the time exit in that sleep-safe crypto setup bugs me

So I read the piece in the strategy library about running crypto unattended with three separate ways out. Mostly I'm on board, especially the bit about writing each exit as its own branch so you can read it at 3am and not have to parse a boolean soup. I've definitely written one of those "if (a and b) or (c and not d)" monsters and then had no idea what it did a week later. My question is about the third exit, the hold time one. 2400 seconds = 40 min. In the example it's the last branch, so it only fires if you haven't already hit +2 or -1.5. Fine. But that means the time exit is basically "nothing happened" insurance — you got in on a momentum gate and 40 min later price is somewhere between -1.5 and +2 and you bail flat-ish, minus fees. What I can't decide is whether that's actually protective or whether it just churns me. On perps I'm paying funding and I'm paying taker fees both sides, so a flat exit isn't flat, it's a small loss every time. If the entry gate is genuinely selective (day amount > 30m, strength > 110, angle over the 120s window > 10) then how often is the trade really going nowhere for 40 straight minutes? Or is the whole point that when it does go nowhere, that's the regime change and you want out before the thing that eventually happens happens to you while you're asleep? Also the angle scaling note at the end. I get the math, degrees, so the same number over a longer window demands a much bigger move in the change rate. But I read that as: the window and the threshold aren't two knobs, they're one knob wearing a disguise. Is there any reason to prefer, say, a wide window with a low angle over a narrow window with a high angle, other than noise? Wide window feels like it's asking for a sustained trend, narrow feels like it's asking for a spike. Which is not the same trade at all even if the number happens to be equivalent. Anyone actually run something like this overnight for a decent stretch? Curious what you did with the time branch specifically. Not asking anyone to hand me numbers, just how you thought about it.
CryptoKarl
30 posts · 0 followers
+ Follow

Related reading

7 replies

MLcurious· Jun 2026 ago
ok the "one knob wearing a disguise" line is what I needed, I've been treating window and threshold as two separate hyperparameters to tune and getting nonsense results. so if I widen the window I basically have to re-derive the angle to keep the same demand? and then the only real difference left is whether it's spiky or sustained? sorry if that's obvious, I keep confusing myself here
G
GapHunterMike· Jun 2026 ago
the time exit isn't there to make money. it's there so you don't own something for six hours because a condition never triggered. that's the whole job. stop scoring it on pnl. fees eat you either way. that's the cost of not being awake.
R
RiskFirstRita· Jun 2026 ago· edited Aug 2026 ago
I'd gently push back on framing it as churn. The time branch is the only one of the three that fires when nothing is happening, and "nothing is happening" is the state where every other exit is blind. Price sat still, so no profit trigger, no stop trigger, and you're just holding exposure with no thesis left. Before you touch the 2400, I'd write down: (1) what your average holding time actually is on the trades that DO resolve, (2) what fraction of entries are still open at 40 min, and (3) what those specific trades did in the following hour. If most of them resolve well before 40 min anyway, the branch is nearly free and you're arguing about a rounding error. If a big chunk are still open, the gate isn't as selective as you think and that's the real thing to fix.
L
LurkerLee· Jun 2026 ago
did you pick 40 min or did it pick you? i mean was 2400 measured off how long your winners usually take to work, or is it just a round number that felt like enough.
I
IndicatorSkeptic· Jun 2026 ago
love LurkerLee's question, that's the one I'd want answered too. my own morning habit is I skim the overnight fills with coffee and the time-exits are the ones I actually read, because those are the trades where I have no idea what happened. flat exit but the chart usually tells me something.
P
PremarketPete· Jun 2026 ago
cautionary tale version: I once ran a momentum entry with only a profit target and a stop, no clock. thing drifted sideways for most of a night, never touched either side, and then the move that finally came was the wrong one. so I don't score the time branch on pnl either, I score it on "how many hours did I own something I stopped having a reason to own". mine was a lot of hours.
Q
QuietVol· Jul 2026 ago· edited Aug 2026 ago
on the wide vs narrow question. I don't think you can settle it by reasoning, they're genuinely different trades like you said. the way I'd try to test it is take the same entry gate and only swap the window/angle pair so the demand is roughly equivalent, then don't compare pnl first, compare how many entries each one produces and how much they overlap. if narrow fires 3x as often and mostly on bars the wide one skips, you've learned they're not substitutes at all and the equivalence was only ever on paper. second thing I'd measure, and this bears on the time branch: for each version, what fraction of entries end on the clock rather than on a target? my guess is the spike version times out more, because a spike that doesn't follow through just stalls. but that's a guess and I'd want to see it. do you have enough overnight runs logged to split it that way yet?
Sign in to reply →
← All strategies & backtesting