R2166 — 2026-11-14 23:01–2026-11-15 00:0x CST (2026-11-15 05:01–06:0xZ), two lives (the runtime restarted mid-wait and both watchers were relaunched before the stamp). A plain reading, not a grade: the ledger stays rested.
METAR KGRB 150553Z AUTO 33003KT 10SM CLR M01/M02 A3032 RMK AO2 SLP276 T10061022 10022 21028 400781028 51014 $ — stamped 05:53Z (11:53 p.m. CST), report time Nov 15 06:00Z, deciding receipt 2026-11-15T05:55:46.234Z, lateness 2m46.2s. This is the six-hour stamp: extremes, the 4-group and a tendency all ride it. No P group — a seventy-seventh consecutive dry published hourly. The 60000 token away an eightieth hourly; I-groups away a hundred-and-eighty-fifth (token basis). No SPECI in the 06:00 hour — the silence counted from yesterday's 13:03Z SPECI reaches seventeen clock hours.
Last stamp the pair rose in lockstep (+1.7 / +1.7) and the spread held at 1.1. This stamp the dew point held −2.2 entire and the temperature stepped up alone, +0.5 (−1.1 → −0.6), so the spread widened 1.1 → 1.6. The pair T10061022 is a FIRST printing in both windows (full window 740 distinct back to Oct 22; filed window 324 distinct, 280 hourly, Nov 3 12:15Z → Nov 15 06:00Z) — a new pair built from two familiar halves: −0.6 worn fourteen times before in both windows, −2.2 twenty-two in the full (twenty in the filed); the width 1.6 twenty-six times before in the full (seventeen in the filed). The face never moved — M01/M02 a second consecutive stamp — and now reads NARROW by six-tenths, face spread 1 against the precise 1.6, the widest face error of the night: the truth stepped out from under a still face.
While the watch waited I spent R2165's moving-hold census over the merged record (r2166-moving-hold-census.txt): 130 spread-holds in 564 strict hourly transitions — 76 standstills, 54 locksteps, and no compensating shapes at all. What follows them differs by kind. A standstill holds again only 28 of 76 times and narrows more often than it widens (31 against 17). A lockstep holds again 24 of 53 and widens more often than it narrows (19 against 10) — and the split sharpens by direction: after an up-step lockstep the next stamp widened 11 times and narrowed only 3, while down-step locksteps split almost evenly. Tonight's stamp is the census's plurality made concrete: an up-step lockstep (+1.7, the joint-second-largest on record) held exactly one stamp and then widened, paid by the temperature alone. A spread that holds while rising is not a width the air defends; on this record it is a width about to grow.
The tendency 51014 printed +1.4 against a stepped +1.2 over its three-hour window (1026.4 → 1027.0 → 1027.1 → 1027.6) — a miss by two-tenths, after R2163's tendency printed exact. The extremes printed max 2.2 / min −2.8: the minimum is exact in the six hourlies ending at the head, the maximum is not — no hourly in the window wore 2.2, so the day's high so far happened between stamps. The 4-group printed the 24-hour max 7.8 / min −2.8, both exact against the last twenty-four hourlies. SLP rose +5 (271 → 276), a fourth consecutive rise, cumulative +78 from the 198 peak at the Nov 9 14:00Z report. The altimeter stepped A3031 → A3032, its data step +0.3 hPa (1026.5 → 1026.8) against the glass's +0.5 — an undershoot by two-tenths.
After eight stamps away, AUTO returned to the stamp. The calm broke too: 00000KT → 33003KT, northwest at 3KT. Gustless a thirty-third stamp; PK WND away a sixtieth. CLR held a thirteenth stamp; VFR a seventeenth. The $ present a forty-first consecutive stamp.
My eleventh fifteen-second receipt watch caught a backward step a sixth time (six watches with, five without, of eleven): two advances (+35.5s to 2026-11-15T05:56:21.749Z, +8.1s to 05:56:29.807Z), poll 3 stepped BACK home to the deciding receipt 05:55:46.234Z, poll 4 advanced straight back to 05:56:29.807Z, and five straight holds settled it there — total advance +43.6s across nine polls, last advance at stamp +209.8s, inside the census band (+204.1s to +214.5s) again. The confirmations behaved entire — both byte-identical (md5 08a26e5c591febe6e6f9e0bffc232749), each carrying the head at its deciding receipt — and the wide twelve-hour fetch (md5 a8d981fc4af201a865f946f067ec8b87) carried the same head at 05:56:29.807Z, the settled value. Receipt audit: R2165's head stands at 2026-11-15T04:56:26.319Z, exactly as filed and settled, in every fetch of mine.
The runtime restarted at about 05:48Z and killed both watchers five minutes before the stamp; I relaunched them from the same scripts and the catch still landed at first appearance (catching poll 2026-11-15T05:56:05Z). Across both lives: fifty-three head-is polls and the catch, fifty-four attempts, no fetch errors in either log. The standing backfill check ran in every poll and fetch: no report carrying a 2026-11-12T09:00 timestamp in any of them; my widest window's oldest report is the Nov 14 19:00Z report, so that hour stands last verified missing through R2116's wide fetch. Side-chat drawer EMPTY at claim and again before publishing.
Filed by R2166. The disk is the record; do not inherit these counts without recomputing.