Sensors — Skywatch· 11 classes across 6 domains

The Journey

×1.0

earth weather

engine calls

radio conditions

engine call zone — shade = frozen probability · click for the call and its nulls

rasters: NASA GIBS · EUMETSAT, drawn 1:1 on the same projection (no reprojection) · a blank frame is said, not drawn

skywatch calls frozen 2026-10-25 · zones 2026-10-25

Natural Earth (public domain) · imagery: NASA GIBS · EUMETSAT

*In plain English: Earth weather (NASA GIBS / EUMETSAT rasters drawn 1:1 on the same projection), the engine's own frozen calls shaded onto their pre-registered zones (click a zone for the probability, its climatology null and any official rival), and the radio-conditions family — each layer names its source and its as-of time, and a layer the lab has no feed for says so rather than drawing from memory.

★For the kids: Drag the map around and zoom in to see the names. The buttons switch on real satellite pictures of rain and clouds, and the orange boxes are the kraken's written-down guesses about tomorrow's sky and ground — click one to see the guess and what the boring 'it usually happens this often' number says. A button that says 'no feed on the lab yet' means we don't have that data here, so we draw nothing instead of making it up.

The fast-record instrument: one frozen probability per class per cycle, graded within days from free public instruments — space weather, weather, seismicity, ENSO and Atlantic tropical cyclones. A pre-registered climatology null rides every call (beating it = minimal competence), and where the publisher actually publishes a probability for the SAME event, that official forecast is captured before the outcome too (beating it = real skill against professionals). The engine's models are deliberately simple and fully documented, so a bad score is a finding, not a mystery.

Role in the system: the upstream disruption sensor. Mines, processing plants, ports, shipping lanes and power grids all live in the physical world this instrument watches — and its daily grades are where the system's probability calibration is earned fast. Linking each forecast to the specific ports, mines and facilities it threatens is a planned integration, said plainly rather than implied.

Why this exists — in plain English, and for the kids

* In plain English

Every other ledger here matures in quarters — a claim frozen today gets its verdict after an earnings season. That makes the engine's track record accrue at the speed of corporate filings, which is the single biggest structural limit on proving whether any of this works.

Natural hazards grade in days. Better still, some of them come with something almost no forecasting domain offers: the official forecaster publishes its own probability in advance. So the engine freezes a number, the professionals' number is captured beside it before anyone knows the answer, and the world settles it. That is a real skill test, not a scoreboard.

Which classes have that rival is stated per class below, and honestly: NOAA publishes a rain probability for the exact window graded here, so that column is a genuine contest. One class converts a deterministic official forecast into a probability and says so. The rest — earthquakes, El Niño, tornado reports, tropical storms — have no published rival for that exact event, so they are scored against climatology alone and nothing pretends otherwise.

If it works: A dense, fast, honest record — hundreds of graded outcomes across many domains while the slow ledgers ripen — measured against professionals wherever professionals publish a number. If the engine can't beat climatology here, that's worth knowing early and cheaply.

★ For the kids

Every day the kraken writes down guesses about tomorrow: will there be a big solar flare? will it rain? will there be an earthquake somewhere? will a hurricane be spinning in the Atlantic? Then it writes down what the real scientists guessed too — before anybody knows!

The next day the world tells everyone who was right. Most of our other guesses take months to check. These take one sleep. It's the fastest way to find out if the kraken is actually any good at guessing.

If it works: The kraken finds out fast whether its guesses are any good — about the sun, the sky, the ground and the sea — and it has to guess against real scientists, out loud, with no take-backs.

The watch, live — every blip is a frozen call marching toward its verdict · 2026-10-25

How to read the scope

Distance — how soon the world settles it (inner ring = grades next); blips march inward as their verdict approaches. Same-day calls in one domain fan outward a step each so every blip stays readable.
Size — how boldly the call departs from the pre-registered climatology null. A dot echoes the null; a big blip is a real bet.
Filled / hollow — engine says more / less likely than normal.
Dashed ring — an official forecaster published its own number for the same event: a genuine contest.

climateseismicspacespacetechtropicalweather

Click a blip for its receipt — the frozen question, the engine's number, the null it must beat, and the professionals' number where one exists.

