| Audited system | martingale |
| Engine version | 1.0.0 |
| Method version | 1.0 |
| Seed | 0 |
| SHA-256 fingerprint of the data | ae4badd1047e27534d61db5914dd8b7232470df67bdc4663c85783394ca19ef5 |
| Monte Carlo iterations | 10 000 |
| Number of trials used (N) | 10 |
| Period covered | 2021-01-05T06:21:46 → 2022-04-03T23:51:20 |
| Usable / supplied trades | 600 / 600 |
| Currency | USD |
| Initial balance | 10 000.00 |
| Service | Audit complet (audit_complet) |
Verdict: NO-GO.
Findings: 1 blocking, 1 major, 0 minor, 6 informational.
A barrier value falls inside its indecision band: human review required (gate E3).
Red light: 1 blocking flaw(s) were found, 1 test(s) could not be run for lack of data, 3 test(s) sit in their uncertainty zone and 1 major flaw(s) were found. The results provided do not allow the claim that this system has a real edge: what was earned can be explained by chance, by a few exceptional trades or by uncontrolled risk. This is not a forecast of future losses; it is the absence of proof.
The verdict covers only what was given to the auditor: the closed trades in the file, over the period covered. It rests on no human judgement and no overall score: every test has a threshold fixed in advance by the method, and the verdict follows mechanically from those thresholds. The same file, audited again tomorrow with the same version, yields exactly the same report.
The five questions put to the results, and the numeric answer for this file:
When several versions of a strategy are tried before one is kept, the best one always looks good, even if none has an edge. This test computes the probability that the performance is real once that effect is corrected, for 10 declared variant(s): 54.0 %. Pass threshold: 50.0 %. Result in the uncertainty zone around the threshold: the method then calls for a human review rather than an automatic decision.
The profit factor is the total won divided by the total lost (1.00 = break-even). The same trades were replayed 10 000 times in a different order, in blocks, to see what the same strategy would yield with less luck: in the 5 % least favourable scenarios the profit factor falls to 1.027. Threshold: 1.00 — below it, the strategy loses money in those scenarios. Result in the uncertainty zone around the threshold: the method then calls for a human review rather than an automatic decision.
The drawdown is the largest fall between a peak and the trough that follows it. The observed one (1 155.00) is larger than in 88.0 % of random reorderings of the same trades. A very low value (under 5 %) would signal a curve too smooth to be honest; a very high value means the losses came clustered, which is harder to live through than the final result suggests. Test passed.
10 000 systems with no edge at all, but winning and losing amounts of the same size as this one, were simulated; 4.8 % of them did as well or better. Threshold: 5.0 % — above it, this system cannot be told apart from a system with no edge. Result in the uncertainty zone around the threshold: the method then calls for a human review rather than an automatic decision.
A strategy tuned on one period tends to shine there and disappoint afterwards. This test compares its performance on periods not used for tuning. It could not be run: it requires one set of results per parameter set tested, which was not provided. Not measurable with the data provided.
Martingale: after a loss, volume is multiplied by ≥ 1.5 in 96% of cases (after a win: 1%), over 135 post-loss trades. In plain words: after a loss, position size is increased to “win it back”. This yields long runs of small gains and then a very heavy loss on the day the losing streak lasts too long. The final balance does not show it until that day comes.
No stop: 100% of trades have no stop-loss. In plain words: part of the positions had no automatic exit level in case of loss. The possible loss on those positions was therefore not bounded in advance.
A red light is not fixed in the report: it is fixed in the strategy or in the data, then verified on a new history. The points above say exactly which test failed and by how much, so that the discussion rests on facts.
This report states facts; it recommends no settings and predicts nothing.
The verdict follows mechanically from the barriers below; no composite score.
| Barrier | Value | NO-GO threshold | Indecision band | Status |
|---|---|---|---|---|
| Deflated Sharpe ratio (DSR) | 0.540 | < 0.500 | [0.450 ; 0.550] | indecision band |
| Profit factor, 5th percentile of the block bootstrap | 1.027 | < 1.000 | [0.950 ; 1.050] | indecision band |
| Percentile of the observed drawdown among order permutations | 87.970 | < 5.000 | [3.000 ; 8.000] | passed |
| p-value against the no-edge null | 0.048 | ≥ 0.050 | [0.030 ; 0.080] | indecision band |
| Out-of-sample walk-forward efficiency | n/a | < 0.400 | [0.350 ; 0.500] | not evaluated |
Martingale: after a loss, volume is multiplied by ≥ 1.5 in 96% of cases (after a win: 1%), over 135 post-loss trades.
