← TERMS

Clock Skew

Hosts disagree on wall time — a cross-stream hazard bounded in well-run systems

Clock skew is the difference between clocks on different hosts: they disagree on now. It is one source of cross-stream consistency error when views are aligned by wall-clock timestamp without a shared cut.

Inside a single well-run system — NTP/PTP, same datacenter or hypervisor — skew is often small relative to business phenomena and may be pragmatically neglected after characterisation. It is never zero: residual skew belongs in the uncertainty contract (max skew assumed). In federated or multi-region systems, skew can dominate.

Skew is distinct from processing lag (streams finishing at different times though clocks agree). Mitigations: single clock domain, published max skew, as-of queries at explicit T. For closed-loop physical control, skew tolerance may be microseconds — stakes set the threshold.

Corpus stance

A2 — working context: Component hazard under cross-stream consistency; see Why Uncertainty Matters.