Function atrTrailingStop

ATR Trailing Stop — a stop a fixed number of ATRs from the close that only ever moves in the direction of the trade, and flips side when the close crosses it. Kernel K6 over atrValues.

d = multiplier × ATR(period)          (Wilder's ATR)

close > prev and prevClose > prev  ->  stop = max(prev, close − d)   long, ratchet
close < prev and prevClose < prev  ->  stop = min(prev, close + d)   short, ratchet
close > prev                       ->  stop = close − d              flip long
otherwise                          ->  stop = close + d              flip short
Copy

Appends two columns:

• ${prefix} — the stop level (ats at the default prefix).
• ${prefix}Trend — +1 while the stop is below price, −1 above.

The two-column shape and the sign convention are parabolicSar's, and for the same reason: the side is what a consumer draws, and on a flip bar the stop can print exactly at the close (when ATR = 0), so stop < close is not a safe derivation of it.

Which trailing stop — the anchor is the close, and the alternative is named

Two families circulate under names this close together that picking one has to be explicit:

• This study: the band is measured from the close (close ± multiplier × ATR), ratcheted, with the four-case flip above. This is the form published as "ATR Trailing Stop" — Sylvain Vervoort's, and the one TradingView's widely-copied ATR Trailing Stop / "UT Bot" scripts implement.
• The Chandelier Exit (Chuck LeBeau): the band is measured from the rolling extreme instead (highestHigh(period) − multiplier × ATR while long, lowestLow(period) + multiplier × ATR while short). That is a different study, not a variant of this one, and it is a composition away from what the package already ships — donchian supplies both extremes and atr the width. It is not implemented here, and this study deliberately does not grow an anchor: 'close' | 'extreme' knob: the two differ in more than one place (the extreme form also needs its own warm-up, max(period, period), and its flip is conventionally tested on the close against the other side's band), so a knob would be two studies wearing one name.

high and low are read only by the ATR — the stop itself is a close-to-close construction.

Warm-up and gaps

Length-preserving; both columns start on the ATR's first bar, which is bar period on gap-free input. As with superTrend, the ATR is the gap rule in practice: an interior hole leaves Wilder's recursion NaN to the end, so the study reads undefined from there. The K6 reset ([PND-SFOLD]) is what governs a gap in close alone.

Edges

• In the units of the price: scaling every bar by k scales the stop by k; adding c shifts it by c. The trend column is unchanged by either.
• A flat stretch gives ATR = 0 and a stop sitting exactly on the close. The fourth case is written otherwise rather than close < prev precisely so that close === prev resolves — it flips (or stays) short, which is the same tie-break the published form uses. No division, so no zero-denominator case.
• multiplier is a width, not a bar count, so it is validated as a positive finite number rather than by assertPeriod.