No stop: 100% of trades have no stop-loss.
Martingale: after a loss, volume is multiplied by ≥ 1.5 in 96% of cases (after a win: 1%), over 135 post-loss trades.
No stop: 100% of trades have no stop-loss.
Variance of Sharpe ratios across variants unknown: Lo approximation (V = 0.0017) for N = 10 trials. Conservative correction: the true DSR is probably lower.
Detector “execution_realism” not run: no price series provided.
Detector “lookahead” not run: no price series provided.
Fields absent from the source: commission, swap. Analyses that depend on them are disabled and reported as such.
The “no-edge” null keeps the empirical distributions of gains and losses and only neutralises their frequency (p₀ = 0.7321). It answers: “is the observed hit rate explainable by luck, given this gain/loss profile?” — it is not an efficient-market null.
Detector “stop_geometry” not run: no price series provided.
PSR: probability that the true Sharpe is positive, corrected for skewness and fat tails. DSR: the same probability evaluated against the best Sharpe expected by chance after N trials.
| Probabilistic Sharpe ratio (PSR) | 0.9315 |
| Deflated Sharpe ratio (DSR) | 0.5401 |
| Expected maximum Sharpe under the null (SR₀) | 0.0644 |
| Number of trials used (N) | 10 |
| Variance of Sharpe ratios across variants (V) | 0.001671 |
| Source of V | Lo approximation (conservative: the true DSR is probably lower) |
This system stops being significant (DSR < 0.50) beyond 12 tested variants.
| N trials | DSR |
|---|---|
| 1 | 0.9315 |
| 10 | 0.5401 |
| 50 | 0.3026 |
| 100 | 0.2294 |
| 300 | 0.1441 |
| 1 000 | 0.0840 |
Three simulations, three questions: the bootstrap measures how unlucky the same edge can get; the permutation tests whether performance owes to trade order; the null counts edge-less systems that would do as well.
The “no-edge” null keeps the empirical distributions of gains and losses and only neutralises their frequency (p₀ = 0.7321). It answers: “is the observed hit rate explainable by luck, given this gain/loss profile?” — it is not an efficient-market null.
| Bootstrap method | stationary blocks (Politis-Romano), automatic length |
| Mean block length | 15.21 |
| Profit factor p05 / p50 / p95 | 1.027 / 1.277 / 1.593 |
| Net profit p05 / p50 / p95 | 280.00 / 2 145.00 / 3 535.00 |
| Drawdown p50 / p95 / p99 | 805.00 / 1 625.00 / 2 140.00 |
| Observed drawdown | 1 155.00 |
| Percentile of the observed drawdown among permutations | 88.0 |
| Permutation drawdown p50 / p95 | 875.00 / 1 300.00 |
| Equilibrium hit rate p₀ | 73.2 % |
| p-value (net profit ≥ observed) | 0.0479 |
| Net profit of null systems p50 / p95 | 10.00 / 2 025.00 |
Classifier used: volatility terciles (rolling standard deviation of P&L, no prices); unclassified trades: 2.
| Regime | Trades | Net profit | PF | Hit rate | Share of net |
|---|---|---|---|---|---|
| low volatility | 218 | 975.00 | 1.508 | 80.3 % | 47.6 % |
| medium volatility | 188 | 575.00 | 1.236 | 75.5 % | 28.0 % |
| high volatility | 192 | 470.00 | 1.134 | 76.0 % | 22.9 % |
| Quarter | Trades | Net profit | PF |
|---|---|---|---|
| 2021-Q1 | 118 | 895.00 | 1.973 |
| 2021-Q2 | 120 | 690.00 | 1.575 |
| 2021-Q3 | 116 | 720.00 | 1.667 |
| 2021-Q4 | 128 | -585.00 | 0.805 |
| 2022-Q1 | 114 | 320.00 | 1.200 |
| 2022-Q2 | 4 | 10.00 | 1.125 |
Statistical drawdown markers from the block bootstrap, to compare with the live drawdown. They are not strategy settings.
| Observed maximum drawdown | 1 155.00 |
| Bootstrap drawdown, 95th percentile | 1 625.00 |
| Bootstrap drawdown, 99th percentile | 2 140.00 |
qaudit 1.0.0 — method 1.0 — schema 1.0 — seed 0 — fingerprint ae4badd1047e27534d61db5914dd8b7232470df67bdc4663c85783394ca19ef5 — service audit_complet.
Reproducible: same inputs + same version + same seed = same report, byte for byte.