“Last 24 hours” sounds a lot like “today,” but they are different measurements. Today starts at midnight. A rolling window starts exactly 24 hours before the page was calculated, so both ends of it move whenever the figure refreshes.
A window with two moving edges
Imagine a creator received 10,000 observed coins at 15:00 yesterday and 2,000 at 10:00 today. At 14:30 today, both events sit inside the preceding 24 hours, so the total is 12,000. At 15:01, yesterday’s larger event has aged out. If nothing new arrives, the displayed total becomes 2,000.
No coins were removed and no correction was made. The window simply stopped including yesterday's 15:00 event. Stream time, comments, likes and gifts all behave in the same way.
Why use rolling periods?
Rolling periods provide a current view without favouring one time zone. A creator streaming around midnight does not have their activity split simply because a calendar date changed. Visitors in different countries also see the same underlying interval rather than a local definition of “today.”
They are useful for questions such as “what has happened recently?” They are less suitable for accounting-style questions such as “what happened on Monday?” A calendar report with an explicit time zone is better for the latter.
Longer windows are steadier, not static
A 7-day or 30-day window smooths individual events because each stream represents a smaller share of the total. It still moves. A particularly successful stream can create a visible step down exactly seven or thirty days later when it expires from the period.
This creates a common analytics illusion: a creator appears to decline despite maintaining ordinary current activity. The actual explanation may be that an exceptional older stream left the comparison window. Inspecting the individual stream history prevents that mechanical change from being mistaken for a new trend.
Partial streams at the boundary
A stream can cross the start of a rolling window. Systems must choose whether to include the entire stream, only events inside the window, or allocate duration across the boundary. Those choices can produce different results even when the source observations are identical.
GiftTok aggregates events and recorded stream activity according to its published methodology. When comparing with another dashboard, check whether both use event time, stream start time or stream end time. The label “24 hours” alone does not specify that implementation detail.
Refresh timing and cached pages
A rolling window is calculated at a particular moment, but a displayed page may be cached. GiftTok’s initial public board snapshot can be up to one hour old. In a normal browser, JavaScript requests fresher board data after the page loads. A crawler or a browser with scripts disabled may continue to see the timestamped edge snapshot.
That staleness is acceptable for broad comparisons, but it matters near a boundary. If an older event is due to expire, two visitors can briefly see different totals because one has a fresh response and another has a cached snapshot. The generation timestamp tells you which calculation you are reading.
Comparing periods fairly
Use complete, equal-duration periods where possible. Comparing the last 24 hours at midday with an earlier calendar day mixes a moving interval and a fixed interval. For week-on-week analysis, compare equivalent rolling windows or complete calendar weeks in the same time zone.
Do not overreact to a single update. Record several points, inspect the underlying streams, and note unusual events. A rolling total is a useful summary, but the movement of its boundary is part of the measurement—not noise to be ignored.