1. What the system observed. Reviewing exit alerts for the weekend of 2026-08-22/24 found four positions — TSLA, GOOGL, INTC, and PALLADIUM — all with their perp leg closing on Saturday and the IB hedge leg deferred 47–59 hours to Monday's market open. Combined naked-leg P&L that single weekend: −$167.56 (INTC alone: −$145.44, flipping an otherwise funding-profitable trade into a net loss).
2. Why it happened — the root cause. A config setting meant to pause exits when the hedge
market is closed only guarded one of four exit paths (the FR sign-flip grace timer), and only
on the INTX venue. The fr_decay debounce timer and every Hyperliquid-venue exit were never
weekend-aware at all.
3. What the model expected. A quick fix — extend the same weekend pause to the other exit paths and both venues — should have closed the gap.
4. What changed. Investigating a separate case (SNDK, −$392.43) to check whether it was the same weekend issue revealed it wasn't a weekend at all — an ordinary Tuesday-evening naked leg. Checking every historical exit's actual market status (not day-of-week) found 7 of the last 9 funding-decay and grace-timer exits since June had fired while the hedge market was already closed — mostly on ordinary weeknights. The mechanism: the 4–8 hour grace timers are shorter than the hedge market's ~17.5-hour daily closed window, so any signal that turns late in the trading day has a good chance of expiring its timer after the close, regardless of which day of the week it is.
5. The trade-off, checked before building anything further. Before extending the fix further, the actual cost of always pausing through a closed market was measured against four trades with reliable naked-leg data:
| Trade | Extra funding cost of waiting | Naked-leg P&L avoided |
|---|---|---|
| NATGAS | −$2.69 | +$61.03 |
| MU | −$1.56 | +$309.56 |
| SNDK | +$2.39 | +$381.84 |
| PALLADIUM | +$6.75 | −$33.25 |
| Total | −$4.89 | +$719.19 |
Three of four trades benefited from pausing, one lost — the cost side is small by construction (funding rates near the exit floor are already small), while the benefit side is a stock-price gap, a different and much larger-magnitude risk.
6. Final result. Replaced the day-of-week check entirely with a live, per-symbol "is the hedge market actually open right now" lookup — the same authoritative source the daemon already used elsewhere to decide whether to defer a close. Applied to all four exit paths. Emergency and max-hold exits were left firing regardless of market status, by design — cutting a catastrophic funding blowup immediately is judged more important than a market-hours pause.
7. What was learned. "Weekend" was the wrong abstraction from the start. The real condition was always "is the hedge market open right now" — and once grace/decay timers are shorter than the hedge's own daily closed window, ordinary weeknight gaps are exactly as likely as weekend ones. See What happens when the perp closes before the hedge for the general mechanism this fix addresses.