Platform updates that quietly change behaviour

[Fairman]

Platform updates that quietly change behaviour

Trading platforms get updated, sometimes automatically. A new build can change how orders are processed, how stops are handled, how history is stored, or how indicators behave.

Most updates are harmless. A few have caused real trouble: an expert advisor that stops working after a new build, custom indicators that break, changes to rounding or margin calculations, and shifts in how the tester generates ticks.

The danger is that the change happens quietly. Results shift, and the trader blames the market or the strategy when the cause was a different version of the software.

A basic habit is noting the platform build number in the trade journal and checking it after any restart or update. If results change suddenly, that is the first thing to look at.

Where possible, keep a stable copy of a working installation and test updates on a demo before moving the live account.

A simple precaution is keeping a text file next to the journal listing the build number, the date of any update, and any behaviour changes noticed. It looks pedantic until the day it explains something.

Has a platform update ever changed results without any change in the strategy?

[FXS HongKong Scalper]

Where possible, keep a stable copy of a working installation and test updates on a demo before moving the live account.

Auto-updates are the reason it is worth pinning the platform build on a VPS that runs an EA. Most brokers allow staying on a build for a while. Running the new build on a demo terminal for a week first, with the same EA, catches most surprises in the tester and in order handling before they reach a live account.

[Fairman]

The demo test is a good first gate, but I would treat it as a regression check rather than proof that live behavior is unchanged. A demo can expose a compile failure, altered indicator output, or a tester discrepancy; it may not reproduce live liquidity, routing, or margin conditions. The useful comparison is controlled: record the platform build, broker server, symbol specifications, EA or indicator version, inputs, and tester settings, then change only the build and compare the same scenarios and logs. That makes it easier to isolate whether a difference came from the platform or from a changed feed or configuration.

I would define a short acceptance checklist before rollout—successful startup, no new errors, expected order types and stop levels, and consistent results on a fixed test set. Keep a backup of configuration and source files, and preserve the prior run’s logs so a change can be investigated later. If an update is mandatory or security-related, delaying it indefinitely may create a different risk; instead test promptly in a separate environment and use a cautious, monitored rollout with predefined limits. The exact build number is helpful, but pairing it with the broker server and symbol settings is what makes the journal entry reproducible.