SRA QUANT

Methodology: how the scoring works

Last updated: July 2026 · Plain-language version of the internal methodology document

Nothing about SRA's analysis is a black box. This page explains exactly how a score is produced, what it means, how it was calibrated, and where its limits are.

1. The pipeline

Every 15 minutes the engine pulls fresh candles for each watched asset (1H, 4H and Daily), computes indicators, detects chart patterns, marks swing support/resistance, and builds a volatility-based setup: invalidation at 1.5× ATR, first target at 3× ATR. A deterministic scorer then counts how many independent conditions align. Before any alert is sent, a second, AI review stage reads the actual headlines and macro calendar attached to the setup and can veto it — for example, a central bank decision inside 24 hours, breaking regulatory news, or an exchange hack.

2. The confluence categories

Points are only awarded when an input agrees with the proposed trade direction.

CategoryMax pointsWhat earns points
Trend alignment20EMAs stacked on 4H (8) and Daily (7); 4H and Daily agreeing (5). Zeroed if the 4H trend contradicts the trade direction.
Key level confluence20Price within 0.3% of a swing support/resistance (12); a detected pattern at that level (+5).
Momentum20RSI at a directional extreme (9), MACD histogram agreeing (7), stochastic extreme (4).
Volume15Volume ≥1.5× its 20-period average (15), ≥1.2× (10), ≥1.0× (5).
Risk/Reward15R:R ≥3.0 (15) down to ≥1.5 (3). Below 1.5 the setup is automatically disqualified, regardless of everything else.
News & macro penalty−20High-impact calendar event inside 24h (−15); ≥3 negative headlines (−8), ≥1 (−3).
Multi-timeframe bonus+101H, 4H and Daily all pointing the same non-neutral direction.

3. Score calibration — what "90+" really is

The raw category points are calibrated before display: displayed = round(raw × 90/76), capped at 100. So a displayed "90+" corresponds to a raw score of 76.

Why: a 75,292-evaluation historical replay (7 crypto majors, 2021–2026, trading costs and pessimistic fills included) showed the raw scale is structurally capped near 93 and never reached 90 — the volatility-based setup fixes R:R near 1:2, capping that category at 8 of 15. The raw ≥76 cutoff was then validated by walk-forward testing (18-month train, 6-month test, 7 rolling windows) before being mapped onto the published 90 threshold. "90+" therefore refers to a level with measured historical edge — not an unreachable number.

4. The honest numbers

At that validated cutoff, the historical baseline produced 289 signals with a 42.2% win rate and an average expectancy of +0.257R per signal. Out-of-sample (walk-forward), expectancy was +0.247R per signal, positive in 6 of 7 windows.

Read that carefully: the system loses more often than it wins. The measured edge comes from expectancy — winners are roughly twice the size of losers — not from a high hit rate. A score of 92 means 92 of 100 conditions aligned; it is not a 92% probability of winning.

5. How outcomes are judged (triple-barrier)

Every signal — sent or silent — is logged and later resolved against the candles that printed after it, using the same rules live as in backtests:

Keeping the live convention identical to the research convention is what makes live-vs-backtest drift measurable.

6. Limitations — read honestly

SRA Quant provides market analysis and educational content only. Nothing on this page or the platform constitutes financial advice, and past or backtested performance does not guarantee future outcomes. Trading involves substantial risk of loss.