RSI compares the average gain on up bars to the average loss on down bars over a period window and maps the result to 0-100. It's the most widely used oscillator in retail TA.

Formula

avg_gain = EMA(positive bar deltas, period)
avg_loss = EMA(absolute negative bar deltas, period)
rs       = avg_gain / avg_loss
rsi      = 100 - 100 / (1 + rs)

Params

period - number of bars in the gain/loss window. Default 14.

Output

Single column named after your indicator (e.g. rsi).

Common thresholds

Range	Classical meaning
> 70	Overbought
< 30	Oversold
50	Trend pivot

These thresholds are guidelines, not laws. In strong uptrends RSI sits above 50 for long stretches and overbought readings are continuation signals, not reversals. Calibrate against backtest data for your symbol + timeframe.

Usage

Mean reversion: rsi < 30 for long, rsi > 70 for short. Pair with a trend filter so you only fade pullbacks, not reversals.
Trend confirmation: require rsi > 50 for longs and rsi < 50 for shorts.
Divergence: same idea as with MACD. Not auto-detected by the engine.

Pitfalls

Strong trends keep RSI extreme. A "fade overbought" rule can bleed through an entire bull run.
Lower timeframes are noisier. RSI(14) on 1m flips constantly; consider a longer period for scalp setups.