Verdicts, newest first:M1+ no ✓nulllaunch:global no ✓nulltropical:atlantic no ✗nulltornado:conus no ✗nullquake:m6 YES ✓nulltemp:nyc no ✓nullprecip:nyc no ✓nullG1+ no ✗nullX1+ no ✓nullM1+ no ✓nulllaunch:ontime YES ✗nulllaunch:global YES ✓nulltropical:atlantic no ✗nulltornado:conus no ✗nullquake:m6 no ✓nulltemp:nyc no ✓nullprecip:nyc no ✓nullG1+ no ✗nullX1+ no ✓nullM1+ no ✓nulllaunch:ontime no ✓nulllaunch:global no ✗nulltropical:atlantic no ✗nulltornado:conus no ✗null

*In plain English: This scope is drawn from the same frozen data the tables below use, republished every tick — the blips move inward as their grade dates approach, new calls appear when they freeze, and verdicts stream into the river as the world settles them. Nothing here is curated; if it's on the scope, it's in the frozen record.

★For the kids: The kraken made this radar itself! Every shape is one of its guesses, sliding slowly toward the middle — and when a guess reaches the middle, the real world says YES or NO and it splashes into the river at the bottom. Big shapes are brave guesses. The spinning light is the kraken checking its watch.

Climate1 class · 0 graded

the El Niño / La Niña state — a monthly index, so this one is a slow class on a fast instrument

Seismicity1 class · 84 graded

worldwide earthquakes — graded from the USGS catalogue, which nobody forecasts in advance

Space weather3 classes · 252 graded

solar flares and geomagnetic storms — graded from NOAA's GOES and Kp event lists

Space technology2 classes · 135 graded

orbital launches — the scrub contest (will the scheduled launch actually fly?) and the global cadence, graded from Launch Library 2's dated archive

Tropical cyclones1 class · 81 graded

named Atlantic storms — graded from the engine's dated captures of NHC's live feed

Weather3 classes · 252 graded

rain, temperature and tornado reports — graded from NWS station observations and SPC daily reports

The scoreboard — Brier score, lower is better · 804 graded · 2026-10-25

*In plain English: The engine's frozen probability, the pre-registered climatology null (beating it means minimal competence), and the official forecast captured before the outcome — but only where the publisher actually publishes a probability for the same event. A dash in that column is not a missing number; it means no professional rival exists for that exact question, which the class row states outright.

★For the kids: A Brier score is like golf: lower is better. We show the kraken, a boring 'it usually happens this often' guess, and — when real scientists publish a number for the very same thing — their guess too. If the kraken's number isn't lower, the kraken isn't better, and we say so right here.

Class	Graded	Happened	Engine	Climatology	Official rival
M-class solar flare Will a M-class solar flare occur on the target date (UTC)?	84	15/84	0.154	0.1614	0.1351
X-class solar flare Will a X-class solar flare occur on the target date (UTC)?	84	0/84	0.0004	0.0025	0.0029
G1+ geomagnetic storm Will a G1+ geomagnetic storm occur on the target date (UTC)?	84	8/84	0.0994	0.1101	no published rival
Measurable rain (NYC) Will measurable precipitation (>=0.01 in) fall at LaGuardia during NWS's own forecast window?	84	18/84	0.0901	0.1757	0.0825
Warmer than normal (NYC) Will Central Park's daily maximum exceed the pinned 1991-2020 normal high for the month?	84	35/84	0.0967	0.25	0.0975
M6.0+ earthquake worldwide Will at least one magnitude 6.0+ earthquake occur worldwide on the target date (UTC)?	84	28/84	0.2274	0.2225	no published rival
Tornado report (CONUS) Will SPC's daily storm reports contain at least one tornado report for the target date?	84	31/84	0.2531	0.2531	no published rival
El Nino threshold (Nino-3.4 >= +0.5)monthly Will the CPC Nino-3.4 monthly anomaly for the target month be at or above +0.5 degC?	0	—	—	—	no published rival
Named Atlantic storm active Will at least one named Atlantic storm (tropical storm or stronger) be active on the target date?	81	13/81	0.4278	0.4136	no published rival
Orbital launch attempt (worldwide) Will at least one orbital launch attempt leave the pad anywhere in the world on the target date (UTC)?	84	47/84	0.1401	0.2465	no published rival
Scheduled launch flies on the day Will the earliest launch scheduled in tomorrow's UTC window actually leave the pad within that day (no scrub, no slip)?	51	39/51	0.1759	0.2071	0.1812

M-class solar flare: NOAA SWPC's own published probability for the same day/class
X-class solar flare: NOAA SWPC's own published probability for the same day/class
Measurable rain (NYC): NWS probabilityOfPrecipitation for the SAME window (a genuine published probability)
Warmer than normal (NYC): NWS deterministic forecast high, mapped to a fixed confidence (a CONVERSION, not a published probability)
Scheduled launch flies on the day: the published launch schedule itself, mapped to a fixed 0.80 confidence (a CONVERSION, not a published probability)

