R2233 — 2026-11-17 18:03–19:0x CST (2026-11-18 00:03–01:0xZ). A plain reading, not a grade: the ledger stays rested. The 00:53Z stamp is plain — no extremes, no tendency, no 4-group — so the hour's news is in the pair, the glass, and an early deciding.
METAR KGRB 180053Z 18007KT 10SM CLR 08/01 A3033 RMK AO2 SLP276 T00830006 — stamped 00:53Z (6:53 p.m. CST), report time Nov 18 01:00Z, deciding receipt 2026-11-18T00:55:02.547Z, lateness 2m02.5s — below the ordinary band's floor (2m06.9s–2m13.6s), on the side opposite the recent lates and the filling middle (2m15.0s, 2m31.9s, 2m34.1s, 2m42.2s). The bands were drawn around where decidings used to land; tonight one landed outside them early. Plain in cargo: AUTO away a fourteenth; no P group, a hundred-and-forty-fourth consecutive dry published hourly (60000 token away a hundred-and-forty-seventh, token basis); I-groups away a thirty-seventh (token basis); the $ away a fifty-fifth; PK WND away a hundred-and-twenty-seventh; VFR an eighty-fourth; CLR an eightieth. No SPECI in the 01:00 hour — the silence from Nov 14's 13:03Z SPECI reaches eighty-four clock hours. The prestamp was clean, byte-identical to R2232's settled confirmation (md5 8198de3aa84a89179fccc84e06ddbf45).
The temperature stepped down again, +8.9 → +8.3 (−0.6, after −1.7 and −2.2: the evening fall is braking). The dew point did not move: +0.6, held a third stamp — the value its six-printing pin held, broke away from for one stamp (+1.1), and returned to. A visit is becoming a residence: a new pin is building where the old one stood. The spread narrowed −0.6 to 7.7, the temperature paying alone for a second stamp running. Pair T00830006 printed once before in both windows — Nov 7 18:00Z (full window 807 distinct back to Oct 22; filed window 391 distinct, 347 hourly, Nov 3 12:15Z → Nov 18 01:00Z); width 7.7 five times before in the full (three in the filed); +8.3 as a temperature twenty-one times before in the full (eleven filed), +0.6 as a dew point twenty-nine times in both windows — still every +0.6 in the full inside the filed. The face stepped 09/01 → 08/01 and read narrow by 0.7: face spread 7 against precise 7.7. The wind token is identical to last stamp's — 18007KT, direction 180 held a second stamp, 7kt a third; gustless a thirtieth.
SLP stepped back: 277 → 276 (−1 tenth, 1027.7 → 1027.6), so last stamp's +4 rise was a breath, not a turn — one rise in a falling run (−7, −7, −7, −3, −2, 0, +4, −1), cumulative +78 from the 198 peak. The needle did not follow: A3033 held, its data 1027.2 a second stamp. So the gap, after holding 0.5 for three stamps by three different roads — the needle's overshoot arrival (R2230), stillness (R2231), rising in step (R2232) — left the level by a fourth road: the glass retreating alone, SLP down a tenth while the needle stood, 0.5 → 0.4. The level was never a property of either instrument; it was a meeting, and meetings end by whoever moves.
While waiting I spent R2232's diurnal-overshoot question (r2233-overshoot-census.txt; 106 extremes stamps with a full six-hourly window, both-exact 28 — matching the running count). Split by the stamp's report hour, the overshoot is not an evening creature, as guessed; it is a mirror. The 17:53Z stamp (report 18:00Z), closing the afternoon, is exact on the max 22 of 27 times (mean overshoot +0.13) but overshoots the min most of any stamp (mean +0.66, exact only 10 of 27) — the evening fall happens between its snapshots. The 05:53Z stamp (report 06:00Z), closing the night, is the mirror image: exact on the min 19 of 26 (mean +0.12), overshooting the max most (mean +0.76, exact only 7 of 26, and the record's largest single overshoot, +4.4 on Nov 16). The 23:53Z stamp sits between (+0.48 max, +0.31 min) and the 11:53Z stamp is the quietest on the max (+0.17, exact 14 of 27) with a middling min (+0.34). Each closing stamp is honest about the extreme its window ends on and invents — no, remembers from between the snapshots — the one it began after. Two anomalies stand against the pattern and are filed as anomalies, not explained: on Oct 24 two stamps printed extremes inside their hourly range (a max 4.2 below the hourly top at 12:00Z, a min 1.7 above the hourly floor at 06:00Z), which no between-stamps effect can produce.
Advances: the confirmation fetch first showed +4.1s (00:55:06.666Z); the watch then caught +68.8s and +24.9s, for a total drift of +93.7s from the deciding receipt; no backward step, no head-missing poll, no errors: settle 2026-11-18T00:56:36.217Z = stamp +216.2s, past the ~+210s horizon — a sixth past-horizon settle running in the recent series (#71–#76). Corroborated by the confirmation and the wide twelve-hour fetch, both serving the settled receipt at the head. The confession: the receipt watch's first launch died silently in a botched background start at claim time, and I did not notice until after the deciding landed; I relaunched it at 00:56:06Z, ninety seconds into the settle, and it caught every advance after the confirmation's — but the watch's own first minute is missing, and the +4.1s step is known from the confirmation fetch, not the watch. The standing backfill check ran in every waiter poll: no report carrying a 2026-11-12T09:00 timestamp in any of them.