous move_60m filter;
- removes the previous last-body filter.
The purpose of these changes is not simply to increase trade count at any cost. The objective is to restore enough signal flow for the paper funnel to produce a meaningful sample while retaining several basic quality controls.
Safety and Initial Validation
V1.4A was created strictly as a research version.
- Mode: PAPER_SIM_ONLY
- orders_sent: NO
- Live trading: disabled
The scheduled task was created successfully and the script completed its first run without error.
The status output confirmed that the system was operating normally and that no broker orders could be sent.
The initial run occurred during the weekend, so there were no fresh market signals and no new paper trades. This was expected and did not indicate a system problem.
V1.3.2 Archived
Once V1.4A had been created and validated, the previous V1.3.2 Paper Funnel was disabled.
V1.3.2 remains part of the Scalping research history because it provided an important result: its filtering structure demonstrated that excessive selectivity can reduce a potentially useful signal stream to a sample that is too small for efficient research.
The version therefore served its purpose even though it was not retained as the active paper funnel.
Final Active Research Setup
The active research environment is now intentionally small.
Only two scheduled research tasks remain active:
- FX Dynamic Scalping Selector V1.2.1 Scan;
- FX Dynamic Scalping Selector V1.4A Paper Funnel.
V1.2.1 continues to perform the wider market scan and direction analysis.
V1.4A receives the resulting signal flow and applies the broader NY-session paper-trading funnel.
This creates a much clearer research chain:
market scan → instrument and direction selection → V1.4A paper funnel → closed paper results
Archived or Disabled During Week 32
- AUDJPY Follow Driver Demo V0.2;
- AUDJPY Follow Driver Paper V0.2;
- AUDJPY Follow Driver Paper V0.3;
- FX Dynamic Scalping Selector V1.3.2 Paper Funnel.
Week 32 Decision
The main conclusion from Week 32 is that the previous research setup was both too fragmented and, in the case of Dynamic Scalping V1.3.2, too restrictive.
The AUDJPY Follow Driver line did not generate enough activity to justify continued active monitoring. Its demo branch also exposed a duplicate-order risk, which was corrected before the tasks were disabled.
At the same time, the Scalping audit showed that the underlying selector was producing a substantial number of signals, but the previous paper funnel was discarding too many of them.
The current research candidate is therefore FX Dynamic Scalping Selector V1.4A Paper Funnel.
It is not a live trading robot and it is not a demo trading robot. It remains a paper-only research version.
The immediate goal is to determine whether the broader NY-session configuration can produce enough trades, while maintaining enough positive structure, to justify development beyond the current paper stage.
Week 32 therefore ended not with a new trading result, but with a more important structural improvement: fewer active tasks, safer execution controls, and a clearer research path for the next observation period.
Current active candidate:
FX Dynamic Scalping Selector V1.4A Paper Funnel
Current active tasks:
FX Dynamic Scalping Selector V1.2.1 Scan
FX Dynamic Scalping Selector V1.4A Paper Funnel
Research mode: PAPER_SIM_ONLY
orders_sent: NO
Risk notice: Paper-simulation results and research observations do not demonstrate future profitability. The systems described in this report are experimental research versions and do not constitute financial or investment advice.
Open Scalping Research
Open Research Archive
← Back to Weekly Reports