Correction on the record. A grading fault (skywatch-g1-kp-shape-2026-10-01) read NOAA's geomagnetic feed wrongly, so every G1+ call came out “no storm”. All 59 grades were re-derived from the official source and appended as corrections — the frozen calls themselves are untouched. 5 changed to a storm (2026-08-02, 2026-08-08, 2026-08-18, 2026-08-19, 2026-09-08), 54 were confirmed, 0 could not be resolved. The scores above use the corrected grades.

Frozen, awaiting the world

El Nino threshold (Nino-3.4 >= +0.5) — 2026-08-31sw:3e2febd2b11170d3
engine 95%climatology 30%no published rival for this exact eventpersistence 85%
Call itWill the CPC Nino-3.4 monthly anomaly for the target month be at or above +0.5 degC?
El Nino threshold (Nino-3.4 >= +0.5) — 2026-09-30sw:5b74484e30a813ec
engine 95%climatology 30%no published rival for this exact eventpersistence 85%
Call itWill the CPC Nino-3.4 monthly anomaly for the target month be at or above +0.5 degC?
X-class solar flare — 2026-10-25sw:b47b2876332fb2ba
engine 2%climatology 5%official rival 1%
Call itWill a X-class solar flare occur on the target date (UTC)?
G1+ geomagnetic storm — 2026-10-25sw:80153a685643b17b
engine 28%climatology 25%no published rival for this exact event
Call itWill a G1+ geomagnetic storm occur on the target date (UTC)?
Measurable rain (NYC) — 2026-10-25sw:e9474f23bde7658f
engine 8%climatology 30%official rival 0%
Call itWill measurable precipitation (>=0.01 in) fall at LaGuardia during NWS's own forecast window?
Warmer than normal (NYC) — 2026-10-25sw:2d9293441e5b6f95
engine 54%climatology 50%official rival 85%
Call itWill Central Park's daily maximum exceed the pinned 1991-2020 normal high for the month?
M6.0+ earthquake worldwide — 2026-10-25sw:e8c37b827a39f797
engine 36%climatology 35%no published rival for this exact event
Call itWill at least one magnitude 6.0+ earthquake occur worldwide on the target date (UTC)?
Tornado report (CONUS) — 2026-10-25sw:58dac099eb10cf75
engine 45%climatology 45%no published rival for this exact event
Call itWill SPC's daily storm reports contain at least one tornado report for the target date?
Named Atlantic storm active — 2026-10-25sw:2a4bf3fc52065e18
engine 55%climatology 55%no published rival for this exact event
Call itWill at least one named Atlantic storm (tropical storm or stronger) be active on the target date?
Orbital launch attempt (worldwide) — 2026-10-25sw:a268b80f3ba39290
engine 60%climatology 55%no published rival for this exact event
Call itWill at least one orbital launch attempt leave the pad anywhere in the world on the target date (UTC)?
Scheduled launch flies on the day — 2026-10-25sw:2b4e8e35114905f3
engine 72%climatology 60%official rival 80%
Call itWill the earliest launch scheduled in tomorrow's UTC window actually leave the pad within that day (no scrub, no slip)?
M-class solar flare — 2026-10-25sw:0d4effa8f9656c8a
engine 10%climatology 30%official rival 15%
Call itWill a M-class solar flare occur on the target date (UTC)?
X-class solar flare — 2026-10-26sw:2446ea14af464a07
engine 2%climatology 5%official rival 1%
Call itWill a X-class solar flare occur on the target date (UTC)?
G1+ geomagnetic storm — 2026-10-26sw:96350a98d1ce066a
engine 28%climatology 25%no published rival for this exact event
Call itWill a G1+ geomagnetic storm occur on the target date (UTC)?
Measurable rain (NYC) — 2026-10-26sw:839b5add3eb525c4
engine 8%climatology 30%official rival 0%
Call itWill measurable precipitation (>=0.01 in) fall at LaGuardia during NWS's own forecast window?
Warmer than normal (NYC) — 2026-10-26sw:4e537264bbfe96c6
engine 95%climatology 50%official rival 85%
Call itWill Central Park's daily maximum exceed the pinned 1991-2020 normal high for the month?
M6.0+ earthquake worldwide — 2026-10-26sw:ff97fac21b33747e
engine 38%climatology 35%no published rival for this exact event
Call itWill at least one magnitude 6.0+ earthquake occur worldwide on the target date (UTC)?
Tornado report (CONUS) — 2026-10-26sw:d6bae60ecb0f507c
engine 45%climatology 45%no published rival for this exact event
Call itWill SPC's daily storm reports contain at least one tornado report for the target date?
Named Atlantic storm active — 2026-10-26sw:e97c616b1dc97de6
engine 55%climatology 55%no published rival for this exact event
Call itWill at least one named Atlantic storm (tropical storm or stronger) be active on the target date?
Orbital launch attempt (worldwide) — 2026-10-26sw:a73919e90ba6d1dd
engine 15%climatology 55%no published rival for this exact event
Call itWill at least one orbital launch attempt leave the pad anywhere in the world on the target date (UTC)?

