Public Document · v1.00

Institutional
Algorithm KPI Debrief

A transparent breakdown of the standards, KPIs, and pipeline stages every algorithm passes through at Sentivue Capital. Built for developers — by traders who've sat on both sides of the screen.

View the Pipeline

We don't run a black-box evaluation process. This document is the same rubric our team uses internally — published openly so developers can build, test, and submit knowing exactly what we're looking for and exactly how every submission gets evaluated. Transparency is the standard.

01 / Who This Document Is For

Builders welcome

Audience & Posture

Sentivue Capital evaluates algorithmic trading systems built by independent developers, retail quants, and small teams across the global trading community. Whether you're a self-taught dev with two strategies or a small shop with a portfolio of systems, this document tells you exactly what we measure, why we measure it, and what to expect from our process.

We don't expect institutional-grade engineering on day one. We expect intellectual honesty, sound risk discipline, and a willingness to validate ideas under live conditions. The rest is teachable. Most of our best partnerships started with developers who were rough around the edges but rigorous about the hard questions.

02 / The Evaluation Pipeline

Five structured stages

Standard & Conditional Path

Every algorithm submitted to Sentivue Capital passes through a defined stage progression. Stage 2.5 is conditional — it only triggers when the data raises specific concerns we cover later in this document. Most algorithms with clean characteristics advance directly from Stage 2 to Stage 3.

Stage 01
Initial Evaluation
Source code review, logic audit, parameter forensics, baseline KPI assessment.
Standard
Stage 02
Optimization Pass
Iteration on identified gaps. Set file refinement under Sentivue risk profile.
Standard
Stage 02.5
Live Observation
Triggered only when concentration, regime, or randomness flags surface. 90-day forward.
Conditional
Stage 03
Formal Alignment
Direct conversation with leadership. Algo discussion, terms, paths forward.
Standard
Stage 04
Forward Test
Post-alignment forward execution period. Real conditions, no retuning.
Standard
Pipeline Philosophy
Forward time is the ultimate validator
No backtest, however clean, is ever sufficient evidence on its own for capital allocation. Stage 4 forward testing exists because backtest robustness is a necessary but not sufficient condition. The pipeline is structured around this principle — each stage exists to build progressive confidence rather than to make a single pass/fail decision.
03 / Submission Requirements

What to send us

Files & Format

Every submission is reviewed manually by a real analyst. We don't gate on file completeness alone — but the more transparent your submission, the deeper the analysis we can produce in return.

.mq5 Source Code
Enables full logic audit — bias detection, position sizing math, execution layer review per Sentivue engineering standards.
Preferred
.ex5 Compiled Binary
Acceptable for trial review when source code privacy is a concern. Limits depth of audit but enables backtest reproduction and live observation.
Trial Accepted
.set Parameter File
Required for accurate backtest reproduction. Send the exact set file that produced your shared backtest report.
Required
MT5 Backtest Report
HTML export from MT5 Strategy Tester. Minimum 1-year backtest period on tick-quality data preferred.
Required
Live Forward Test Data
6-12 months of demo or live forward execution if available. Significantly accelerates evaluation and reduces likelihood of Stage 2.5.
Strongly Preferred
Strategy Description
Plain-language explanation of the edge thesis. What pattern is the algo trying to capture? Why does it work?
Optional
Optimization Reports
Parameter sensitivity surfaces help us assess whether your settings are robust optima or fragile peaks.
Optional
⬢ Privacy-Friendly Trial Review
.ex5 binaries accepted for first-pass evaluation
Most evaluation programs require full source code on day one. We don't. Compiled binaries are accepted for trial-stage review, with the understanding that source code is required before Stage 3 alignment. This lets developers protect proprietary logic during initial conversations while we build mutual trust.
04 / The KPIs We Look For

Performance standards

Quantitative Bar

No algorithm passes or fails on a single number. We assess each submission holistically — but the following KPIs are the floor we measure against. Algorithms that fall significantly below these markers will be advised on specific upgrades before re-evaluation.

