Methodology

1. Signal states

Every signal has a lifecycle: DRAFT → REVIEW → LIVE, then a terminal state:

• HIT_TARGET — last traded price reached the target before stop-loss and before expiry.
• HIT_SL — stop-loss was tagged first.
• EXPIRED — neither target nor stop-loss was reached within the signal’s validity window.
• CANCELLED — withdrawn by the publisher; excluded from hit-rate math.

2. Hit-rate over time

On the analytics dashboard, the “Hit-rate over time” chart aggregates weekly buckets over the last 90 days, separated by direction (bullish / bearish). For each week:

hit_rate(week, direction) =
  count(signals where status = HIT_TARGET)
  / count(signals where status IN (HIT_TARGET, HIT_SL, EXPIRED))

Buckets with no resolved signals are plotted as gaps, not zeros, to avoid misleading the viewer.

3. Confidence calibration

For each confidence bucket [0–40, 40–60, 60–80, 80+], we compute hit-rate across resolved signals in the 90-day window. A well-calibrated model shows monotonically increasing hit-rate with confidence. If the 80+ bucket fails to outperform the 60–80 bucket, the model is re-trained.

4. Leaderboard — Potential vs. Realized

Potential ranks live bullish signals by projected return (target − entry) / entry, with risk/reward ratio shown alongside. This is a forward-looking projection, not a guarantee.

Realized ranks symbols by the actual percentage return on closed trades:

realized_return(symbol) =
  sum(pnl of closed trades)
  / sum(entry_price * quantity of closed trades)

Realized numbers are aggregated per symbol; they reflect closed-trade outcomes logged on the platform, not per-signal attribution.

5. What we do not do

• We do not cherry-pick successful signals. Expired and stop-loss outcomes count toward the denominator.
• We do not claim guaranteed returns. Signals are probabilistic, not deterministic.
• We do not backfill hit-rate with synthetic data. Empty buckets stay empty.