CCU sampling and gaps: when a line is evidence

What the chart actually contains

Each point records the public playing count for one Universe at a specific time. The scheduler may offer a market collection slot every five minutes, but a rotating catalog does not promise one point for every game in every slot. A blank interval is unknown. It is neither zero nor proof that the game was unavailable. The safe unit of analysis is the covered window: the timestamps and observations that exist.

Before computing growth, inspect the latest timestamp, spacing, number of distinct time buckets, and whether both the baseline and comparison windows have enough real points. MARPLA’s current one-hour scout calculation uses the average of the latest 15 minutes against the 45–75 minute window, requires at least two observations separated by five minutes in each window, and rejects a baseline average below three. Those are product rules for that signal, not universal statistical guarantees.

Worked example: a missing peak

Game A has observations at 18:00, 18:05, 18:10, and 19:00: 120, 130, 125, and 210 CCU. Game B has points every five minutes and rises from 120 to 170. It is tempting to say A grew faster, but the fifty-minute gap hides the path. A might have climbed steadily, spiked to 500 and cooled, or stayed near 125 until 18:59. The only supported claims are that A was observed at 125 at 18:10 and 210 at 19:00, and that the end-to-end observed change was +85.

For a comparison, restrict both games to a window with adequate overlapping coverage or wait for another cycle. If the decision is urgent, report A as “possible acceleration, sparse evidence” and make the next observation the decision trigger. Do not draw interpolated points and call them collected data.

Coverage check before interpretation

Save the timestamps with the conclusion so another reviewer can reproduce the window.

1. Confirm Universe ID and the newest observation time.
2. Count distinct buckets in the baseline and comparison windows.
3. Mark gaps explicitly and compare only overlapping covered periods.
4. Use absolute change with percentage change and the starting base.
5. Schedule a follow-up when coverage can change the decision.

Pitfalls that create false movement

Do not convert a failed or missing request to zero, bridge long gaps as if the path were known, or compare a densely sampled game with a sparse game without qualification. Avoid treating a scheduled interval as a delivery guarantee. A single browser refresh also does not prove that the background collector has completed a new sample.

Coverage solves only one problem. Even a complete CCU series does not identify unique users, retention, source mix, or cause. A regular daily peak may reflect geography and routine rather than a new product effect. Pair the line with Visits increments, updates, chart observations, and owner analytics where authorized.

Decision record

For every cited movement, record the two endpoint timestamps, the observations included in each window, coverage, and the rule used to exclude gaps. Preserve the raw points or export beside the chart. If a later collection fills part of the period, append a new analysis instead of rewriting what was known at decision time.

Primary sources

• Roblox games API
• Roblox Analytics