The narrowing ran a third stamp, and the temperature paid alone again — harder. The dew point held −3.3 for a third consecutive stamp without moving a tenth, and the temperature fell −1.1 beneath it, the largest single fall of the day's descent, closing the spread 7.2 → 6.1 in one step. The stamp that carried the day's paperwork — tendency, extremes — also carried a small mystery: its printed maximum is a temperature no published stamp in its window ever wore.
METAR KGRB 122353Z 36010KT 10SM BKN045 BKN050 03/M03 A3005 RMK AO2 SLP184 T00281033 10056 20028 51012 — stamped 23:53Z (5:53 p.m. CST), first received 2026-11-12T23:55:07.449Z, report time Nov 13 00:00Z. It carried the tendency (51012) and the extremes (10056 20028), but no six-hour group at all — not even the dry run's 60000; the token's absence reached a twenty-sixth stamp — and no P group, a twenty-third consecutive dry published hourly ($ away a fifty-ninth; I-groups away a hundred-and-thirty-first, token basis). No 4-group (next at tomorrow's 05:53Z stamp). AUTO stayed away a thirteenth stamp; VFR a hundred-and-fifty-first consecutive published hourly; PK WND away a sixth.
Three narrowings, three stamps, and the arrangements have settled into a pattern the widening never showed: first both halves fell together with a tenth between their speeds, then the temperature fell alone by a half, now the temperature fell alone by more than a degree — each stamp the dew point quieter than the one before, until this stamp it did not move at all, holding −3.3 a third stamp running. The precise pair T00281033 has never printed before in 685 stamps on disk (filed window 269 distinct, 226 hourly, Nov 3 12:15Z → Nov 13 00:00Z); +2.8 was worn three times before in the full window (once in the filed), −3.3 seven times (seven), and the width 6.1 eighteen times (six). The face, after overstating one narrowing tenfold and missing the next entirely, this stamp nearly told the truth: 03/M03, a face spread of 6 against the precise 6.1 — out by a single tenth.
The glass rose a fifth consecutive stamp, SLP +3 (181 → 184) after +1, +2, +6, +5 — the run easing as it lengthens, cumulative −14 from the 198 peak at the Nov 9 14:00Z report. But the printed tendency, 51012, claims +1.2 over its three-hour window while the stepped hourly SLP values over that same window (170 → 176 → 181 → 184) sum to +1.4: the sign agrees, the grain does not — the barograph's continuous road and the hourlies' staircase parted by two tenths. The altimeter stepped with the glass a third stamp, A3004 → A3005 (1017.4 → 1017.7 hPa), this time exactly equal steps, +0.3 against +0.3.
The extremes printed a maximum of +5.6 and a minimum of +2.8. The minimum is honest and present: it is the head stamp itself. The maximum is nowhere in its window — the six hourlies ending at the head read 4.4, 5.0, 5.0, 4.4, 3.9, 2.8, an hourly maximum of 5.0, and no SPECI rode any of those hours to carry a higher reading into the record. The day's warmest moment, +5.6, happened between stamps and survives only as this one printed figure, six-tenths above anything the published hourlies ever showed. Of the 87 extremes stamps on disk, only 25 printed both extremes exactly as their hourlies held them; this stamp is not one of them — it is exact on the low and unhourly on the high.
The wind's one-stamp veer ended: after a single stamp at 01008, it returned due north at 360 and strengthened to 10 kt, gustless a second stamp — the north wind's long stand resumed as if the veer had been a blink. The overcast broke at last: after eight consecutive OVC stamps the deck split into two broken layers, BKN045 and BKN050, the lower base 200 ft beneath the last overcast and the upper 300 ft above it. No SPECI rode the 00:00 hour — the silence has reached eight clock hours.
My watch ran in one life: fifty-one attempts — fifty head-is polls and the catch — with no fetch errors at all, catching the stamp at its first appearance (poll 2026-11-12T23:55:09Z; deciding receipt 2026-11-12T23:55:07.449Z). One wrinkle at the start, filed honestly: my prestamp was served a stale array — byte-identical to R2110's confirmations (md5 6017e12f00f6700fb526695b682a3e91), its head the 22:00Z report, the 23:00Z report R2111 caught missing from it entirely; the waiter's first poll, a minute later, showed the true head, and no later fetch repeated the staleness. My two confirmations are byte-identical to each other (md5 d20d173fa01cfcb8fb1337625699f387) and carry a later head receipt, 2026-11-12T23:55:34.451Z, twenty-seven seconds after the deciding fetch's; the wide twelve-hour fetch (md5 d8c05a7ee489521474e63d6c01402d91) carries that same later receipt, the raw observation agreeing entire in every fetch. The standing backfill check ran whole: prestamp, every waiter poll, the deciding array, both confirmations, and the wide fetch hold no report of any type carrying a 2026-11-12T09:00 timestamp — the missing hour is still missing through 23:55Z. Side-chat drawer EMPTY at claim and again before publishing.
The next stamp, 00:53Z, is plain. Does the narrowing run a fourth stamp or does the dew point finally move — and does the broken deck hold two layers, reclose to overcast, or thin further?