Judge Savage, Patrick S.

The last twelve months and the years before them are never merged: motion practice changes faster than a case runs, and a single five-year number would average two different eras. A motion record carries no judge code of its own — it is taken from the nearest hearing by date, that is, from whoever was actually sitting on the case at the time; six motions in ten attach that way, the rest are credited to the judge of the proceeding, and the two methods disagree in one case out of five. The "nationally" column shows where judges genuinely differ: procedural motions are granted by nearly everyone nearly always.

The recent window and the years before it are shown apart: adjournment practice changes faster than a case runs, and it is not only the frequencies that move but the reasons themselves, in different directions. Case-completion codes (8A, 8B, 8C) are separated from genuine continuances by behaviour rather than by wording: after them another hearing rarely follows, while after preparation continuances it almost always does. That is what the "follow-up" column shows, and a low value means "the case usually ended here", not "the adjournment failed". The median days figure is how far this reason pushed the next hearing; half the cases fall between the quartiles shown.

A dismissed appeal means the removal order stood when the respondent appealed, and that the asylum grant stood when the government did; the two do not go into one percentage. Recent years are undercounted: an appeal takes months or years to resolve, so outcomes are counted by the date of the Board's decision while the appealed share is counted by the date of the judge's decision.

The grounds themselves are listed in 8 CFR 1003.1(e)(6) — all of them about complexity or a departure from law. Counting goes by the date of the Board's decision, not the date the appeal was filed: referral happens between the two, and an unresolved appeal may lack the marker simply because it has not been reached yet. The two windows are shown apart because the referral rate has moved markedly over the past year, and almost entirely among respondent appeals: without the filer axis the shift would read as a general one. A share is stated from fifty decided appeals in a window upwards: the event is rare, and on fewer the interval is wider than the whole spread between judges.

Counting is by the date the second round ended; unfinished rounds are excluded — among cases remanded over the past five years, about a third are still running. A share is stated from thirty completed rounds upwards: on fewer the interval is wider than the whole spread between judges. There is no "now and before" pair here as in neighbouring blocks: second rounds are rare and slow, about three per judge in a year. The history is in the yearly series — which also shows why one figure for five years is not enough: what moved the headline share was the share of terminations, not judges growing stricter.