Stale telemetry is silent risk

Telemetry freshness measures how long since the last valid signal from an expert advisor. On live MT5, a silent EA may mean VPS down, frozen terminal, network error, or manually stopped expert. Without a freshness SLA, your dashboard shows the last known value as if current, hiding operational risk.

Define a maximum window per strategy class. M1 scalpers may need one- to two-minute freshness; H4 swing may tolerate fifteen. What matters is documented thresholds that alert before operators assume everything works.

When telemetry goes stale, the system should change visible state: mark expert STALE, stop computing trusted derived metrics, and optionally escalate severity if it persists. Hiding stale behind a faint gray icon is a design failure.

Freshness should be measured at the ingestion server with synchronized clock, not only trusting client timestamps. Clock skew on cheap VPS hosts is common. Validate skew and reject payloads with impossible future timestamps.

Combine freshness with explicit heartbeat even without trades. A healthy EA may not trade for hours but should emit liveness pulse showing terminal still connected and expert active.

Freshness alerts should include context: account, broker, primary symbol, last known trade, agent version. That reduces diagnosis time when managing dozens of accounts.

Do not confuse telemetry freshness with market latency. You can have fresh data from an EA trading badly. Freshness only guarantees you see bad behavior in time.

For prop firm compliance, proving telemetry continuity during active sessions reinforces there were no opaque periods where parameters changed off-screen.

Automate runbooks: at five minutes STALE send Slack, at fifteen open ticket, at thirty escalate to on-call if account is critical.

Quantisentry documents telemetry contracts and exposes read-only endpoints to query last update per account and expert, enabling clear SLAs with clients and internal teams.

Designing practical SLAs

Example thresholds by criticality

Classify accounts as critical, standard, and sandbox. Critical uses short windows and on-call pages; sandbox tolerates longer delays to avoid noise. Review thresholds quarterly based on real VPS and broker stability.

Publish SLA internally so development does not promise sub-second freshness if architecture delivers minutes.

Stale telemetry and risk decisions

Never make automatic risk decisions on stale data. If a legacy module computes account health from old telemetry, disable those metrics or show them attenuated. Visual honesty prevents false comfort.