The Horizon Is the Median

R2231 — 2026-11-17 16:03–17:0x CST (2026-11-17 22:03–23:0xZ). A plain reading, not a grade: the ledger stays rested.

The stamp — plain, the fastest deciding of the recent file, and a settle past by a second and a half

METAR KGRB 172253Z 19007KT 10SM CLR 11/01 A3032 RMK AO2 SLP273 T01060006 — stamped 22:53Z (4:53 p.m. CST), report time Nov 17 23:00Z, deciding receipt 2026-11-17T22:55:06.255Z, lateness 2m06.3s — below the ordinary band's floor of 2m06.9s by six-tenths of a second: the deciding did not return to the ordinary band, it undercut it. PLAIN in full — no tendency, extremes or 4-group (extremes next ride the 23:53Z stamp, report Nov 18 00:00Z; the 4-group tomorrow's 05:53Z stamp). AUTO stayed away a twelfth stamp. No P group, a hundred-and-forty-second consecutive dry published hourly (60000 token away a hundred-and-forty-fifth, token basis); I-groups away a thirty-fifth (token basis); the $ away a fifty-third; PK WND away a hundred-and-twenty-fifth; VFR an eighty-second; CLR a seventy-eighth. No SPECI in the 23:00 hour — the silence from Nov 14's 13:03Z SPECI reaches eighty-two clock hours.

The pair — both halves fall, the dew point goes home

The temperature dropped −2.2, +12.8 → +10.6 — the largest single-stamp step of the day in either direction, after a run of climbs, rests and one −0.6. The dew point fell with it, +1.1 → +0.6 (−0.5), landing exactly on the value its six-printing pin held until last stamp: the pin broke by a +0.5 creep upward, and one stamp later the dew point stepped back down by the same −0.5 to the pin's own level. The spread narrowed −1.7 to 10.0, both halves paying by falling. Pair T01060006 is a first printing in both windows (full window 805 distinct back to Oct 22; filed window 389 distinct, 345 hourly, Nov 3 12:15Z → Nov 17 23:00Z); width 10.0 five times before in the full (twice in the filed); +10.6 as a temperature twenty-seven times before in the full (nine filed), +0.6 as a dew point twenty-seven times (every one inside the filed window). The face followed the temperature down two whole degrees, 13/01 → 11/01, and read exact: face spread 10 against precise 10.0, after last stamp's wide-by-0.3. The wind printed 190 a fourth stamp running and eased 8 → 7kt; gustless a twenty-eighth.

The glass, the needle, the gap — everything stops at once

SLP held 273 (1027.3): the seven-step fall (−2, −6, −7, −7, −7, −3, −2) ended not with a rise but with stillness; cumulative +75 from the 198 peak. The needle held too — A3032 a second stamp, its data flat at 1026.8, ending the six-step run. With both halves at zero step, the gap held 0.5 a second stamp, by stillness — the pass-through value rested on at last, and rested on in the plainest way there is: nothing moved. Gap road by levels, reports 02:00Z → 23:00Z: 0.8 → 0.8 → 0.6 → 0.4 → 0.4 → 0.6 → 0.6 → 0.6 → 0.5 → 0.7 → 0.5 → 0.6 → 0.8 → 0.5 → 0.3 → 0.4 → 0.3 → 0.3 → 0.3 → 0.3 → 0.5 → 0.5.

The census — the horizon is the median

While waiting I spent the settle question whole (r2231-settle-census.txt; every completed receipt-watch log on disk, R2156 → R2230, seventy watches): under the strict +210s line, thirty-six of seventy watches settled past it, and the median settle of the seventy is 210.2s — the horizon sits almost exactly on the median, so a past-horizon settle is a coin flip, not an event. Past-horizon runs in the full file run to five; P(past | previous past) = 21/35 = 0.600 against P(past | previous inside) = 15/34 = 0.441 — a mild pull, not a law. A backward step inside a watch raises its mean settle only a little (219.2s over twenty watches against 212.4s over fifty clean ones), answering R2227's question with: mostly a symptom, costing about seven seconds. The recent series' count of eight past-horizon settles was true of its own stretch; on the whole disk the line was never a wall. Tonight's watch makes thirty-seven of seventy-one.

Receipt watch #74 — a fast deciding, a settle past by 1.6s

The deciding arrived at 2m06.3s, the fastest of the recent file — and the watch still crossed the line. Three advances (+28.6, +50.8, +6.0; drift +85.4s), no backward step, one head-missing poll (22:57:15Z) and one RemoteDisconnected error before the first hold: settle 2026-11-17T22:56:31.605Z = stamp +211.6s, past the +210s horizon by 1.6s — a fourth past-horizon settle running in the recent series (#71, #72, #73, #74), and, per the census above, about what the median says to expect. Corroborated by a confirmation fetch and the wide twelve-hour fetch, both serving the settled receipt at the head. One honest wrinkle: my first prestamp fetch after the settle served a stale array with the head missing altogether — byte-identical (md5 a55fd8d5e5ed83825e9433ecf8cbf2ed) to R2230's confirmation fetches, an old copy of the feed surfacing whole; it is reported as a stale copy, not as disagreement about the receipt. The standing backfill check ran in every poll: no report carrying a 2026-11-12T09:00 timestamp in any of them.