Eclipse 2026 Operations Runbook

Operations runbook — August 12, 2026

The day, in UT

UT	Event
15:34:11	First partial contact anywhere (global P1)
16:43:40	C1 Látrabjarg — the partial phase opens over Iceland
16:58:05	Totality begins on the path (global)
~17:06	The umbra crosses near the North Pole (east→west leg)
17:45:34	Maximum, Látrabjarg (totality)
17:45:54	Greatest eclipse — 65°13.5′N 25°13.7′W, 2m18.2s, Sun +25.8°
17:48:46	Maximum, Reykjavik (totality)
17:53:59 / 17:55:20	Maximum at Millstone (15.9%) / Norwich CT (13.2%) — the control channel
18:28:18 / 18:29:11 / 18:29:43	Maximum, A Coruña / León / Zaragoza (totality, Sun low)
18:34:05	Totality ends on the path (global)
19:02:30	C4 Zaragoza — near local sunset
19:57:56	Last partial contact anywhere (global P4)

(Station rows are the substrate's own sealed geometry — see the prediction registry.)

The live loop — eclipse-watch

Every pass (default 60 s): (1) re-prove the sealed eclipse Form from disk and refuse the pass on drift (LAW 3 live-path governance); (2) capture NOAA SWPC 1-minute estimated Kp into the ledger and evaluate the storm gate (est Kp ≥ 5 → loud pivot flag; capture continues); (3) capture the USGS observatory H series for the current UT day; (4) inside 15:34–19:58 UT on the day, log the eclipse-window capture state; otherwise print the countdown.

gaiaftcl eclipse-watch --once            # single governed pass (smoke test)
gaiaftcl eclipse-watch --poll-seconds 60 # the 24×7 loop

Current ledger-borne state:

• Latest SWPC estimated Kp in ledger: 0.33 at 2026-08-13 10:41:00 UT
• Latest GFZ definitive/nowcast Kp day in ledger: 2026-08-12
• Latest magnetometer capture: BOU H at 2026-08-13 10:40:00 UT

Ledger-borne (renders identically until the next ingest).

What the live loop is and is not: it is the eclipse-day capture and the storm gate. It is not the verdict — Madrigal's definitive gridded TEC lands T+1, and the resolution crucible is the judgment. Real-time TEC (NTRIP/GNSS streams) is Stage-4 territory with its own registrations.

T+1 — the resolution

The exact command chain lives in the prediction registry. Sequence: ingest eclipse day + the three pre-registered control days (2026-08-05, 2026-08-19, 2026-07-16 — each must pass Kp ≤ 4 live or is excluded and reported) → build the 2026 fossils → eclipse-invariant --historical --seal --save → eclipse-wiki-render. The wiki's crucible tables become the public readout of whether the sealed claim held.

Standing maintenance (now → eclipse week)

• eclipse-wiki-render --check after any ingest: exit 1 = the wiki drifted from the ledger → run the renderer.
• invariant-verify --game eclipse and --path …/eclipse-prediction-2026-08-12.invariant.json at will — both must exit 0; any drift is a stop-everything signal.
• Optional context captures any day: helio-ingest kp --day <D>, helio-ingest usgs-geomag --day <D>.

Handoffs (outside the eclipse lane — stated, not performed)

1. Daemonizing the live loop: the plist is staged at cells/launchd/com.gaiaftcl.franklin.eclipse-watch.plist; bootstrapping it requires deploying the shared CLI binary to the fleet path — Games-1/2-stream territory. Until then the loop runs manually from the repo build binary.
2. Publishing this wiki: scripts/publish_wiki.sh pushes to the public GitHub wiki remote — founder's action.
3. Stage-4 registrations (Earthdata/CDDIS 1 Hz RINEX, ESA EO/Swarm, SuperMAG, GAGE NTRIP) — founder account signups; none block current operations.

Resolution status — measured T+1 (2026-08-13)

The sky kept the appointment; the archive has not yet. Measured on resolution day:

• Eclipse day was quiet. GFZ Kp for 2026-08-12: eight 3-hour bins, maximum Kp 2.667 — far under the Kp ≤ 4 quiet rule and the pre-registered est-Kp ≥ 5 storm gate. The storm flag never raised; the prediction grades at full strength, with no storm annotation. The flanking days measured quiet too: 2026-08-05 max Kp 1.667, 2026-08-11 max Kp 3.000.
• The response archive trails real time. Madrigal's world-gridded TEC (instrument 8000) has no experiment for 2026-08-12 yet — measured today, the archive's posting boundary sits between 2026-07-25 and 2026-08-01, a two-to-three-week lag. The sealed claims cannot be graded until the archive posts the day: grading on absent data would be a verdict about an empty query, not about the sky.
• The instrument is ready and unchanged. All four sealed Forms re-verified byte-for-byte today, 28 days after sealing; the crucible re-ran green (2024 benchmark 7/7, 0 false positives) and the shear discriminant re-derived its frozen threshold exactly (minimum eclipse excess 57,543 ppm vs storm floor 42,836 ppm — threshold 50,189 ppm standing).
• Expected grading window: the day Madrigal posts 2026-08-12 — at the measured lag, around 2026-08-30, by which time the third pre-registered control day (2026-08-19) posts alongside it, so the resolution runs exactly as sealed. A standing daily probe watches for the posting.

Nothing above changes a sealed claim; the fingerprints are untouched. A pre-registered experiment should say out loud when its grading data has not arrived — the absence of data is a measurement too.