Needed today: root-caused the "trades are losing" report to a second,
deeper persistence bug -- gridPlaced (and the rest of SaveGridState's
GlobalVariable-based state) never reaches disk in this environment at
all (no .gvr file exists anywhere under the MT5 data dir). It survives
a same-process EA reinit (input edit) fine, but a full container
restart loses it completely, so every restart made every EA think no
grid was resting and place a fresh one on top of whatever was already
there, uncancelled. Confirmed via ~29 stacked pending orders on some
symbols spanning 24+ hours of normal hourly cycles, not just last
night's restarts -- this has likely been happening on every watchdog
restart for weeks.
Ran it today: cancelled 177 stacked/duplicate pending orders fleet-
wide, 0 remaining. Open positions untouched -- each EA's own OnTick
"Cycle closed?" check self-heals from an empty pending book and places
exactly one clean grid on its next tick.
