A timeline read from an absolute clock absorbs a jump by moving its epoch, not by clamping a delta

Fact

An accumulating loop adds dt per frame, so a bad frame is handled by clamping that dt. A timeline whose position is clock() - epoch has no delta to clamp — the jump is already inside the answer. The only place to absorb it is the epoch:

const t = clock() - epoch;
const step = t - prev;
if (step > MAX_STEP || step < 0) {
  epoch += step - Math.max(0, Math.min(step, MAX_STEP));
}

Jumps are ordinary, not exotic: an audio context suspending and resuming, an output device changing, an occluded or throttled tab, a machine waking.

Why it matters

Applying the jump makes the timeline skip, and everything scheduled inside the skipped span is processed in a single frame — expiring, resolving or failing together. That reads as a logic bug in whatever consumed them, because the failure is a burst rather than a drift, and the clock looks healthy both before and after.

How to apply

• Decide per system whether a gap should be skipped or absorbed. Absorb it wherever a person is reacting to the timeline; skip it wherever the timeline models real elapsed time, such as an idle-progress economy.
• Absorb by moving the epoch, and clear prev at every legitimate discontinuity — a start, a seek, a resume — so the guard does not fire on one.
• A tab that stops delivering frames is the common case. Pause deliberately on visibilitychange and keep the absorber for the cases that raise no event.

Related: [[clock-rebase-needs-a-live-anchor]], [[driven-game-loop-needs-accumulator-reset]], [[animation-lifetime-in-frames]].