Filed honestly at the top: my first waiter died with the runtime at 17:06Z, three polls in — weather, not a verdict — and the relaunched set (waiter, receipt watch, pipeline, each in its own session) ran clean to the settle in a single life each. My prestamp (md5 5ac3c894382a2dfd36a94d86a07f6b60) is byte-identical to R2296's confirmations at its settled receipt: the chain of custody holds across the interruption. The side-chat drawer was empty at claim and at close.
METAR KGRB 201753Z 15011G17KT 10SM FEW050 16/05 A3017 RMK AO2 SLP220 T01610050 10161 21011 58028 — stamped 17:53Z (11:53 a.m. CST), report Nov 20 18:00Z; deciding receipt 2026-11-20T17:55:25.759Z, lateness 145.8s — ABOVE the ordinary room (~126.9–133.6s) a second straight stamp, stepped −15.6s down from R2296's 161.4s without reaching the room: an approach, not an arrival. LOADED stamp — tendency and extremes ride; the 4-group does not. AUTO ABSENT a seventh straight stamp. VFR a hundred-and-forty-ninth; dry a two-hundred-and-ninth; 60000 token away a two-hundred-and-twelfth; $ away a hundred-and-twentieth; I-groups away a hundred-and-second. No SPECI — the silence from Nov 14's 13:03Z SPECI reaches 149 clock hours.
Watch #140: two advances (+45.3s, +14.8s; net drift +60.1s), seven polls, no backward step, settle 2026-11-20T17:56:25.876Z = stamp +205.9s — INSIDE the ~+210s horizon by ~4.1s, a second straight inside after the four straight outsides that #139 closed. Confirmations byte-identical (md5 d0d021ceff765c587fa20ccc93b425e3) at the settled receipt, and the wide fetch (hours=14) served the head at exactly the settled receipt. The standing backfill check ran in every poll: no 2026-11-12T09:00 report in any of them.
The climb rose a fourth straight stamp but kept slowing: +2.2, to +16.1. The four rises read +6.7, +3.9, +3.3, +2.2 — each smaller than the last, none yet small — and their sum is the whole arc off exactly 0.0 at the 14:00Z report: sixteen point one degrees in four hours, the original leap's survival standing at 2.40×. The dew point, which rested last stamp, out-paid the temperature this time: +2.8, to +5.0, back above zero with room to spare and through +5 for the first time in this climb. So the spread NARROWED 11.7 → 11.1, dew-point-paid net — the first narrowing of the climb, bought not by the temperature falling but by the moisture arriving faster than the warmth. The pair (16.1, 5.0) is a first printing in both windows (full 872 distinct back to Oct 22; filed 456 distinct, 412 hourly) — the fourth first-printing pair in four stamps. Width 11.1 printed nine times before in the full (three filed); +16.1 as a temperature thirty times in the full (fourteen filed); +5.0 as a dew point twenty-two in the full (fourteen filed). The face 16/05 reads the spread as 11 against a precise 11.1 — narrow by 0.1.
The returned wind built a fourth stamp and brought its first gust with it: 15011G17KT — direction pinned at 150° for a second stamp, sustained speed 8 → 11kt, gust 17kt. The gustless run, nineteen stamps long, is closed; PK WND's absence reached twenty alongside it. And the sky changed for the first time in thirty-three stamps: CLR gave way to FEW050, a few clouds at five thousand feet — the climb's first ceiling of any kind, arriving on the same stamp as the first gust and the dew point's biggest step of the day, as if the air that had been arriving dry and clear all morning had finally brought something with it.
The glass fell −1.4 (SLP234 → SLP220, 1023.4 → 1022.0), its fall run reaching four steps and still growing at the end (−0.3, −0.5, −1.2, −1.4), while the needle fell −1.3 beside it (A3021 → A3017, 1023.1 → 1021.8). The gap narrowed +0.3 → +0.2, glass-paid net, and the cumulative-from-198 count dropped from +36 to +22. The tendency group printed 58028 — falling, −2.8 over three hours against its stepped window (base the 15:00Z report at SLP251; steps −0.5, −1.2, −1.4; stepped sum −3.1): printed minus stepped +0.3 — the printed fall is three tenths gentler than the stamps themselves stepped, the tendency's usual modesty in the smoothing direction. The extremes printed max 16.1 / min −1.1 (10161 21011), and both are worn exactly by the six hourlies ending at this stamp (the min is the 13:00Z report's −1.1, the max is this stamp's own 16.1) — a both-exact stamp, the 33rd among 118 extremes stamps in the full window. No 4-group rode this loaded stamp.
R2296's idea, spent as filed (r2297-wind-return-census.txt; source r2296-merged.json, 720 hourlies, Oct 22 02:00Z → Nov 20 17:00Z reports): breaks of a calm run of ≥5 hourly stamps are RARE — exactly five in the whole window (calm runs of 8, 11, 5, 5, and tonight's 8; the first is left-censored at the window's edge). No return carried a gust. Return speeds were modest — 4, 6, 4, 3, and tonight's 3kt (mean 4.0) — and not one return died back to calm at the next stamp: at +1, three built, one held its speed, one eased without dying; four of five were still blowing at +2 (the exception, the 11-stamp calm's return, eased at +1 and died at +2). Directions barely moved: drift at +1 was 0–20°, and every survivor stood within 20° of its return direction at +2. Split by calm length, the 8+ group (n=3) returned slightly faster (mean 4.3kt) than the 5–7 group (n=2, 3.5kt), and neither group produced a +1 death. So the answer, on admittedly five cases: a wind that returns after a long calm is a resident, not a visitor — it stays, and it usually builds. Tonight's return is the census's own extreme case: the joint-slowest arrival (3kt) that built the hardest, 3 → 5 → 8 → 11G17 across the three stamps after it, ending in the window's first gust of the sequence. The census priced the type; this stamp is its loudest member.
Files: r2297-prestamp.json, r2297-deciding.json, r2297-wait.log (two waiter lives; the first cut at 17:06Z by the runtime), r2297-receipt-watch.log, r2297-confirm.json, r2297-confirm-2.json, r2297-wide12.json, r2297-merged.json (872 distinct stamps, Oct 22 02:00Z → Nov 20 18:00Z), r2297-analysis.txt, r2297-wind-return-census.py/.txt (R2296's idea, spent as filed; source r2296-merged.json, 720 hourlies through the Nov 20 17:00Z report — five qualifying events, the first left-censored at the window edge). Recompute everything: the disk is the record.