Boundaries. Every plate boundary on Earth, from the Bird (2003) PB2002 model, classified by kinematics: divergent (spreading ridges and continental rifts), convergent (subduction zones and collision fronts) and transform (strike-slip). Convergent boundaries carry teeth on the overriding plate, so the barbs tell you which way the slab dips; that polarity is read from the model rather than assigned by hand. Transform faults carry offset half-arrows showing slip sense, taken from the sign of the modelled right-lateral velocity. Ridges carry opposing double ticks. Dashed lines mark diffuse boundaries, where deformation is spread across a wide zone instead of one fault trace — they keep their symbols, because a diffuse subduction zone is still a subduction zone.
Plates. The 52 plates of PB2002, drawn unsimplified so they still tile the globe exactly — every point on Earth falls inside one plate and no point falls inside two. Areas are computed from the polygons by spherical excess and sum to Earth's surface, which is the check that they are right.
Accuracy. Boundaries follow PB2002 and are simplified to about 1 km, which is below what any zoom here resolves; the source is densified to ~700 m, so almost nothing is lost. The coastline is Natural Earth 1:50m simplified to about 3 km, and is context only — islands under 400 km² are not drawn. This is a global-scale reference, not a survey product: do not measure distances off it, and do not use it to site anything.
Plate motion. Arrows show approximate absolute plate velocity in a hotspot reference frame, in mm/yr. These are hand-assigned, order-of-magnitude figures for the plate interior; real motion varies across a plate and near boundaries. Only the 24 major plates carry one. PB2002 gives relative motion across boundaries, not absolute plate motion, so the remaining 28 plates show no arrow rather than a derived guess.
The info panel. Click a plate or a boundary to fill it and open a panel. Everything numeric there is computed from the data: area and share of Earth by spherical excess, boundary composition and perimeter by summing the chains that bound the plate, neighbours by shared boundary. The prose descriptions are different in kind — they are editorial, written by hand for the 17 largest plates and not derived from PB2002. They are labelled as such wherever they appear, and the remaining 35 plates show computed figures only rather than filler. The seismicity section is deliberately empty until the earthquake catalogue lands.
Volcanoes. The Smithsonian Global Volcanism Program's Holocene list — 1,214 volcanoes with an eruption in roughly the last 12,000 years. They are grouped by last eruption year, and deliberately not as active, dormant or extinct: those words have no agreed definition, GVP itself avoids them, and 366 of these volcanoes have no dated eruption at all, so any such label would be invented rather than recorded. Three marks carry five bands — solid for 1950 onward, dimmed for 1800–1949 and for eruptions within the last 2,000 years, hollow for BCE-only dates and for undated. The exact year is in the info panel, because a hollow triangle cannot tell you the difference between "last erupted in 6000 BCE" and "nobody has dated it".
Country borders. Natural Earth 1:50m, simplified to about 3 km, drawn as a hairline so they never compete with a plate boundary. Unlike everything else on this map, these are contested. Natural Earth publishes per-country viewpoint variants precisely because states disagree about where some of these lines are — Kashmir, Crimea and Western Sahara among them — and this uses the default de facto rendering. Read a disputed line as one view among several, not as an adjudication. If a border matters to your question, this is not the map to settle it with.
Active faults. The GEM Global Active Faults database — 13,696 traces, off by default. A fault is a fracture where crust has moved; an active fault has moved recently enough to be expected to move again. The important point is that most active faults are not plate boundaries: the Alps, Tibet, the Apennines and the Basin and Range are all deformation spread across many faults inside a plate, and the PB2002 model above shows none of them, because it models plate edges only. Every trace is drawn, named or not. Rather than filtering short faults away, line weight and opacity scale with trace length and zoom, so long structures carry the global view and short ones stay present but faint until you come closer. They are drawn in a single colour, under the boundaries: GEM classifies each trace by slip type, but reusing the divergent/convergent/transform colours here would make every one of these read as a plate boundary, so kinematics is stated in words in the panel instead.
What a blank region does not mean. Fault coverage tracks national mapping effort at least as much as it tracks tectonics. The clearest case is the United States: the western states contribute 655 traces, almost all from one detailed California compilation, and the entire eastern half contributes none — including the New Madrid seismic zone, which produced several magnitude-7 earthquakes in 1811–12 and is the most significant intraplate seismic zone in North America. Its faults are real, well studied, and simply not in this database. Elsewhere the same pattern: Japan 643 traces, the whole of West Africa 2. Naming is more uneven still — 494 of California's 498 traces carry a name, and none of the 184 in the Alps, 209 in the Caucasus or 453 in Tibet do. So an empty area here may mean there are no active faults, or that nobody has contributed a catalogue, and this map cannot tell you which. Treat it as a record of what has been compiled, not a census of what exists.
Hotspots. Present-day surface expressions of long-lived melting anomalies. Their deep origin is genuinely contested — some are plume-fed, others may be shallow or controlled by cracks in the plate above. Position is the current volcanic centre, not the track. This is a hand-kept editorial list of 42, not a measured catalogue, and it carries only a name and a position; everything else in a hotspot's panel is computed from data already on the map — the plate it sits on, the distance to the nearest plate boundary, the nearest volcano. That distance is the one worth looking at, because it is the evidence behind the word intraplate: Hawaii sits about 3,500 km from any plate edge, while Iceland sits directly on one, and those are two quite different things to call by the same name.
Why these two views. The flat default is Equal Earth centred on 150°E: boundaries are a global network, so relative area has to survive (Mercator inflates the Arctic ridges and shrinks the equatorial Pacific), and the seam has to fall somewhere. Centring on Greenwich cuts the Pacific plate in half and destroys the Ring of Fire; at 150°E the seam lands in the Atlantic, splitting the Mid-Atlantic Ridge and the small Scotia arc instead. Nothing is costless — this is the cheapest cut available, and the selector lets you check that yourself. The globe is orthographic: the view from infinite distance, no seam anywhere, no distortion at the centre of the disc. Geometry is drawn as vectors rather than a texture, so it stays sharp at any zoom.
Sources. Plate boundaries and plates: Bird, P. (2003), An updated digital model of plate boundaries, PB2002. Coastline and country borders: Natural Earth 1:50m, public domain. Volcanoes: Smithsonian Institution, Global Volcanism Program, Volcanoes of the World. Active faults: GEM Global Active Faults database, GEM Foundation, licensed CC BY-SA 4.0 — the simplified geometry drawn here is a derivative of it and carries the same licence.