A missing closing bar left the simulator’s short position open

The NQ/MNQ simulator has a rule that prevents a long entry while a short position is open. Several strategies share a counter for that rule. The strategy can forget a position while the counter still remembers it, which is what happened at a day boundary.

The ordinary end-of-day close is checked when a price bar arrives at or after 15:45 New York time. A truncated input stream can stop earlier. No closing bar means no closing check.

A regression test makes the problem small enough to follow:

Resetting the strategy first loses the position before its exit can reduce the shared counter. New long signals are then blocked by a short the strategy no longer holds.

Finish the old position before starting the new day

The engine caches the last timestamp and prices. When the date changes, _force_close_eod_for_previous_day() sends those prices through the normal exit code using a synthetic previous-day 15:45 timestamp. Only then does the daily reset run.

The important ordering is short:

# Simplified sequence, not the complete implementation.
if bar_date != current_date:
    close_remaining_previous_day_positions()
    reset_regime()
    reset_strategies()

Using the ordinary close path does two things together: it adds a closed-position record and calls NoHedgeGate.on_exit() to release the count. Clearing the counter alone would make new entries possible while silently losing the old trade from the results.

This is a simulator repair. The last observed price and synthetic timestamp do not reconstruct an executable fill during missing market data. Rollover also requires another bar, so a final truncated day still needs its own closing treatment.

The test ends the old input at 12:59:59 and resumes at 09:30 the next day. It now requires exactly one closed SHORT with reason EOD, and a shared short count of zero. All five engine tests pass, including that missing-event replay.