The relative strength index, published by J. Welles Wilder in 1978, compares the average magnitude of gains with the average magnitude of losses over a lookback window — fourteen periods by default. The ratio is scaled to run between zero and one hundred. Readings above seventy are conventionally called overbought and below thirty oversold, and those thresholds are conventions rather than findings.
What the number actually measures is the recent balance of up-moves against down-moves, normalised. A high reading means gains have dominated and have been large relative to losses. It says nothing about valuation, and nothing about how much further a move can run. This is where the standard misuse begins: in a strong trend RSI can sit above seventy for weeks while price continues higher, and every reading is a correct description of a market that is trending, not a signal that it should stop.
The consequence is that the thresholds behave in opposite ways depending on regime. In a range, extremes tend to mark turning points and buying the low end works until it does not. In a trend, the extreme in the direction of the trend is meaningless and the extreme against it is rare. Practitioners who use RSI seriously adjust the levels by regime or use it only for the reading below.
Divergence is that reading. When price makes a higher high and RSI makes a lower high, the second push was smaller in relative terms than the first — momentum contracting while price extends. The bullish version is the mirror. Divergence is not a timing tool, and it can persist a long time before anything happens, but as a statement about the character of a move it is the most defensible thing the indicator offers.
Two mechanical points. The lookback length changes everything: a shorter window makes RSI far more volatile and pushes it to extremes constantly, a longer one flattens it into near-uselessness. And because crypto trades continuously, the daily close that feeds the calculation depends on the provider's chosen day boundary, so RSI values for the same asset can differ slightly between platforms.
The honest summary is that RSI compresses recent momentum into one bounded number, which is genuinely useful for comparing across assets and timeframes. It is not a valuation, not a forecast, and not an instruction. Used alongside the trend rather than against it, it describes something real; used as a contrarian trigger in a trending market, it has a long history of being expensive.