Everything on this page is counted, not claimed

The convention field is required, and here are three of them verbatim

“Why does your RSI differ from theirs in the first fortnight” is the support question an indicator library exists to be able to answer. So the convention is a required field on every study — the registry refuses a definition with an empty one — and it is shown in the settings dialog beside the parameters, not buried in documentation.

Relative Strength Index

Wilder’s RSI: average gain and average loss are RMAs (alpha = 1/length) seeded with the simple average of the first `length` CHANGES, so the first value sits at index length — length changes need length+1 closes — and not at length-1 like every other windowed indicator here. Rejected: Cutler’s RSI, which uses simple moving averages of the gains and losses and draws a visibly faster line; and EMA(2*length-1) smoothing, whose alpha matches Wilder’s but whose seed does not. A zero average loss reads 100, and a flat market 50, per core/guards.ts. Volume is not read.

Average True Range

Wilder’s RMA of true range, imported from ../ma/rma rather than re-derived. True range is undefined on bar 0, so the first input arrives at bar 1 and ATR-14’s first value is at index 14. Rejected: Pine’s ta.atr = rma(tr(true)), which plots high − low on bar 0 and therefore starts at index 13 with a bar-range averaged in as though it were a true range. Seeding is contested and is a parameter: Wilder’s SMA-of-the-first-length seed by default, the first-true-range seed as seed="first". Volume is not read.

Volume Weighted Average Price

Cumulative sum(price x size) / sum(size) since the session open, where the session is src/lib/session.ts’s answer and not a second calendar invented here — so the reset survives DST rather than sliding an hour twice a year. Quote volume is folded when the venue reports it, which makes the number EXACT and unchanged by timeframe; otherwise typical price x volume, which assumes every trade happened at (H+L+C)/3, and the `exact` output says which one this bar’s value was built from. Rejected: a rolling N-bar VWAP under the same name, which answers a different question; and standard-deviation bands, which are a second set of conventions. Anchored VWAP is the same accumulator started at an epoch second — not a bar index, which moves when the timeframe changes. A bar with no reported volume blanks the VWAP until the next reset.

The strongest evidence we have is external: this guide prints Wilder’s own 1978 RSI column beside the output of the engine that runs on this site, computed when the page is built.