From raw strategy to prop-firm certificate

A complete, real walkthrough of AlgoCrucible's six-stage validation funnel — every screenshot below is from one actual NinjaTrader 8 strategy (goADXMALinReg on MNQ) going from first optimization to a printable certification report.

The six stages

Setup — connect to NinjaTrader
Configure & Optimize
Walk-forward
Audit — the ROBUST verdict
Perturbation — the plateau test
Monte Carlo stress-test
Certification — your pass odds

Before you read the numbers: everything below is hypothetical, model-based simulation — not a prediction or guarantee. Backtested results have inherent limitations, past performance does not indicate future results, and trading futures carries a substantial risk of loss. This walkthrough shows how the tool interrogates a strategy; it is not a claim that any strategy makes money. Full guide · see the disclaimer in every report.

0/Setup — AlgoCrucible talks to your NinjaTrader

AlgoCrucible runs on your machine and drives NinjaTrader 8's own backtest engine through a small bridge add-on the installer deploys. That matters: every number you'll see is produced by NT8 itself, on your data, with commission and slippage modeled — not by a re-implementation that fills at prices NT8 wouldn't.

AlgoCrucible home screen with the six-stage pipeline and a green NinjaTrader connection check — **The setup check.** The green banner means the bridge is loaded and a Strategy Analyzer window is open — that's all the setup there is. The six cards above it are the funnel this strategy is about to run.

Strategies tab listing every compiled NinjaTrader strategy with parameters detected automatically — **Pick a strategy.** Every strategy compiled in your NinjaTrader appears here automatically, parameters detected by reflection. We'll follow `goADXMALinReg` through all six stages.

1/Configure & Optimize — is there any edge at all?

Each parameter can stay at its default or be searched across a range. Fewer free dials means less room to fool yourself — only free the settings you genuinely aren't sure about. Here four parameters get ranges; everything else stays fixed.

Parameter grid with four parameters set to Optimize with ranges and the rest on defaults — **Choosing which dials the optimizer may turn.** Target, Stop, HMA Period and Lin Reg Period get ranges. The search is Bayesian (Optuna) — it samples the space intelligently rather than stepping a brute-force grid.

Optimization running with live trial results streaming from the NinjaTrader engine — **The search, live.** Trials stream in from NinjaTrader's engine as each batch finishes — every row is a full backtest with its trade count, net and profit factor.

Best found panel showing the top candidate parameters and a table of runners-up — **The audition result.** Best found: 338 trades, profit factor 1.407. The app says it itself: *a good score here only means "worth validating."* An optimizer can always find something that looked great in the past — the next four stages try to break it.

2/Walk-forward — does the edge survive data it never saw?

The strategy is re-tuned on one slice of history, then tested on the following slice it had never seen — repeated across the whole data set. Only the unseen slices count toward the result. That's the honest number, and it is expected to be smaller than the optimizer's.

Walk-forward result with per-window log and the combined out-of-sample stats — **Held up out-of-sample.** Eight windows, pooled unseen-data result: 194 trades, $1,057 net, profit factor 1.227 — down from the optimizer's 1.407, exactly as an honest test should be, and still profitable.

Setting stability table showing each parameter's best value staying in a tight band across windows — **Do the "best" settings hold still?** If the winning values jump around window to window, the optimizer is chasing noise. All four parameters graded *stable* — the edge doesn't need a different strategy every month.

3/Audit — the deploy gate

The audit freezes one parameter set and grades it on two separate out-of-sample periods it never touched. The verdict takes the worst of the two windows — no cherry-picking. Verdicts are blunt: ROBUST or MIRAGE.

Audit verdict panel with a ROBUST badge and the candidates table — **ROBUST.** "This setting made money in BOTH unseen periods with consistent months. It earned the right to be deployed." Most strategies die here — and a failure here is the tool doing its job, catching a mirage before it costs real money.

Equity curve and underwater drawdown chart of the audited candidate — **The road, not just the destination.** The audited candidate's full-range equity curve and its underwater chart — every decline from a peak, how deep and how long. Worst dip: −$488.

First page of the printable Strategy Audit Report — **The paper trail.** The audit as a shareable, printable report: what was tested, on what data, with what settings, and the verdict — for a trading partner, a prop firm, or future you. *(Click to see the full page.)*

4/Perturbation — plateau or cliff edge?

