Scott Alexander makes predictions at the start of every year and scores them at the end of each year.

Judged as of 2016-01-25. True: from when he started the practice up to January 2016, Slate Star Codex published a set of predictions at the start of every year and a scoring post at the end of every year (or soon after it), with no year missing. Mostly true: he did this as a regular yearly practice, but with a minor irregularity (for example, a scoring post published in the following January, or the practice only a few years old). Mixed: he did it in some years but skipped others. Mostly false: he did it only once, or he made predictions but rarely scored them. False: there is no such yearly predict-and-score practice.

Set before the research began.

1. E1 Predictions For 2014 Slate Star Codex · 28 January 2014 Own record

   “Unless otherwise specified, predictions are about whether something happens before the end of 2014, except LW Survey predictions which will be judged whenever next survey is finished”

2. E2 (Late) Predictions for 2015 Slate Star Codex · 13 June 2015 Own record

   “I was supposed to institute a tradition of making predictions at the beginning of each year, then grading them at the end to test my calibration. Everything went according to plan last year – last January I made predictions for 2014 and then this January I scored them.”

3. E3 (Late) Predictions for 2015 Slate Star Codex · 13 June 2015 Own record

   “I ended with “2015 predictions coming soon,” then totally forgot to do that.”

4. E4 2015 Predictions: Calibration Results Slate Star Codex · 2 January 2016 Own record

   “In keeping with the SSC tradition, here are the results for my (late) predictions for 2015.”

5. E5 Archives Slate Star Codex Own record

   “01/25 - Predictions For 2016”

Slate Star Codex posted predictions for 2014 in January 2014 (E1). The June 2015 post says those 2014 predictions were scored in January 2015 (E2). The 2015 predictions, however, came out late, in June 2015, because the author "totally forgot" to post them in January (E2, E3). They were scored on 2016-01-02 (E4), and predictions for 2016 followed on 01/25 (E5). So by January 2016 the predict-and-score practice had run every year since it began, but it was only two years old and the 2015 predictions were published mid-year, not at the start of the year. That fits the criteria for "mostly true" rather than "true".