Trading intelligence

Two-sided decisions, one shared market view, evidence-gated learning.

This page describes what the RSForex platform actually does on every evaluation, how customers share the same market decision while executing it individually, and how the system improves without ever modifying itself unchecked.

Every evaluation

What is scored, for BUY and for SELL, on each pass.

The engine reads the live tape from your terminal, classifies the regime, looks for validated setup families and then asks the same question of both sides: given this session, this regime and this setup, what does the evidence say?

Selection is by utility: the expected value per trade shrunk toward its pessimistic bound by how much history backs the context. Thin evidence is treated cautiously; strong evidence earns the trade. WAIT is chosen whenever neither side qualifies, and every WAIT carries its reason.

Numbers such as the exact edge threshold, session rules and setup parameters are private. The concepts and the recorded fields are public so you can understand and audit what the system did.
P(success)
Estimated probability that a BUY (or SELL) in this context wins, from the account's realised history and the fleet's evidence
Expected value and pessimistic bound
Average win and loss in the context, combined with P(success); the pessimistic bound uses a lower-confidence probability
Expected MFE / MAE
How far comparable trades typically ran in favour and against, from live outcomes
Confidence and uncertainty
How much history backs the estimate; thin samples widen the interval and lower utility
Regime and setup family
Trend, range, chop, extremes; setups such as range-extreme fades and trend pullbacks
Execution conditions
Spread, session, market state, broker stop levels; a bad tape blocks both sides
Opportunity quality
Structure, confirmation and location within the day's range
Result
BUY, SELL or WAIT, plus the rejected side and why, written to the decision record

Shared decision, individual execution

One MARKET_DECISION_ID. One ACCOUNT_EXECUTION_ID per customer.

Market decisionA qualified opportunity is one shared decision: symbol, side, setup, regime, session, reference price.
Fan-outEvery eligible account evaluates that decision independently.
EligibilityTrading Ready, correct MT5 account, EA and connection alive, access active, no risk stop.
Account riskBalance, equity, daily-loss room, maximum-drawdown room, prop-firm rules, broker volume and stop constraints.
Execution or blockBUY 0.20 lots on one account, BUY 0.05 on another, blocked by risk on a third, offline on a fourth. Each outcome and reason is recorded.

Because the market decision is one object, learning treats it as one prediction. The many account executions of the same decision are execution observations: they teach the system about spread, slippage, latency, broker behaviour, fill quality, sizing and management, not about whether the market call was right.

Customers therefore never compete for a signal and never see anyone else's data; they receive the same intelligence and execute it within their own limits.

Fleet learning

What is collected, how it is protected, and what it teaches.

Market evidence

Regime, setup family, both direction scores, price, spread, timing and the reason for every WAIT. One row per market decision.

Execution evidence

Fill price versus proposed, slippage, latency, size, account constraints and blocks. One row per account execution.

Outcome evidence

MFE, MAE, capture ratio, giveback, exit reason, management actions and the post-trade autopsy.

Identity
Internal account identifiers and hashed logins only. Names, emails, passwords and payment details are never part of training data.
Isolation
A customer sees only their own account; aggregate evidence is shown without account attribution.
Data quality
Freshness, completeness, timestamp order, duplicate detection and lifecycle completeness are scored; low-quality rows are excluded from training.
Where it runs
On the RSForex server. No customer PC is used for training and no paid third-party AI service is required.
Your terminal sends what the executor needs to trade safely and what the platform needs to measure the trade. It does not send your broker credentials, your other symbols or anything outside the gold position it manages. The privacy policy describes this in full.

Self-learning, governed

Observe, decide, execute, measure, learn, validate, improve.

ObserveLive decisions, executions and outcomes accumulate as time-ordered evidence.
LearnNightly: outcomes labelled, data quality scored, candidate models trained on the pool.
ValidateWalk-forward blocks and an untouched holdout; no future data leaks into training.
ShadowCandidates score live decisions in parallel without any authority.
CanaryA limited live stage with its own evidence requirements.
ChampionPromotion only through evidence gates; a known-good champion is always kept.
MonitorDrift in outcomes, predictions, execution and data quality is watched continuously.
RollbackDegradation restores the previous champion automatically and records why.

Machine learning may improve direction selection, setup ranking, entry and management quality and opportunity thresholds. It has no authority over authentication, customer isolation, licensing, maximum-loss protections, prop-firm hard rules, drawdown controls or security. Those remain deterministic governors that no model can weaken.

Today production runs the deterministic Tier-0 intelligence; candidate models are trained only once enough labelled live executions exist and stay in research or shadow until the gates are met.

Production evidence, last 30 days

Market decisionsunavailable
Executionsunavailable
BUY / SELLunavailable
Live outcomesunavailable
Open the Evidence page

See it run on your own account.

Six months complimentary. Setup takes minutes; verification is automatic.