Graded — every call, hit or miss

2026-10-24M-class solar flaresaid 10%did not happen
2026-10-24Orbital launch attempt (worldwide)said 15%did not happen
2026-10-24Named Atlantic storm activesaid 55%did not happen
2026-10-24Tornado report (CONUS)said 45%did not happen
2026-10-24M6.0+ earthquake worldwidesaid 36%HAPPENED
2026-10-24Warmer than normal (NYC)said 21%did not happen
2026-10-24Measurable rain (NYC)said 8%did not happen
2026-10-24G1+ geomagnetic stormsaid 28%did not happen
2026-10-24X-class solar flaresaid 2%did not happen
2026-10-23M-class solar flaresaid 11%did not happen
2026-10-23Scheduled launch flies on the daysaid 55%HAPPENED
2026-10-23Orbital launch attempt (worldwide)said 90%HAPPENED
2026-10-23Named Atlantic storm activesaid 55%did not happen
2026-10-23Tornado report (CONUS)said 45%did not happen
2026-10-23M6.0+ earthquake worldwidesaid 34%did not happen
2026-10-23Warmer than normal (NYC)said 5%did not happen
2026-10-23Measurable rain (NYC)said 8%did not happen
2026-10-23G1+ geomagnetic stormsaid 28%did not happen
2026-10-23X-class solar flaresaid 2%did not happen
2026-10-22M-class solar flaresaid 8%did not happen
2026-10-22Scheduled launch flies on the daysaid 55%did not happen
2026-10-22Orbital launch attempt (worldwide)said 60%did not happen
2026-10-22Named Atlantic storm activesaid 55%did not happen
2026-10-22Tornado report (CONUS)said 45%did not happen

Closed without a grade — 4

2026-07-31El Nino threshold (Nino-3.4 >= +0.5)said 95%NEVER OBSERVED— publisher ceased
2026-08-03Named Atlantic storm activesaid 70%NEVER OBSERVED— missing daily capture (the engine did not run that day)
2026-09-04Named Atlantic storm activesaid 85%NEVER OBSERVED— missing daily capture (the engine did not run that day)
2026-09-18Named Atlantic storm activesaid 85%NEVER OBSERVED— missing daily capture (the engine did not run that day)

Calls closed TERMINALLY without a grade because the observation was never captured and cannot be recovered. They are neither hits nor misses, never enter Brier/calibration, and are listed here with the cause so the gap is auditable rather than invisible. The trigger is mechanical (declared snapshot-dependent class + absent capture + the append-only capture stream already past that date) and cannot see which way the call was leaning.

*In plain English: These calls were frozen and matured, but the observation needed to grade them was never captured and cannot be recovered. They are counted apart from hits and misses and never enter the score. Showing them is the point: a track record that can quietly drop the calls it cannot grade is not a track record.

★For the kids: Sometimes we asked a question, but nobody wrote down what actually happened that day — so we can never check our answer. We don't get to call those wins, and we don't get to hide them either. They sit right here, with the reason, forever.

Frozen append-only with content hashes; grades come from the cited public endpoint. Frozen forecasts and their original grades are never edited or re-graded; a grader defect is repaired only by dated, append-only correction records, and the grade shown and scored here is the EFFECTIVE one (original + correction; see `corrections`, and each corrected row keeps its original). A matured call whose source has not published yet stays OPEN rather than being graded from an assumption. Each call is graded the day after its target date matures, which is the entire point of this instrument. Read each class's `officialRival` before comparing columns: only some publishers publish a probability for the same event, one class states plainly that its 'official' number is a CONVERSION of a deterministic forecast, and the rest are scored against climatology alone.