One fixed long-only trend rule on TLT, played bar by bar on real history or on synthetic shock paths. Your choices (B) are scored next to the rule taking every signal (A).
NO EPISODE
—Bar
—Open R
—Closed R (B)
TLT daily · strategy tf_breakout_long v1Start an episode to see bars
SMA50SMA20020-bar high (breakout line)Stop in forcePrice axis in % from the close before the episode. Dates stay hidden until the episode ends.
New episode
Data
Real TLT bars 2002–2026, raw prices. You may have seen any part of this history before.
Length
Costs, sizing and rules
Strict mode is always on: no entry without a signal, no lower stop.Rule check at the last close
Valid signal on this close
Entry would fill at the next bar's open. Stop = fill − 2 × ATR14 of this bar. Decide before the next bar is shown.
Right-arrow key also advances one bar. "To next event" stops at a signal, a fill or an exit.
Off-rule actions (logged)
Episode result
Account equity per run (sizing assumption, not the R sum)
Record
—
Rows are kept apart by data mode and shock path. Synthetic paths appear at chosen frequencies, not real-world odds, so no row here is an expected return.
No finished episodes yet. Start one above; it is saved here when it ends.
Assumptions and limits
Rule v1 (long only). Signal on a closed bar when close > SMA50, SMA50 > SMA200, SMA50 above its value 5 bars earlier, and close > the highest high of the 20 bars before it. Entry at the next open. Initial stop = fill − 2 × ATR14 (signal bar); 1R is fixed there. Trailing stop = max(stop, highest close since entry − 3 × ATR14), effective from the next bar, never lowered. No target, no pyramiding, one position at a time. A filter turning off does not close the trade.
Fills are modelled. A stop is taken as hit when the bar's low reaches it; a bar that opens below the stop fills at the open. Daily bars cannot show the intraday path, quotes or halts, so the real fill could differ.
Prices are raw, not adjusted. TLT's monthly distributions are not added to returns, which understates a held long. One bad bar (2007-01-06, a copy of 2007-01-05) was removed.
History is not unseen. Every historical bar may already be familiar from earlier practice, and the same history feeds the synthetic generator. The held-out evaluation seeds are new paths, but built from the same history.
Synthetic data re-samples real daily bar shapes in blocks (mean length 20) and adds an assumed path (trend, volatility multiplier, gap events). Volume and yields are not generated. Policy events are scenario assumptions and may fade.
Open at the end. A position still held when an episode ends is valued at the last close less exit slippage and shown apart from closed trades.