Document your personal trading journey. Track daily equity curves, review winning and losing streaks, share trade screenshots, and get constructive feedback.
// Inside CNewsFilter::FetchCalendar() after parsing the JSON node
string json_date_string = "2026-10-01T10:00:00-04:00"; // Extracted from JAson
m_next_danger_time = ParseISO8601ToBrokerTime(json_date_string);
PrintFormat("Next High-Impact Event localized to Broker Time: %s", TimeToString(m_next_danger_time));
A note on Daylight Saving Time (DST) gaps: Europe and the United States shift their clocks on different weekends. Because this function calculates the instantaneous broker offset dynamically, checking a Friday event on Monday during that weird 2-week DST gap might calculate 1 hour off. However, because your EA continuously refreshes the data, as the broker server rolls over the DST boundary, the dynamic TimeTradeServer() offset instantly recalculates, and your 30-minute danger window will snap to perfect accuracy before the actual event fires.
Preserve your own money. Scale with the market's money. Exponential growth is the ultimate key.
The example date is a good one to test with. The string in your snippet, 10:00 on 1 October at UTC-4, is 14:00 UTC, the ISM manufacturing release, which always lands at 10:00 New York on the first business day of the month. Because the string carries its own offset, the conversion to UTC is exact, and the only thing that can go wrong is the broker offset you add afterwards.
That offset is the one taken at the moment of conversion. Most MT servers run on New York daylight saving dates, so a broker switching from GMT+3 to GMT+2 does so on the same weekend as the US clocks. If the calendar is fetched before that weekend and the event falls after it, the stored broker time will be an hour out until the next refresh. Your refresh loop fixes it as long as it runs at least once after the Sunday open, which it will if the EA is running continuously. After a VPS reboot I'd force a refetch on initialisation rather than relying on cached times.
TimeGMT() also depends on the VPS clock being correct. A drifting clock shifts every event by the same error, so a periodic time sync on the machine is worth having.
Network and restart behaviour before CPU every time. I have wasted enough hours comparing instance prices while the real failures were weekend relaunches and a clock that had drifted three seconds.
What I measure on a new box now: round-trip to the broker gateway under London load, what the EA does after a forced disconnect, and whether the system clock is still sane on Monday morning. If those three are boring, the $22 versus “premium” argument can wait.
Uptime dashboards make people feel professional. A reconnect that re-subscribes quotes cleanly is the actual test.
Days without an unhandled critical error is the right month-one metric. Cost-adjusted P&L in the first thirty days is mostly noise next to a clean reconnect and a sane rollover.
Hosting choice is easy to over-argue. Surviving disconnects, clock drift, and a weekend restart that still subscribes quotes — that is the live test. Manual intervention count belongs on the dashboard next to fills.
The separation of trade logic and risk management is good architecture. One more layer worth adding: a "sanity check" before each order that verifies the lot size and stop distance are within expected ranges. Bugs in lot size calculation are one of the most common causes of large unexpected losses in EAs.
A good complement is to test the EA's behaviour when the broker connection drops and returns, since some EAs send duplicate orders after reconnecting.