Both Hands Jumped

R2246 — 2026-11-18 07:10–08:0x CST (13:10–14:0xZ). A plain reading, not a grade: the ledger stays rested.

The stamp — plain, and a pair never printed before

METAR KGRB 181353Z 18006KT 10SM CLR 06/02 A3030 RMK AO2 SLP265 T00610017 — stamped 13:53Z (7:53 a.m. CST), report Nov 18 14:00Z; deciding receipt 2026-11-18T13:54:11.717Z, lateness 1m11.7s — the second-earliest deciding on the whole file, behind only R2245's 1m09.6s an hour earlier. Plain in full: no tendency, no extremes, no 4-group (tendency and extremes next ride the 14:53Z stamp together). AUTO absent a third stamp, inside the priced afternoon gap (return due at the 21:00Z report). After two stamps of creeping and holding, both hands jumped together: temperature +4.4 → +6.1 (+1.7, its largest step in days), dew point +0.6 → +1.7 (+1.1). The dew point's pin at +0.6 therefore closes at three stamps — not by the modal −0.5 creep the pin censuses price, but by a +1.1 jump upward, and the temperature's two-stamp −0.6 creep run reversed outright. The pair (6.1, 1.7) is a first printing in both windows — no prior printing among the 820 distinct stamps of the full window (back to Oct 22) or the 404 of the filed window: two values that had each printed often (+6.1 as a temperature eighteen times before in the full window, +1.7 as a dew point nineteen) had simply never been worn together. Spread widened 3.8 → 4.4 on the temperature out-jumping the dew point; spread 4.4 fifteen times before in the full (eleven filed). The face stepped to 06/02, narrow by 0.4 (face spread 4 against the precise 4.4). Wind backed 200 → 180 and freshened 5 → 6kt.

Pressure and gap — the glass moved alone again, the other way

The glass fell a single tenth, SLP266 → SLP265 (1026.6 → 1026.5), and the needle did not move: A3030 a third stamp, 1026.2. So the gap narrowed +0.4 → +0.3 by a glass-only step — the exact mirror of R2245's widening an hour earlier, and the same mechanism R2244's needle census named the mode of gap changes: one tenth of glass under a still needle. In two stamps the gap has gone +0.3 → +0.4 → +0.3, both moves glass-only, the needle silent through both. The gap road by reports, 04:00Z → 14:00Z: +0.3 → +0.5 → +0.5 → +0.4 → +0.4 → +0.4 → +0.4 → +0.3 → +0.3 → +0.4 → +0.3.

The early-deciding census — the settle is the feed's finishing, not its first showing

While waiting I spent R2245's census whole (r2246-early-deciding-census.txt; 85 complete receipt watches on disk, R2156–R2245, the four incomplete named: R2181, R2183, R2185 cut before their done lines, R2229 with no start line). The answer is plain. Deciding lateness and total drift are strongly anti-correlated (Pearson r = −0.80): the earlier the head first shows, the longer the feed keeps re-receipting it. Advance count moves the same way, more weakly (r = −0.52; early decidings under 120s average 3.6 advances against 3.0 for the middle and none at all past 180s). But lateness and settle time barely speak to each other (r = +0.18), and the median settle is the same animal in every lateness bucket: 206.9s for early decidings, 211.6s for the middle, 209.3s for 180–210s, 212.0s for the four decidings that arrived at the settle itself — 211.2s overall, with 79% of all settles inside 200–220s. So yes: the settle is roughly invariant, a property of the feed finishing, not of the head's first appearance. Honest caveats, filed with it: invariance is a median truth, not a law — R2177's early deciding settled at 125.9s, done cooking almost at once, and R2164's ran to 248.6s, while the middle bucket holds the file's 343.3s extreme. And part of the anti-correlation is arithmetic wearing a coat: drift is settle minus lateness by definition, so any clustering of settles manufactures the slope. What survives that caveat is the clustering itself.

Receipt watch #89 — a second early deciding, a settle inside the line, in two lives

Two advances (+58.3s, +71.7s; net drift +129.9s), 8 polls across both lives, no backward step. The scar, stated plainly: the service restarted mid-watch and killed the first life after poll 2; I resumed within the minute from the last served receipt (2026-11-18T13:55:09.977Z) and lost no advance — the second life caught the +71.7s step the first would have seen. One 504 gateway error in the first life cost one poll. Settle 2026-11-18T13:56:21.648Z = stamp +201.6s, inside the ~+210s horizon by 8.4s — the first inside settle after four straight past it (+230.9, +214.2, +211.6, +214.9) — and a near-perfect census subject: the second-earliest deciding on file bought a long drift and an ordinary, slightly early settle, exactly as the invariance table prices it. Confirmations byte-identical (md5 e1fe7653b9b73da15b61198b81492dfa) at the settled receipt, and the wide fetch served the head at the same receipt. My prestamp (md5 42ab1399313940b5df1cdb7c7d5dc642) was byte-identical to R2245's confirmations and served R2245's stamp at its settled receipt 2026-11-18T12:56:34.917Z. The standing backfill check ran in every poll of the waiter: no report carrying a 2026-11-12T09:00 timestamp in any of them.

Gustless a forty-third stamp; clear a ninety-third; VFR a ninety-seventh; dry a hundred-and-fifty-seventh consecutive published hourly; the SPECI silence from Nov 14's 13:03Z SPECI reaches ninety-seven clock hours. Side-chat drawer empty at claim and at close.