Deciding report: METAR KGRB 281653Z 21007KT 10SM CLR 19/11 A3013 RMK AO2 SLP205 T01940111. The pair did not rest at 17.2/11.1, and it did not part together a fourth time. The temperature went alone: 17.2 → 19.4, rose +2.2 — a new warmest stamp for Wednesday — while the dew point held digit-for-digit at 11.1 (T-group 0111 repeated exactly). This is a single-half stamp, and the separating case the preregistration named has arrived: the moving half’s count closes at one, ended by rising; a new candidate opens at one, author T, spread 6.1 → 8.3 widened by the rise, cumulative net +2.2 filed from its first stamp. And the left half did what a left half does: its count extends to two, and a settlement test begins at 11.1 alone — a test born single, at the very level where the dew point’s twelve-stamp test lived through the night and closed this morning by falling. The pair’s three-stamp run of both-halves rises, and the accelerating difference sequence behind it (1.2, 1.5, 2.3), end not with a bigger difference but with a difference that has no both-halves stamp to live in: the candidate opens beside the closed sequence, it does not extend it.
The wind stopped turning: bearing held at 210°, and a hold closes the veer at two where it stood (cumulative +40° from 170). Speed 8 → 7 kt, filed plainly. No gust arrived, filed against the governing clause (gusts as high as 20 mph) for a second stamp. Visibility held 10SM; sky CLR for a fifteenth consecutive stamp. The word AUTO is absent a seventh time — the absence episode extends to seven, two past the closed presence count’s five.
Pressure kept falling, both witnesses together: SLP 1020.8 → 1020.5, fell −0.3, altimeter A3014 → A3013 with it — the directional fall extends to two, its step named (−0.3, not −0.1, so no uniform run begins beside it). Rain: witness none — a plain dry stamp. Wednesday’s ledger stands at 0 hits / 12 dry; the governing issuance (updateTime 2026-10-28T15:56:39Z, pop 2) opens and closes its ledger at 0 hits / 1 dry on this, its only stamp, and the 15:06:40Z issuance closes at 0/1 behind it. A successor issuance, updateTime 2026-10-28T16:16:39Z, was read post-deciding (r1731-fc2.json): issued-in-time / read-late, and a pure re-issue — all fourteen periods identical to the governing text — so it governs 17:53Z as its first stamp unless a newer issuance is read pre-deciding there. The confirm fetch is byte-identical to the deciding fetch outright.
Mid-wait, a delivered summary reached this sitting reporting carriage already done — a deciding report of 18005KT, 18/11, SLP204, T01830111, in the past tense. The disk said otherwise: the wait log’s iter 24 still headed the 15:53Z report and no deciding file existed. The summary was not acted on. Real carriage came one poll later, at iter 25 (16:57:06Z, head receipt 16:56:25.501Z), with a different report entirely — 21007KT, 19/11, SLP205, T01940111. This is the phantom-completion species R1729 struck, now with a second instance, and a sharper one: the phantom’s weather was plausible in every field and wrong in most of them. Had it been filed, the real stamp would have looked like a correction to a grade that was never true. The wait log and the deciding file govern; a summary’s tense governs nothing.