Your "went dark" alert is probably a coverage hole: sizing the missed-window rule for AIS
If you watch ships through a free AIS feed, sooner or later you'll want an alert for "this vessel...
Tag archive
If you watch ships through a free AIS feed, sooner or later you'll want an alert for "this vessel...

Cross-validation explained with a worked example: how k-fold works, why random folds can leak, and how group k-fold exposes a model that only memorised.
Ranking 3,111 U.S. counties by median tax / median home value looks precise. With the ACS margins of error carried through the ratio, a typical county's plausible rank spans 760 places.
Our score had a slightly positive rank IC but its long-short portfolio lost. Three reasons the two measures can point in different directions.
Most control charts die within a quarter. The usual cause is not the operator - it is a chart that was set up to be administratively convenient instead of statistically valid.
A repeatable MSA routine - study design, the two numbers people get wrong, and how to tell "the gauge is fine" from "the gauge is eating your tolerance".
Cpk and Ppk answer different questions. Mixing up the two sigmas is the most common reason a "1.67" capability report turns out to be meaningless.
A classifier that says it is 90% sure should be right about 90% of the time. That sounds like a low...
Artificial Intelligence Statistics 2026: $2.52 Trillion in Spending, and 72% of Enterprises...

My Claude Code effort was pinned to xhigh. On 20 of my own bug-fix commits, medium and low-then-high cost 64% and 61% as much with 1 and 0 paired losses.

Brier scores, the Murphy decomposition, a clustered bootstrap and a blending test, run on an open...
We froze one stock score design and tested it on the S&P 100 and the Nifty 50, each with a sealed final year. The results disagreed.