Cases

Documents about what the system measured and about where the system was wrong. There is no publishing schedule: one gets written when there is something verifiable to tell. Four of the five tell of a mistake of its own.

• Detection

  Detecting the Gulf airspace closure with open data

  Two independent sources recorded the event on the same day and narrowed the collapse to seven hours. Of the eight countries that closed their airspace, only one shut down the internet.

• Discard

  When the engineering is right and the measurement is wrong

  The best-built pipeline in the system was capturing the wrong metric. No code review would have found it: it took checking the source against another one during a real event.

• Audit

  A finding that did not survive its own audit

  A perfect separation across thirteen nodes, at p = 0.00078. On review, two thirds of the data did not exist and the clock was wrong. What held, and what did not.

• Rebuild

  A bug that never threw an error

  No exceptions, no error logs and no symptoms for months: the system never knew for certain what hour it had measured at. Diagnosis and rebuild of 242 days.

• Contrast

  Nine countries without internet that never disconnected

  On 2 September, connectivity in nine countries fell at once according to the signal the system uses. Routes, traffic to Google and flights stayed the same: the only thing that fell was the response to the probes.

The same, once a day, on Telegram.

This site carries no advertising, but it does cost money to run.If you find it useful, you can support it.