SPEC-EAV-001 · v0.2
ENTRION EA Verification Standard
Twenty-one items across three layers. Judged from the output a trading platform already produces — no source code, no parameters, no strategy description.
Three layers
L2 Risk management | Position sizing, leverage limits, stop and exit conditions, capital allocation, maximum loss limits. The core of the judgment. Defects in this layer leave traces in the output regardless of the sample. |
|---|---|
L3 Execution and verification | Order transmission, fill confirmation, slippage and latency handling, backtest methodology, live monitoring. Judged. Detects the structural causes of divergence between backtest and live trading. |
L1 Entry logic | Rules that open a position when price, indicator or pattern conditions are met. Not judged. The merit of this layer depends on the sample and cannot be judged structurally. Reference figures only. |
Judgment items
L2 Risk management
K-01 | Position sizing independence K-01 tests whether position size is calculated from account equity or fixed inside the entry logic. It is judged from the trade record alone: a regression of trade volume on the preceding balance, read together with the coefficient of variation of volume. |
|---|---|
K-02 | Per-trade risk consistency K-02 tests whether the amount risked per trade follows a rule. It is judged from the distribution of (entry price minus stop price) times volume divided by prior balance; where no stop price is recorded, realised loss divided by prior balance is used instead. |
K-03 | Drawdown circuit breaker K-03 tests whether the system reduces or halts trading while losses accumulate. Drawdown intervals are identified on the equity curve, and trade frequency and average volume inside those intervals are compared with the same figures outside them. |
K-04 | Absence of stop loss K-04 tests whether a loss ceiling is actually observed in the record. The tail thickness and maximum of the losing-trade distribution are examined to distinguish losses that cluster near a bound from losses that spread without one. |
K-05 | Position increase after loss (martingale) K-05 tests whether the system increases position size following losing trades. It is judged from the correlation between the sign of the previous trade's result and the change in the following trade's volume, together with the rate of volume increase across consecutive losing sequences. |
K-06 | Correlation of concurrent positions K-06 tests whether positions held at the same time are genuinely diversified. Overlapping positions are grouped, and the directional agreement of their results is measured alongside their combined exposure. |
K-07 | Concentration of profit and loss K-07 tests how far the reported result depends on a small number of trades. The share of total loss held by the five largest losing trades, and the share of total profit held by the five largest winning trades, are computed. |
K-08 | Response to volatility K-08 tests whether the system reduces exposure when volatility rises. Each trade is assigned to a volatility regime and differences in volume and frequency across regimes are tested. |
K-09 | Effective leverage K-09 measures leverage as it was actually carried, not as an average. The time series of total concurrent notional exposure divided by equity is constructed, and its maximum and upper percentiles are reported. |
L3 Execution and verification
E-01 | Walk-forward degradation E-01 measures how far performance falls between the fitting window and the validation window. The ratio of the two is computed, and the trend of degradation across successive windows is read alongside it. |
|---|---|
E-02 | Backtest to live fill price divergence E-02 measures the difference between the price a backtest assumed and the price a live account received for the same signal. The distribution of that difference is examined and tested for directional bias. |
E-03 | Slippage assumption verification E-03 compares the slippage and spread assumed in a backtest against what was actually observed in live trading, separated by session and by volatility regime rather than aggregated into a single figure. |
E-04 | Inclusion of trading costs E-04 tests whether commission and swap were included in the reported result. The presence and distribution of those fields are examined to determine whether the equity curve was drawn from gross or net profit. |
E-05 | Duplicate orders and re-entry anomalies E-05 detects duplicate fills at the same timestamp or at extremely short intervals, and patterns of immediate re-entry after an exit. |
E-06 | Traces of look-ahead bias E-06 tests whether the recorded accuracy is attainable. Entry timing is examined against the price movement immediately following it, and the combination of win rate and payoff ratio is tested against the statistically possible range. |
E-07 | Extreme events in the sample period E-07 records how many volatility extremes the test period contained and how long they lasted, and compares them against known periods of market stress. This item states the valid range of the verification rather than passing or failing it. |
E-08 | Stability of trade frequency E-08 detects trend and structural break points in the number of trades per unit period. |
E-09 | Signal to order to fill attrition E-09 measures the proportion of signals that become orders and orders that become fills, together with the time taken at each stage, separated by session and volatility regime. It requires the optional extended input; without it the item is reported as not judgeable. |
L1 Entry logic
L-01 | Stability of signal frequency L-01 reports the stability of signal generation over time. It is presented, not judged, so that the reader can determine whether the logic operates only within a particular regime. |
|---|---|
L-02 | Performance dispersion by regime L-02 classifies trending and ranging intervals and presents the distribution of results within each. It is presented, not judged. |
L-03 | Parameter sensitivity L-03 shows whether performance collapses at neighbouring parameter values. It is produced only where a parameter set has been supplied, and is presented, not judged. |
Design principles
- P-01We do not judge profitability.What is judged is structure, not result. Judging profitability creates an implicit promise about future performance, blurs the boundary with unregistered investment advice, and produces a sample-dependent judgment with no reproducibility.