⬢ Risk Discipline · Highest Priority
Maximum Drawdown
The single most important number we evaluate. Drawdown is measured against starting capital under Sentivue's risk profile — meaning we re-test your algorithm with parameters tuned to our risk standards before scoring this metric. See the Tiered Capital Protection table below for full bands.
Tiered
See Below
⬢ Risk-Adjusted Return
Annualized Sharpe Ratio
Computed on monthly returns, properly annualized. 1.0 is our minimum bar, 1.5+ is preferred, 2.0+ is excellent. Above 3.0 with limited trade history will trigger additional scrutiny — exceptional Sharpe ratios on small samples are often artifacts.
≥ 1.0
1.5+ Preferred
⬢ Drawdown Recovery
Calmar Ratio
CAGR divided by maximum drawdown. Tells us how efficiently your algorithm generates return relative to the worst pain it produces. 1.0 minimum, 2.0+ preferred. Algorithms with Calmar above 5.0 require regime validation before we trust the headline.
≥ 1.0
2.0+ Preferred
⬢ Resilience Indicator
Recovery Factor
Net profit divided by max drawdown. Indicates whether the strategy can earn back from its worst periods. 3.0 minimum, 5.0+ preferred. Below 3.0 typically indicates a strategy that can't survive its own bad days.
≥ 3.0
5.0+ Preferred
⚠ Critical Diagnostic
Profit Distribution Across Time
This is where most algorithms fail. We evaluate whether profit accrues consistently across the test window or concentrates in a narrow recent period. Backtests showing the majority of profit clustered in a final quarter — while prior months are flat — will trigger Stage 2.5 Live Observation regardless of how strong the headline numbers look. Adaptability across regimes is the signal we're looking for.
Anti-Overfit
Forensic Check
⬢ Sample Quality
Backtest Period
1-year minimum backtest required. We strongly prefer submissions that include 6 months to 1 year of live forward test data alongside historical backtest results. Live data dramatically accelerates evaluation and often eliminates the need for Stage 2.5.
1y + 6-12mo
Backtest + Live
⬢ Statistical Health
Z-Score (Wald-Wolfowitz)
MT5 reports this natively. We pay attention to the absolute value. Z-scores between -1.5 and +1.5 indicate healthy outcome distributions. Scores beyond ±2.0 with high confidence indicate non-random clustering — typically a sign of regime-locked behavior that requires live observation to validate.
±1.5
Healthy Range
05 / Tiered Capital Protection

Drawdown standards

Three Bands · One Standard

Maximum drawdown is the metric we weight most heavily. Algorithms are evaluated under Sentivue-tuned parameters — meaning if your hardcoded set file produces 22% drawdown, we'll re-test under risk-adjusted parameters before applying these tiers. The bands below reflect post-adjustment performance.

Drawdown Tier System
Maximum Drawdown · Sentivue-Tuned Parameters
▸ Tier 1 · Qualifies
≤ 10%
Clean institutional risk profile. Algorithms in this band advance through the pipeline with the smoothest review process. Most professionally engineered systems land here once parameters are properly tuned.
▸ Tier 2 · Conditional
10–15%
Acceptable with conditions. Typically requires Stage 2.5 Live Observation regardless of other metrics. Position sizing recommendations may be issued. Algorithm enters the pipeline but with reduced initial sizing if approved.
▸ Tier 3 · Refinement Required
15%+
Outside our capital protection profile. Will not advance in current form. We'll provide specific guidance on parameter adjustments or risk architecture changes needed to bring the algorithm into Tier 1 or 2.
⚠ Important Context
Tier assignment uses Sentivue-tuned parameters, not your originals
Many submissions arrive with aggressive parameter sets that maximize backtest return at the cost of higher drawdown. This is a legitimate developer choice, not a flaw. However, Sentivue evaluates every algorithm as if it were trading our capital under our risk discipline — meaning we'll re-test your strategy under conservative parameters before applying tier assignment. You'll receive both your optimized set file and our tuned version for comparison.
06 / Engineering Standards

Code quality

Mandatory & Preferred

We assess code quality during Stage 1 review. The list below distinguishes between mandatory standards (where violations are auto-rejecting) and preferred standards (where gaps are flagged but workable through Stage 2 iteration).

