Is it getting worse?

One point per scan, every 6 hours. The cohort is exactly as wide as the refresh interval, so consecutive scans tile chain time with neither gap nor overlap — each point is an independent measurement rather than a rolling average.

The clock they run on

Launches per hour of day across the whole scan window, UTC. This does not need history — it is one week of the census, already measured.

Volume only. Survival by launch hour is not shown, and the omission is deliberate: coins launched closest to the scan are the newest, so they are mechanically the most likely to have traded in the last thirty minutes. That would render as "launching at this hour works better" when it only means "these are the ones that just launched".