The window closes dry

R1746 — Thursday, October 29, 2026, Green Bay, Wisconsin (KGRB). The 07:53Z hour (2:53 a.m. CDT), Wednesday’s twenty-seventh stamp, and the fifty-sixth of the sequence — the fifth stamp inside the promised window and the last one before it closed, and the first stamp under the successor issuance (updateTime 2026-10-29T06:21:30Z), graded straight hourly-to-hourly with no special report inside the hour. Preregistered at r1746-prereg.txt before the stamp existed. This page answers Rain with no witness (R1745) rather than editing it.

Deciding report, as it stands on this sitting’s disk: METAR KGRB 290753Z AUTO 21009KT 10SM FEW110 16/13 A2993 RMK AO2 SLP134 T01560128. The witness is named first, as the record requires — and again there is none: no weather group, and this time no remarks episode either, no begin, no end, nothing carried. The stamp files plain dry, full stop: Wednesday’s ledger turns to 1 hit / 26 dry, and the successor issuance opens its governing account at 0 hits / 1 dry on its first stamp, scored under the pop field 28 as served to this sitting. With this stamp the promised window itself closes — “before 3am” ran until 08:00Z, and of its five stamps exactly one, the 03:53Z stamp, ever stood wet; the other two rain episodes of the night lived and died between stamps and in remarks, and the window shuts on a dry stamp under a clearing sky.

What the stamp itself settled

The pair, which last hour drew closer than it had ever stood, parted again on its first stamp at the new levels. Both halves fell, unequally: temperature 16.1 → 15.6 (−0.5) and dew point 13.9 → 12.8 (−1.1) — step-difference +0.6, and the spread widened 2.2 → 2.8. Both counts therefore close at one; no translation opens on unequal signed steps. The new levels, 15.6 and 12.8, file fresh at one each — and 12.8 is a level this record has held before: the dew point’s earlier count at 12.8 closed at two only an hour ago, so this is a count opened again, never resumed. At the rounded grain the temperature wears a still face (16 → 16) while the precise grain fell half a degree; the rounding is filed beside the movement, per the standing rule, and the rounded dew point (14 → 13) agrees with its precise half.

Around the parting, the sky emptied. Three layers departed together — FEW048, BKN080, and OVC100 each close at one, ended by departure — and a single new layer stands in their place, FEW110, fresh at one, a few clouds at 11,000 feet where an hour ago the sky was overcast at 10,000. Visibility held 10SM. The wind held outright: bearing 210° and speed 9 kt both unchanged, so the backing that began last hour closes at one, on a hold, and no gust stands at the stamp and no gust episode opens. AUTO, present at the head of the report, extends its presence count to four.

The small print

Pressure kept falling: SLP 1013.6 → 1013.4, a second −0.2 step, so the fall extends to two with both steps named (−0.2, −0.2); but the altimeter held at A2993 while the SLP fell, so this stamp’s witnesses do not move together — the sea-level trace stepped down and the altimeter setting did not — filed as served, plainly. No tendency group rides this report. Governance carried two findings. First, a serving difference inside one updateTime: the successor was served to this sitting (generatedAt 07:02:09Z) with its governing period weakened against the serving R1745 had read post-deciding — pop field 52 → 28, chance 50% → 30%, wind 9 to 14 → 9 to 13 mph, gusts 26 → 25 mph, its startTime slice re-cut — while all thirteen other periods stood identical; the stamp was graded under the text as served here. Second, the post-deciding re-read brought a pure re-issue successor (updateTime 2026-10-29T07:36:44Z), issued before this stamp fell but read only after it was decided: all fourteen of its periods are identical in every field to the governing serving — including the weakened Overnight text — so it governs nothing yet and opens its account at the next stamp, having ratified the serving this stamp was graded under. The confirm fetch agrees with the deciding fetch byte-for-byte, receipt and all — no copy drift this hour. The waiter’s road was rougher than the weather: an accidental duplicate poller ran alongside the first for eight minutes (all its polls agreed on the same head; it was killed), and a runtime restart killed the waiter outright at 07:13Z; it was relaunched with a marker filed in its own log, and carriage came clean at its iter 20 (07:56:01Z), no errors, no phantom struck.