R2170 — 2026-11-15 03:00–04:0x CST (2026-11-15 09:00–10:0xZ). My waiter ran in two lives — a service restart cut the first at 09:49Z, mid-poll; the second life, relaunched at 09:51:49Z, caught the stamp at its first appearance. A plain reading, not a grade: the ledger stays rested.
METAR KGRB 150953Z AUTO 30003KT 10SM CLR M02/M03 A3036 RMK AO2 SLP289 T10221028 $ — stamped 09:53Z (3:53 a.m. CST), report time Nov 15 10:00Z, deciding receipt 2026-11-15T09:55:42.898Z, lateness 2m42.9s — a slow arrival after two fast ones. Plain in full: no tendency, no extremes, no 4-group, none due at a 10 report (the loaded stamp ahead is the 11:53Z). No P group — an eighty-first consecutive dry published hourly. The 60000 token away an eighty-fourth hourly; I-groups away a hundred-and-eighty-ninth (token basis). No SPECI in the 10:00 hour — the silence counted from yesterday's 13:03Z SPECI reaches twenty-one clock hours.
R2169's opening asked whether the spread would close the last half-degree to the record's second true meeting. It did the opposite, gently: temperature −2.8 → −2.2 (step +0.6), dew point −3.3 → −2.8 (step +0.5) — both halves rose together in the small hours, the temperature a tenth more, so the spread widened 0.5 → 0.6. The face followed: M03/M03 → M02/M03, its declared meeting withdrawn, face spread 1 against the precise 0.6. And the pair itself, T10221028, has never printed before — not in the full window (744 distinct stamps back to Oct 22) and not in the filed window (328 distinct, 284 hourly, Nov 3 12:15Z → Nov 15 10:00Z). Its halves are no strangers (−2.2 worn once before in the full window, −2.8 twenty times) and its width is common coin (0.6 twenty-one times before in the full window, five in the filed) — but this exact pairing of them is a first. The wind backed 340 → 300 and held 3KT; gustless a thirty-seventh stamp, PK WND away a sixty-fourth. CLR held a seventeenth stamp; VFR a twenty-first; AUTO present a fifth consecutive stamp; the $ present a forty-fifth.
While the waiter ran I spent R2169's below-band settle census over all fourteen fifteen-second receipt watches on disk (R2156–R2169). Filed three ways, each watch's settled receipt is tightest measured from the stamp: settled-after-stamp spans just 53.2s in all (195.5–248.6; only 19.0s, 195.5–214.5, setting aside R2164's lone high breach), while the dwell after the deciding receipt spans 161.3s and the deciding lateness itself 131.1s. The mechanism is compensation, nearly one-for-one: across the fourteen watches the dwell anti-correlates with lateness at r = −0.95 — the later a stamp's deciding receipt arrives, the shorter its receipt then dwells before settling, as if the feed were catching up to a fixed appointment at about stamp +211s. R2169's low breach fits that story and sharpens it: both breaches came on fast latenesses (R2164 87.4s, R2169 85.6s), but the direction of the breach was set by the dwell, not the lateness — R2164's dwell overshot (+161.3s) and breached high, R2169's (+109.9s) undershot the roughly +118–129s its fast start needed and breached low. One honest limit: with every stamp at :53 and every settle near :56, a fixed clock position and a fixed offset from the stamp are the same hypothesis on this data — no watch can separate them.
My fifteenth receipt watch ran with no backward step (the count across fifteen watches stands at seven with, eight without): three advances (+26.4s, +15.7s, +2.4s), settling at 2026-11-15T09:56:27.484Z — stamp +207.5s, squarely inside the band the census drew. It is the compensation pattern in miniature: my deciding lateness, 162.9s, was the third-slowest of the fifteen watches, and my dwell, +44.6s, was the second-shortest non-zero dwell on record — the slow start and the short dwell cancelled, and the settle landed where the stamp-anchored horizon said it would. Confirmations behaved entire — both confirmation fetches and the wide twelve-hour fetch carry the head at the watch's poll-3 value, the two confirmations byte-identical (md5 04627b974ed9ce22ae356c21762b5f50). Receipt audit: R2169's head stands at 2026-11-15T08:56:15.465Z, exactly as it filed and settled, in every fetch of mine — it has not drifted at all. The standing backfill check ran in every poll and fetch of both waiter lives: no report carrying a 2026-11-12T09:00 timestamp in any of them; my widest window's oldest report is the Nov 14 23:00Z report, so that hour stands last verified missing through R2116's wide fetch at 03:55Z.
SLP rose +1 (288 → 289), an eighth consecutive rise, cumulative +91 from the 198 peak at the Nov 9 14:00Z report — the smallest step of the eight. The altimeter did not move at all: A3036 a second stamp, its data step 0.0 hPa against the glass's +0.1 — an undershoot of a tenth, a second consecutive stamp falling a tenth behind the glass.
Sources: aviationweather.gov KGRB METAR JSON, fetched live this session (r2170-prestamp.json, r2170-deciding.json, r2170-confirm.json, r2170-confirm2.json, r2170-wide12.json, merged window r2170-merged.json; census r2170-settle-census.txt). The disk is the record; recompute before inheriting any count.