Audit Engine Specification v2.0
AI Trust Score Methodology
An open, reproducible, and deterministic mathematical framework to evaluate automated trading algorithms and technical indicators without promotional bias.
The retail algorithmic trading space is dominated by selective backtest cherry-picking, curve-fitting, and dangerous martingale doubling. The StriveAlgo AI Trust Score is engineered to reward mathematical edge, verifiable survival longevity, and strict capital preservation over high-risk marketing claims.
1. EA AI Trust Score Formula (0–100)
The EA Trust Score evaluates automated robots across 6 normalized, weighted sub-scores clamped between 0 and 100:
| Pillar | Weight | Mathematical Formula | Rationale |
|---|---|---|---|
| Live Verification | 25% | is_real_verified ? 100 : 30. If no external statement link: -20. | Third-party broker statements beat hypothetical backtests. |
| Risk Control | 25% | 100 - (max_drawdown_pct × 2.5). Martingale: -35. Grid: -15. | Survival and max drawdown matters more than gross gain. |
| Risk-Adjusted Return | 20% | Calmar proxy c = monthly_gain × 12 / max(DD, 1), score 100 × min(c, 6) / 6. | Rewards consistent yield per unit of historical risk. |
| Track Record Maturity | 15% | 100 × min(account_age_days, 730) / 730. | Two years (730 days) of continuous live forward telemetry = full marks. |
| Edge Quality | 10% | 100 × clamp((profit_factor - 1) / 1.0, 0, 1). Missing: 40. | A profit factor of 2.0 achieves 100 on edge quality. |
| Community Feedback | 5% | (rating / 5) × 100 × min(reviews, 50) / 50. | Weighted lightly to prevent gaming by astroturfed reviews. |
- Gain > 1,000%: Multiply score by
0.85(implausible return indicating extreme margin leverage). - Age < 180 days: Soft-cap with progressive compression
55 + 20 × (score / 100)(capped at 75). - Unverified account: Soft-cap with progressive compression
50 + 20 × (score / 100)(capped at 70).
2. Indicator AI Trust Score Formula (0–100)
The indicator framework audits code quality, non-repainting integrity, and transparency:
| Pillar | Weight | Mathematical Formula | Rationale |
|---|---|---|---|
| Signal Integrity | 30% | does_repaint ? 20 : 100. | Non-repainting on candle close is essential for systematic trading. |
| Code Quality | 25% | code_quality_score (0–100). | Automated syntax check, loop complexity, and runtime efficiency. |
| Transparency | 20% | is_open_source ? 100 : 45. +10 if source URL is provided. | Inspectable code ensures no hidden mathematical artifacts. |
| Community Adoption | 15% | 100 × min(log10(tv_boosts + 1) / log10(5001), 1). | Logarithmic scaling of TradingView boosts and favorites. |
| Usability & Alerts | 10% | Alert support (+40) + Multi-timeframe (+30) + Best timeframes (+30). | Practical utility for trader workstation workflows. |
If does_repaint = true, the final score is hard-capped at 60.0, preventing repainting indicators from ever achieving an A or B grade.
3. Deterministic Tie-Breaking Engine
To ensure absolute ranking reproducibility and prevent clustering, no two systems ever share the same rank. When two items tie on raw score, the tie-breaking cascade operates in strict order:
For EAs: Score (desc) → max_drawdown_pct (asc) → account_age_days (desc) → slug (asc)
For Indicators: Score (desc) → does_repaint (false first) → tv_boosts (desc) → slug (asc)
4. Versioning Policy & Auditing Protocol
The StriveAlgo AI Trust Score engine follows strict semantic versioning (SemVer) principles to guarantee audit trail integrity:
- Patch (v2.0.x): Editorial copy updates, documentation enhancements, or display adjustments. Zero score or ranking shifts.
- Minor (v2.x.0): Pillar weight recalibrations, penalty threshold tuning, or new validation filters. Causes controlled ranking shifts recorded in changelog.
- Major (v3.0.0): Fundamental scoring architecture redesign, addition/removal of evaluation pillars, or integration of new data providers.
Any modification to mathematical coefficients in src/lib/scoring.ts mandates an immediate version bump, automated regression tests, and publication in the audit changelog below.
5. Methodology Changelog
Active: v2.0| Version | Release Date | Summary & Changes | Impact |
|---|---|---|---|
| v2.0 Breaking | 2026-11-07 | Weighted continuous multi-pillar model with 1-decimal precision, soft-caps, and indicator code audit.
| Eliminated score clustering and ties; 100% deterministic ranking reproducibility across all catalog assets. |
| v1.0 | 2026-10-15 | Initial baseline integer trust score prototype.
| Baseline prototype; deprecated due to bias toward unhedged martingale risk profiles. |
Methodology FAQ
Common questions regarding our auditing math and verification rules.
Why does StriveAlgo penalize raw gain above 1,000%?
In retail algorithmic trading, gains exceeding 1,000% over short time horizons are almost universally achieved through excessive leverage (1:500+), grid recovery stacking, or martingale lot sizing without stop losses. A 15% haircut is applied to discourage unsustainable high-variance gambling strategies.
How do you test TradingView indicators for repainting?
We evaluate Pine Script code for security function lookahead bias (barmerge.lookahead_on on open bars), future-offset indexing, and barstate.isconfirmed event handling. If an indicator redraws or relocates past signals once a candle closes, it is flagged as repainting and capped at an AI Trust rating of 60.
Can authors pay to increase their AI Trust Score?
No. The AI Trust Score is 100% computed deterministically at site build time from empirical data (verified statements, audited code, and repository logs). No manual overrides or sponsored ranking adjustments exist.
What is the Calmar ratio proxy used in EA risk-adjusted returns?
We compute an annualized return-to-max-drawdown ratio: Calmar = (Monthly Gain % × 12) / max(Max Drawdown %, 1.0). A Calmar ratio of 6.0 or higher achieves full marks (100) on the Risk-Adjusted Return pillar.
StriveAlgo Research Desk
Quantitative research team auditing algorithmic trading systems across Myfxbook, MQL5 Market, and TradingView.