A real edge sits on a plateau: nudge every setting slightly off its chosen value and it should still work. An edge that only works at one exact magic number is a cliff — and live markets push you off cliffs. AlgoCrucible backtests dozens of nearby parameter sets and re-grades each on held-out data.

Robustness cloud verdict: ROBUST, with 95% of neighboring parameter sets holding out-of-sample — **The neighborhood test.** 38 of 40 nearby parameter sets (95%) held the edge out-of-sample. "The edge held up all around the winning values — this isn't a lucky spike. Small drift in the parameters (or the market) shouldn't break it."

Per-parameter sensitivity table and the out-of-sample robustness map, all cells green — **Which dials are dangerous?** Each parameter swept ±10–20% off its value, every step re-graded on the held-out tail. An all-green row is drift-proof on that axis; a red cell is where the edge stops surviving unseen data. Here: green across the board.

5/Monte Carlo — what could luck have done?

A backtest is one ordering of history. Reshuffle the trade sequence thousands of times and you see the range of outcomes the same edge could have produced — including drawdowns worse than the single one your backtest happened to show. These are the numbers a prop firm's rules actually act on.

Monte Carlo equity-curve fan with the actual backtest as the bold line and a confidence table — **The fan.** 2,000 reshuffles of the trade sequence; the bold line is the actual backtest, the band is where luck could have taken it. The table turns it into planning numbers: the backtest shows $1,971 net / $538 drawdown, but at **95% confidence** you plan for as little as $459 net and a drawdown up to $1,057.

Final P&L distribution and max-drawdown distribution histograms across reshuffled trade orders — **Size for the distribution, not the anecdote.** Across all reorderings: where final P&L lands, and how deep the worst dip typically gets. Your account — and your nerves — should be sized for the 95th percentile, not for the single lucky drawdown one backtest showed.

6/Certification — odds against a real evaluation's rules

Now the question that started all of this: would this strategy pass a prop-firm evaluation? Enter the evaluation you're actually facing — profit target, daily loss limit, trailing drawdown, time limit — and AlgoCrucible steps thousands of simulated evaluations through those exact rules, trade by trade.

Certification configuration form with prop-firm evaluation rules: account size, profit target, daily loss, trailing drawdown — **Your firm's rules, not a generic score.** Load a preset, then edit anything — here a typical 50K one-step: $3,000 target, $1,100 daily loss, $2,000 end-of-day trailing drawdown, pay-monthly (no time limit), 3 contracts.

Certification result: 82.7% modeled probability of eventually passing, with bust rate, months to pass, and estimated fees — **The answer, with its costs.** Modeled on 180 days of history: eventually passes **82.7%** of the time, busts 17.3% along the way, typically ~3.3 months and ~$601 in evaluation fees before it resolves. Every bust came from the trailing drawdown — not the daily-loss limit. Know this *before* paying for an eval, not after.

Evaluation outcomes donut and the distribution of days-to-pass across simulations — **What a normal month looks like.** Outcomes across 5,000 simulated evaluations, and — when it passes — how long that takes (median 70 trading days). This sets expectations: what's normal, and how long is long enough to suspect something has changed.

First page of the printable Prop-Firm Certification Report — **The certificate.** Every stage's verdict on one printable page: walk-forward PASSED, audit ROBUST, perturbation ROBUST (38/40), the exact rules tested, and the modeled pass odds — with the methodology and its limits stated right on the document. This is what "validated" looks like when it's earned rather than claimed.

Put your own strategy through the crucible

The full pipeline — Optimize, Walk-forward and Audit — is free for 30 days on up to 3 strategies. Perturbation, Monte Carlo and Certification are in the 7-day Trader trial, no card required.

Download AlgoCrucible Installation guide →

Results shown are hypothetical, model-based estimates computed from limited historical data under simplifying assumptions — not a prediction or guarantee that any strategy will be profitable or that any evaluation will be passed. Evaluation profiles are generic models of common industry structures, not the official rules of any firm; AlgoCrucible is independent and not affiliated with, endorsed by, or sponsored by any proprietary trading firm or by NinjaTrader. Trading futures carries a substantial risk of loss and is not suitable for everyone. Not financial advice.