How to read these charts

What is a cohort? A group of CPIDs (researcher accounts) bucketed by the calendar month they first appeared on the chain. Specifically, the month containing the first block their CPID staked. So a cohort labelled 2024-01 is every CPID whose first-ever staked block landed in January 2024.

What does each curve plot?Month-by-month survival of that cohort. The X axis is months since the cohort was formed (0, 1, 2, …); the Y axis is the share of the original cohort still active in that month, from 0 to 1. A CPID counts as "active" in month N if it staked at least one block in that calendar month. Month 0 is always 1.0 by construction; every member is active in its own joining month.

Worked example. Say 2024-01 has cohortSize 87 (87 CPIDs first appeared that month). If 71 of those 87 also stake a block in February, month 1 reads 81 %; if 41 stake in July, month 6 reads 47 %; and so on out to the 12-month horizon. The CPIDs don't need to stake every month in between, only in the specific month we're looking at.

What the shape tells you.

• Tall, slowly-decaying curve. Sticky cohort, most members kept staking. Healthy retention.
• Cliff in the first 1-3 months."Tourists": researchers who joined, tried it, and dropped off. Common during hype cycles.
• Stable middle, late drop-off. Long-term participants who eventually rotated out (hardware retired, BOINC project shut down, etc.).
• Curves of recent cohorts much lower than older ones. The chain is leaking new researchers faster than it used to. Worth investigating what changed.