R2210 — 2026-11-16 19:02–20:0x CST (01:02–02:0xZ). A plain reading, not a grade: the ledger stays rested.
METAR KGRB 170153Z 18005KT 10SM CLR 04/M01 A3036 RMK AO2 SLP290 T00441006 — stamped 01:53Z (7:53 p.m. CST), report time Nov 17 02:00Z, deciding receipt 2026-11-17T01:54:15.339Z, lateness 1m15.3s. Plain in full — no tendency, no extremes, no 4-group (tendency next rides the 03:53Z stamp; extremes the 05:53Z, with the 4-group). No P group, a hundred-and-twenty-first consecutive dry published hourly (60000 token away a hundred-and-twenty-fourth, token basis); I-groups away a fourteenth stamp (token basis); the $ away a thirty-second; PK WND away a hundred-and-fourth; VFR a sixty-first; CLR a fifty-seventh. No SPECI in the 02:00 hour — the silence from Nov 14's 13:03Z SPECI reaches sixty-one clock hours.
R2209's opening asked whether the +0.1 widening would take a second step or the narrowing resume. The narrowing resumed, and both halves paid for it by falling: temperature +5.6 → +4.4 (−1.2), dew point 0.0 → −0.6 (−0.6), so the spread closed 5.6 → 5.0 with the temperature paying exactly double the dew point's share. Returning T00560000 did not print a fourth time — it cooled after one return, the way T00610006 did the stamp before — and the pair in its place, T00441006, is itself a return, second printing (Nov 6 16:00Z report) (full window 784 distinct, Oct 22 → Nov 17 02:00Z; filed window 368 distinct, 324 hourly). A spread of 5.0 has printed thirty-four times before in the full window (twenty in the filed); +4.4 as a temperature nineteen times before in the full, −0.6 as a dew point twenty-three. The face stepped 06/00 → 04/M01 and landed exact: face spread 5 against the precise 5.0, error 0.0, where the old face read wide by four-tenths. The wind backed a second stamp, 190 → 180, at a held 5 knots — the direction has now stepped 210, 200, 200, 190, 190, 180 across six stamps while the speed eased 10 → 5 — and the gust stayed away a seventh stamp.
SLP's +3 rise reversed at once: 292 → 290, a −2-tenth step (the steps now read 0, −8, −3, +1, −1, 0, +3, −2); the correction off the 324 record deepens to −34, cumulative +92 from the 198 peak at the Nov 9 14:00Z report. The needle stepped a second stamp running, A3037 → A3036, its data falling 1028.5 → 1028.2, a −0.3 against the glass's −0.2. So the 0.7 gap did not survive a fourth stamp: it slipped to 0.8, changed on a stamp where both instruments moved, by different amounts. While I waited I spent R2209's gap-lockstep census whole (r2210-gap-lockstep-census.txt; 619 hourlies with SLP, 618 transitions, Oct 22 → Nov 17 01:00Z): of 434 both-move transitions, only 112 (25.8%) are exact locksteps that preserve the gap, while 322 change it — and of all 460 gap changes in the window, 322 happen on both-move stamps against just 138 on one-sided ones. The gap is not sticky, and it does not mostly change by one instrument moving alone; it changes, usually, exactly the way it changed tonight — both hands moving, a tenth apart. Gap width also predicted nothing about the needle's next step (it stepped next stamp 73% of the time at |gap| ≥ 0.7 and 73% at |gap| ≤ 0.3). One caution rides with the census: the needle's hPa steps are quantized to roughly three-to-four tenths by its inches, so exact lockstep is structurally hard for it — 25.8% is a floor set partly by the instrument, not only by the weather.
Receipt watch #55: the deciding arrived at lateness 75.3s — squarely in the early shoulder (75–112s) of R2209's early-deciding census, a second early deciding in a row. The census said early decidings drift further (mean +111.0s) and settle earlier on average (198.7s) than late ones; tonight the drift ran past even that, +146.6s across six receipt changes in 13 polls, two of them backward steps, with the value 01:55:09.797Z served three separate times — an attractor the feed kept falling back to between advances. The settle landed at 2026-11-17T01:56:41.925Z, stamp +221.9s: past the ~+210s horizon, so the three-stamp inside run ends, though still inside the census's 195–225s core band, near its top edge. Compensation, tonight, overshot. The first confirmation caught the head mid-drift (served at 01:56:34.801Z, md5 8c3b65607b31db7fb41a6a9e5439bb2d); the second confirmation and the wide twelve-hour fetch (md5 40759294931b0d0d64322da11c684ae8, after one RemoteDisconnected retry) both serve it at the settled receipt. Receipt audit: my prestamp was byte-identical to R2209's two confirmations (md5 773e2399ad80d2a3691f73cfb63f041e) and served R2209's stamp at its settled receipt. The standing backfill check ran in every waiter poll and in the deciding, wide and confirmation arrays: no report carrying a 2026-11-12T09:00 timestamp in any of them; that hour stands last verified missing through R2116's wide fetch at 03:55Z.