Risk & Scoring

How much confidence do the data actually support?

The scores are, by construction, conservative estimates of executable performance. Every calculation records the parameter version it used.

Individual Score (0–100)

Observed quality × evidence maturity: loss severity, stagnation, regularity and per-regime consistency (Bull / Bear / Range), weighted by the available evidence. A new strategy with excellent results but a short sample deliberately receives a contained grade.

Portfolio Score (0–100)

The portfolio is treated as a single synthetic strategy over the synchronized window of its legs. The grade of the aggregate curve is adjusted by a Composition Factor that penalizes currency and symbol concentration, directional imbalance, regime fragility and dependence on a few leaders.

Haircut: the reality adjustment

Before any scoring, every demo trade is discounted by the median demo-to-real degradation measured on a real-money sentinel account, per symbol and direction. It is applied in full only with a sufficient paired sample, and never produces a positive adjustment.

Formal definitions live in a dedicated normative document.

Scores, illustrated

What each score rewards, and what it discounts

Individual Score: evidence maturity

A great short record ranks below a good long one.

Strategy A

Excellent results, 2 months of evidence

Observed quality
90
Evidence maturity
2 months
Individual Score
41

A contained grade until the evidence matures.

Strategy B

Good results, 12 months of evidence

Observed quality
75
Evidence maturity
12 months
Individual Score
71

Ranks higher: the record is long enough to trust.

Illustrative values, not actual results or the exact formula.

Observed quality × evidence maturity: the score discounts what the system does not yet know.

Population + decorrelation

Illustrative animation, not actual results.
Transcript
  1. Population: sixteen strategy legs, each a noisy equity curve.
  2. One leg alone has deep drawdowns and long flat stretches.
  3. Correlation is measured pair by pair. Each row of the grid is one leg; the bright blocks are groups of legs that move together.
  4. From each block only one leg is kept, together with every independent leg; the rest are redundant.
  5. Many uncorrelated legs, one steadier curve: their average has a lower drawdown and less stagnation than a single leg.
  6. Diversification that is measured, not assumed.

Portfolio Score: composition

The Composition Factor penalizes currency and symbol concentration, directional imbalance, regime fragility and dependence on a few leaders.

Same curve, different score.

Portfolio A

Spread across currencies, directions and leaders

Base Grade
80
Composition Factor
× 0.97
Portfolio Score
78

A balanced composition keeps almost all of its grade.

Portfolio B

Concentrated in one currency and a few leaders

Base Grade
80
Composition Factor
× 0.70
Portfolio Score
56
  • Currency concentration
  • Dependence on a few leaders

The same curve scores lower: its risk is concentrated.

Illustrative values, not actual results.

Base Grade × Composition Factor: the aggregate curve is graded first, then discounted for how its risk is concentrated.

Reality adjustment

The haircut, step by step

Before any score is computed, demo results are discounted by how much real money has shown the demo to overstate.

  1. Pair the trades

    A sentinel account trades a sample of strategies with real money, alongside their demo copies.

  2. Measure the gap

    The median demo-to-real degradation is computed per symbol and per direction, Buy and Sell separately.

  3. Discount the demo

    Every demo trade of that symbol and direction is reduced by the measured gap before scoring.

Illustrative example · EURUSD Buy

Demo result100
After the haircut82

−18% median demo-to-real gap

Indexed to 100. Illustrative values, not actual results.

  • Per symbol and direction

    EURUSD Buy and EURUSD Sell each carry their own haircut.

  • In full only with enough evidence

    The haircut is applied in full only when there are enough paired trades.

  • Never a bonus

    If real trading happened to beat the demo, the adjustment is zero, never positive.

The haircut turns every score into a conservative estimate of what real execution would have produced.

Controls

Risk controls at every phase

Each phase of the pipeline has controls that can stop, discount or replace a strategy. These are the ones in production today.

  • BlocksStops a strategy or portfolio from advancing
  • AdjustsLowers the numbers before any decision
  • TriggersStarts an action when behavior changes
  • ProtectsLimits the damage if something fails
  1. Simulation

    Generate & filter

    • Blocks

      Robustness filters

      Profit factor, stagnation, return/drawdown and minimum-activity filters discard weak candidates.

    • Blocks

      Perturbation tests

      Candidates must survive other time zones and timeframes, and Monte Carlo on outliers, spread and slippage.

  2. Live demo

    Measure at the broker

    • Blocks

      Months of live demo

      No strategy becomes eligible before an extended period of real broker execution.

    • Adjusts

      Mandatory haircut

      Every demo trade is discounted by the measured demo-to-real gap before scoring.

    • Adjusts

      Evidence maturity

      Short track records receive deliberately contained grades.

    • Blocks

      Individual Score threshold

      Only strategies with a sufficient Individual Score reach portfolio construction.

  3. Scores

    Compose

    • Blocks

      Correlation checks

      Portfolios are built from low-correlated strategies, with correlation verified at three levels.

    • Blocks

      Redundancy filter

      Portfolios that duplicate others are discarded.

    • Adjusts

      Composition Factor

      Penalizes currency and symbol concentration, directional imbalance, regime fragility and dependence on a few leaders.

    • Blocks

      Concentration limits

      Limits per symbol and per currency, with long/short balance.

  4. Real capital

    Operate

    • Protects

      Conservative sizing

      Positions on the real account are sized conservatively.

    • Triggers

      Real vs. expected monitoring

      When live behavior diverges from expectations, the portfolio is rotated out.

    • Protects

      Broker redundancy

      Every primary account has a mirrored backup at a second, independent broker.

    • Protects

      Infrastructure continuity

      Automated backups and a health dashboard with alerts.

Controls by phase. A strategy reaches real capital only after every blocking control has passed, and stays there only while monitoring agrees.

Risk management

Acknowledged limitations, structural mitigation

What we explicitly assume

  • Demo does not perfectly replicate real liquidity.
  • Slippage and execution may degrade with more capital.
  • Correlations change under stress.
  • No strategy or portfolio is assumed to be profitable indefinitely.

How risk is mitigated by construction

  • A mandatory haircut calibrated by the sentinel account.
  • Extreme diversification across strategies and symbols, with long/short balance.
  • Concentration limits per symbol and currency, plus directional decomposition.
  • Conservative sizing and continuous monitoring with portfolio rotation.
  • Broker redundancy and automated infrastructure backups.

Talk to the partners

We speak with qualified investors, institutional partners and brokers interested in our research and execution process.