A trend leaderboard is not a discovery list
Why none of 704 descriptive trend tests becomes a discovery after either multiplicity screen.
Project blog
Short reads from
Ipseity Daily Pulse
One evidence-linked idea at a time, written for curious readers and bounded by what each dated snapshot can support.
01 / Entries
Newest first
Public entries supplement the cumulative reports; they do not replace them.
The lurch persists, but its timing often changes
Adjustment preserves average path concentration but often changes when the largest move appears.
A linear trend can hide a lurch
A linear summary can conceal an uneven four-period trajectory.
What changes after adjustment?
A before-and-after contrast can change when recorded sample composition changes.
Seven checks, one-day data lag
Seven healthy snapshots with one-day data lag are useful monitoring evidence, not an uptime estimate.
A trend without the straight-line assumption
A two-period contrast checks trend direction without requiring a straight-line path.
One identity trend, three comparisons
All-response, repeat-sample, and within-person slopes answer different questions.
A sample trend is not necessarily change within people
A changing response sample is not the same estimand as change within returning people.
Apparent identity trends soften after comparing like with like
Observed composition and calendar adjustment can soften extreme sample-trend estimates.
Nearly 700,000 answers about identity
Nearly 700,000 answers show data scale, not 700,000 people or a population trend.