- P-02We do not ask for source code.An EA's source code is the developer's entire asset. A tool that demands it is not adopted. Structural defects leave traces in the output — the trade record and the equity curve — so judgment is possible without code.
- P-03Local execution. Data never leaves the device.Input data is not transmitted to a server. Judgment runs on the user's machine and only the result is displayed. This resolves the trust problem in P-02 structurally rather than by assurance.
- P-04Every judgment is published with its evidence and its threshold.A tool that presents a verdict while concealing its basis asks to be trusted in itself. That is the same structure this project exists to criticise. The reader must be able to re-examine the judgment.
- P-05Insufficient data is reported as such, never as a pass.Items lacking data are not filled by estimation. A pass and a judgment that could not be made are different results, and a tool that does not distinguish them manufactures safety that does not exist.
Judgment states
| Pass | The structure is observed and meets the threshold. |
|---|---|
| Caution | The structure is observed but sits at the boundary of the threshold. Evidence figures are presented alongside. |
| Defect | The structure is not observed, or clearly falls outside the threshold. |
| Not judgeable | The required input data is absent. Marked distinctly from a pass. |
What is not judged
- Profitability or future performance. Neither judged nor predicted.
- Direction, instrument, or entry timing. No opinion is offered.
- The merit of entry logic. Reference figures only, never a verdict.
- Suitability for managing third-party money. This standard is not used for that determination.
Input specification
Required fields
| Entry time / exit time | Concurrency, time-based exits, change in trade frequency over time |
|---|---|
| Direction (buy / sell) | Directional correlation of concurrent positions |
| Volume (lots) | The central variable for position sizing independence |
| Entry price / exit price | Realised risk per trade, tail of the loss distribution |
| P/L and balance after trade | Equity curve reconstruction, drawdown intervals, sizing-to-balance test |
Recommended
| Commission / swap | Verification that costs were applied. Detection of cost-free backtests |
|---|
Optional
| Stop price / take-profit price | Direct computation of per-trade risk. Otherwise estimated from the loss distribution |
|---|---|
| Parameter set / walk-forward window results | Activates L-03 sensitivity and E-01 degradation |
| Audit-trail event log (open standard format) | Activates E-09; converts K-03, K-06 and K-09 from estimation to observation |
Supported formats
| MetaTrader 5 Strategy Tester report (HTML / XML) | First-class support. Parser written in-house. |
|---|---|
| MetaTrader 5 account history (Deals / Orders export) | Used for live comparison. Required input for L3 judgment. |
| MetaTrader 4 report (HTML) | Second phase. The installed base remains large. |
| Generic CSV | A table whose columns carry the required fields. Minimum common input, platform-independent. |
Citation
Refer to items by code and source.
ENTRION EA Verification Standard, item K-05 (SPEC-EAV-001 v0.2)