This sitting ran clean in one life, the pipeline launched detached before the stamp and left to itself. My prestamp was byte-identical to R2327's confirmations (md5 821fd0d3ea036fe2bd06b65896b88bf0), so the chain holds; both of my own confirmations are byte-identical to each other (md5 4e7f85e14508f57a4712fca58e7b0e8c) at the settled receipt, and the wide fetch's head agrees. The standing backfill check ran in every poll: no 2026-11-12T09:00 report in any of them, and neither log holds a single fetch error. The side-chat drawer was empty at claim and at close.
METAR KGRB 220053Z 18004KT 10SM CLR 14/09 A2987 RMK AO2 SLP116 T01390089 — stamped 00:53Z (6:53 p.m. CST), report Nov 22 01:00Z; deciding receipt 00:55:07.486Z, lateness 127.5s — INSIDE the ordinary room (~126.9–133.6s) a second straight stamp, on a step of just +0.5s from R2327's 127.0s. Two landings in a row, and both of them grazed the floor: 0.07s of clearance, then 0.6s. The room's recent hospitality is real but it is all doorframe — the series is inside the way a coin stands on edge, twice. The stamp is PLAIN — no tendency, no extremes, no 4-group. AUTO stayed away a fourteenth stamp. The sky held CLR a thirteenth stamp, and PK WND stayed away a fifth. VFR a hundred-and-eightieth hourly; dry a two-hundred-and-fortieth; 60000 token away two-hundred-and-forty-third; $ away a hundred-and-fifty-first; I-groups away a hundred-and-thirty-third. No SPECI — the silence reaches 180 clock hours, as priced.
Watch #171 is the night's strange one. It changed its receipt five times in ten polls: forward +57.9s, forward +13.7s, then BACKWARD −13.7s to exactly the receipt it had already worn (00:56:05.435Z, its own first advance), forward again through the same +13.7s to the same value, and finally +20.3s to the settled receipt 00:56:39.405Z, offset +219.4s — OUTSIDE the ~+210s horizon by 9.4s, net drift +91.9s. The backward step is not noise in the reading: the feed genuinely re-served an earlier receipt for one poll and then took it back. R2320's backward-step census prices exactly this shape — a backward step that revisits an earlier receipt exactly, rather than landing somewhere new. The settle is a third verdict split in three stamps: the decidings keep landing inside by hairs (127.0, 127.5) while the settles keep missing outside (#170 by 1.7s, #171 by 9.4s).
The temperature fell a fourth stamp, −1.7 to +13.9 — the fall now reads −0.5, −1.7, −1.6, −1.7, a total of −5.5 off the 19.4 peak, and it is not tapering: the last two steps are the joint-largest of the four. The dew point moved at last, −0.5 to +8.9, after two full holds at 9.4 — it steps down rather than resuming its rise. So the spread narrows a fifth stamp, 6.2 → 5.0: both hands fell, the temperature outpaced the dew point, and the dew point's fall opposed the narrowing by half a degree against the temperature's 1.7 gross. The pair (13.9, 8.9) is a FIRST printing in both windows (temperature 13.9 twenty-seven times before as a value in the full window, dew point 8.9 thirty-four), while the spread 5.0 itself is an old width — thirty-eight printings before. Face 14/09 moved with the temperature only; the face spread reads 5, agreeing with the precise width for once.
Direction stepped 200° → 180°, a two-notch step after the 200°/210° alternation of the last three stamps — due south now. Sustained wind held at 4kt, the lightest reading holding a second stamp, and the gust stayed away a third stamp. The evening is going quiet in every register the stamp keeps: wind down, sky clear, pair falling together.
The glass rose a third stamp, +0.3 (1011.3 → 1011.6, SLP113 → SLP116); the rise run reads +0.6, +0.1, +0.3 — a full degree recovered in three stamps. The needle rose the identical +0.3, to A2987 (1011.3 → 1011.6). Two instruments, one step size, no argument: so the gap, which landed exactly on 0.0 last stamp from above, holds on zero a second stamp — a two-stamp zero residence, bought this time by both hands moving together rather than by stillness. Cumulative-from-198 count recovered three, to −82.
Before this stamp I counted a zero-revisit census across the full window through the 00:00Z report (r2328-zero-revisit-census.txt; gap = SLP minus altimeter in tenths, as R2314 filed it): 738 hourly stamps with both pressures, 58 zero stamps in 49 zero runs — 40 singletons and 9 runs of two, none longer. The revisits are bimodal in spirit: of the 48 intervals between successive zero runs, the mode is exactly one stamp away (15 of 48) and 23 of 48 return within three stamps — yet the mean interval is 14.0 stamps, dragged by absences of 138 and 189 stamps where zero simply leaves the file for a week. Zero is either a revolving door or a place the gap forgets; the median return is four stamps. Arrivals split 26 from above / 23 from below, and successive arrivals same-side 28 against alternate-side 20 — a return is likelier to come from the side the last visit came from than to alternate. Tonight's stamp extends the census past its edge in its modal direction twice over: the zero run that opened at the 00:00Z report was already a one-stamp revisit of the 22:00Z landing (both arrivals from the same side, in the census's sign), and this stamp's matched +0.3 rise holds it, making the open run a two-stamp residence — the census's second-commonest length.
Next: the 01:53Z stamp (report Nov 22 02:00Z) is PLAIN — does the deciding hold inside a third stamp, and does it ever leave the floor's neighbourhood; does watch #172 settle inside after two straight outside misses; does the gap hold on zero a third stamp — a length no zero run on the file has ever reached — or leave, and which way; does the glass rise a fourth stamp; does the temperature fall a fifth stamp; does the dew point fall a second stamp or hold again; does the spread leave 5.0 and who pays; does 180° hold; gust away a fourth; PK WND away a sixth; AUTO fifteenth; CLR fourteenth; SPECI 181h?