Worldview

USGS earthquakes (past hour): no data yet

Automatic, unreviewed solutions: magnitude and epicentre are revised, sometimes substantially, and events are occasionally withdrawn entirely. The summary feed publishes NO horizontal error, so every epicentre here carries a declared 5 km floor rather than a measurement (USGS_EPICENTRE_UNCERTAINTY_FLOOR_M). Detection completeness is a function of instrument density: M2 events are catalogued in California and invisible in the South Atlantic, so a sparse region means sparse SEISMOMETERS, never sparse earthquakes.

NASA FIRMS active fire detections (VIIRS S-NPP NRT): key presence not checked (FIRMS_MAP_KEY)

A thermal anomaly is not a fire. VIIRS flags ~375 m pixels that are hot: gas flares, industrial furnaces, volcanic vents and controlled burns all appear, and a real fire under cloud or between overpasses does not. Coverage is polar-orbit sampling — typically a few passes a day at mid latitudes — so absence between passes is absence of OBSERVATION. `confidence` is carried through verbatim as the source spells it (low/nominal/high) and is not collapsed into a number.

adsb.lol aircraft (ADS-B): no data yet

Volunteer receivers. Coverage is a map of where people run antennas, not of where aeroplanes are: dense over western Europe and North America, thin over Africa and central Asia, and essentially absent over open ocean, where an aircraft in level flight simply is not there. Aircraft without a transponder, with it switched off, or filtered upstream do not appear. `alt_baro` is a PRESSURE altitude referenced to 1013.25 hPa, not a height above the ellipsoid.

Open-Meteo surface conditions: no data yet

A numerical weather MODEL's nowcast, not a station observation — so every value is graded `estimated`, however current it is. Coverage is THE SAMPLED POINTS AND NOWHERE ELSE — this surface’s anchors plus whatever a caller added through `extraPoints` — and the space between them is not interpolated, because this package will not manufacture a value it was not given. The count is deliberately not written here: it is `weatherPointsFor(req).length`, and a number in this sentence would go stale the first time a deployment sampled one more place. Every value is reported at a model GRID node, up to several kilometres from the point it was requested for.

CommandAGI live threads: no data yet

PUBLIC threads only, and only the ones whose owner attached an origin pin. A pin is a place the owner NAMED, not a measured fix, so it carries a declared 500 m floor rather than an accuracy; a thread with no pin is invisible here and is not absent from the platform. This is where work was STARTED from, never where a cloud VM physically runs.

CommandAGI compute regions: no data yet

Every value here is graded `estimated` and the reason is structural: `Region` is a STRING in `packages/core/src/types.ts` and the platform holds no region→coordinate table, so the position is the vendor’s published metro location with a 60 km extent — a cloud region is several availability zones across a metropolitan area, not a point anybody surveyed. Free CAPACITY is not carried: there is no DB-level capacity read to project, so this layer says WHERE the platform runs and never HOW MUCH is left.