▸ Mandatory · Auto-Rejection If Missing
Required Engineering

These represent fundamental correctness requirements. Submissions failing these standards cannot proceed regardless of backtest performance.

  • ◆No lookahead bias. Signal logic must use shift = 1 or later. Bar-close gating required for intra-bar signal generation.
  • ◆No hidden martingale logic. Position sizing must not scale up after losses. Risk-per-trade must be bounded.
  • ◆Bounded worst-case exposure. Maximum aggregate position risk must be enforceable and quantifiable at any moment.
  • ◆No future-bar data leakage. Indicators that repaint historical values are disqualifying.
  • ◆Defensive break-even logic. Stop-loss modification logic must be asymmetric — only narrow stops, never widen.
▸ Preferred · Gaps Workable in Stage 2
Strongly Preferred Engineering

These represent professional engineering practices. Gaps don't auto-reject submissions but will appear as recommendations in your Stage 1 report.

  • ◆Per-instrument position sizing using SymbolInfoDouble(SYMBOL_TRADE_TICK_VALUE) and TickSize rather than hardcoded pip values.
  • ◆Execution-quality filters: spread guards, max deviation parameters, session filters, news/calendar gates.
  • ◆Equity-based emergency kill switch — flatten and disable trading on configurable equity drawdown threshold.
  • ◆Symbol-aware magic numbers for clean multi-instance deployment without ID collisions.
  • ◆Structured telemetry via PrintFormat for forensic analysis when live behavior diverges from backtest.
  • ◆Min-stop compliance via SYMBOL_TRADE_STOPS_LEVEL check before order submission.
07 / Stage 2.5 — Live Observation

When backtests aren't enough

Conditional · 90-Day Forward
⊙ Conditional Stage

Live Observation Period

When a backtest displays patterns that cannot be resolved through additional backtesting, we trigger a 90-day live observation period before advancing the algorithm. This isn't a punishment — it's the only honest way to distinguish real edge from recent-regime artifacts. Forward time is the irreplaceable input.

▸ Trigger 1
Profit Concentration
Backtest shows the majority of net profit clustered in a final-period window while prior months were flat or declining. Strong indicator of recent-regime curve fit.
▸ Trigger 2
Z-Score Beyond ±2.0
High-confidence non-random outcome clustering. Suggests regime-locked behavior that backtest data alone cannot validate or refute.
▸ Trigger 3
Tier 2 Drawdown
Maximum drawdown landing in the 10–15% conditional band. Live observation confirms whether the risk profile holds in real-time conditions.

What happens during Stage 2.5: the developer runs the algorithm under Sentivue-tuned parameters in either demo or live mode (developer's choice) for 90 consecutive days. No retuning during the window — that's what makes the data valid. Monthly informal check-ins with Sentivue's evaluation team. Clean post-period passes advance directly to Stage 3.

What we're looking for: profit accrual across multiple market regimes — not necessarily replication of backtest magnitudes. 15-30% annualized returns across diverse conditions is stronger institutional evidence than 80% concentrated in a single quarter. Adaptability beats magnitude.

08 / Ready to Submit?

Three simple next steps

Path to Pipeline
01
Self-Audit
Run your algorithm against the KPIs and engineering standards above. If you find gaps, address them before submitting — it'll dramatically accelerate your evaluation.
02
Package & Submit
Send your .mq5 or .ex5, set file, MT5 backtest report, and any forward test data. Include a brief description of your edge thesis.
03
Receive Your Debrief
Within 5–7 business days, you'll receive a full institutional analysis report just like the ones our team produces internally. Verdict, KPI breakdown, code review, and clear next steps.
A note from the team
We read every submission personally
No automated rejections. No black-box scoring. Every algorithm is reviewed by a real analyst who walks your code line-by-line, runs your backtest under our parameters, and writes you a personalized institutional report. Whether your algorithm advances or not, you walk away with insights you can use to make your next build better. That's the partnership we offer — even before we know if we're working together long-term.