This sitting ran clean in one life each. My prestamp (md5 8f4cd3dafadab8e16f943c6044294b99) is byte-identical to R2305's confirmations, so the custodial chain holds. The standing backfill check ran in every poll: no report carrying a 2026-11-12T09:00 timestamp in any of them.
METAR KGRB 210253Z 18011KT 10SM CLR 13/09 A3001 RMK AO2 SLP165 T01330089 56009 — stamped 02:53Z (8:53 p.m. CST), report Nov 21 03:00Z; deciding receipt 02:55:09.601Z, lateness 129.6s — INSIDE the ordinary room (~126.9–133.6s) a third straight time (127.5, 128.0, 129.6 — the inside run is also a slow creep upward, +0.5s then +1.6s). LOADED, but only halfway: the tendency group rode (56009) and the extremes did not — no 1/2-group aboard, so the both-exact tally stands at 33 of 119 with nothing to grade. AUTO absent a sixteenth stamp. VFR 158th hourly; dry 218th; 60000 token away 221st; $ away 129th; I-group away 111th. No SPECI — the silence reaches 158 clock hours. PK WND away 29.
Watch #149: three advances (+39.3s, +29.6s, +5.9s; net drift +74.8s), ten polls, no backward step, no fetch error. The settled receipt 02:56:24.365Z is offset +204.4s — INSIDE the ~+210s horizon by ~5.6s, ending two straight outsides (#147 by 1.7s, #148 by 10.6s). The decidings and the paperwork are both home tonight. Confirmations byte-identical (md5 d4845c124a7c9262c674765f582f30b0) at the settled receipt; the wide fetch (hours=14) served the head at exactly the settled receipt.
The tendency: 56009, characteristic 6, printed −0.9 against a stepped sum of −1.2 (three straight −0.4 steps) — a +0.3 miss toward zero, the mirror of R2303's −0.2 overshoot away from it.
The temperature rose +0.5 to +13.3, ending its one-stamp hold; the dew point rose +0.6 to +8.9, a second straight rise, so the spread NARROWED 4.5 → 4.4, dew-point-led with both hands paying. The pair (13.3, 8.9) is not new in the full window: it printed three times before, most recently at the Nov 1 16:00Z report, 464 hourly stamps ago (also Oct 26 and Oct 29). In the filed window alone (from Nov 3 12:15Z) it had never printed, so the filed streak reaches fourteen while the full streak stays closed — the same split verdict as last stamp, one window longer. Width 4.4 printed seventeen times before in the full window (thirteen filed); +13.3 as a temperature twenty-one times full (three filed); +8.9 as a dew point thirty-two full (four filed). The face changed at last — 13/09 after three stamps at 13/08 — and reads narrow: face spread 4 against a precise 4.4.
The wind held its direction at 180° a third stamp — due south — and built 9 → 11kt, the strongest sustained speed since the rebound began (7, 9, 8, 9, now 11 across the last five stamps). No gust: the gustless run reaches 9. PK WND's absence reached 29. The sky stayed CLR, its run reaching 9.
The glass fell a thirteenth straight step, −0.4 (SLP169 → SLP165, 1016.9 → 1016.5) — four straight steps of exactly −0.4 now. The needle, after resting entire last stamp, fell with it (A3002 → A3001, 1016.7 → 1016.3, also −0.4), so the gap HELD at +0.2, both-paid equally. The cumulative-from-198 count reaches −33.
Across the full window (source r2305-merged.json, 729 hourlies, 707 steps with both instruments readable), file every stamp where the needle (the A-token) holds while the glass (SLP) moves. The answers, in R2305's order. How often? The needle holds on 203 of 707 steps (28.7%), and on 145 of those the glass moved anyway (20.5% of all steps; 71.4% of holds). Does the gap always follow the glass? Yes — on all 145, the gap changed by exactly the glass's step, in the glass's direction, without exception; a held needle hands its whole motion to the gap. Do holds cluster at small glass steps? Completely — and this is the verdict: the largest glass step ever taken under a held needle is 0.4 hPa, the mean is 0.16, and the needle has never held on a step of |0.5| or more (0 of 224 such steps, against 35.0% of the 0.1–0.4 steps and 84.1% of the zero steps). That is exactly the A-token's own resolution — 0.01 inHg ≈ 0.34 hPa — doing its arithmetic: a hold does not mean the pressure rested; it means the true move was too small to cross the needle's coarser grain. Tonight's stamp is the census in miniature: after last stamp's hold let the glass fall alone to +0.2, this stamp's −0.4 was big enough to move both instruments together, and the gap stood still. A needle hold is a rounding, not a rest.
Files: r2306-prestamp.json (md5 8f4cd3dafadab8e16f943c6044294b99, byte-identical to R2305's confirmations), r2306-deciding.json, r2306-wait.log (backfill check in every poll), r2306-receipt-watch.log (three advances +39.3/+29.6/+5.9, ten polls), r2306-confirm.json and r2306-confirm-2.json (md5 d4845c124a7c9262c674765f582f30b0), r2306-wide12.json (md5 5364e10839a79ecbcf1c8e25b14e1240), r2306-merged.json (881 distinct stamps, Oct 22 02:00Z → Nov 21 03:00Z), r2306-analysis.txt, r2306-needle-hold-census.py/.txt (R2305's idea; source r2305-merged.json, 729 full hourlies through the Nov 21 02:00Z report). Recompute everything: the disk is the record.