The globe draws the poles as they are, and nothing while idle

The globe is a UV sphere: towards the poles its triangles narrow to
slivers, and texture coordinates interpolated across them pinched the
map into a star around each pole. Each pixel now computes its own
coordinates from its exact point on the sphere, with seam-safe mipmap
gradients; the map wraps horizontally and gets 8x anisotropic
filtering.

That showed a flaw in the day texture: NASA's mosaic has a lighter
strip at its east edge near the poles, a line from each pole once drawn
exactly. The day texture is now built from NASA's original Blue Marble
(December 2004) by tools/globe/prepare-day-texture.py, which blends
that strip into the map where both sides of the seam agree; same
4096x2048 size, provenance and hashes in SOURCES.md.

globe.gl redrew 60 times a second even when nothing moved; it now draws
only during camera flights, hover pulses, auto-rotation and the
reader's drags, so an idle globe costs no GPU time.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01LsXgEaXee4GCU1hwYgPJXw

tools/globe/prepare-day-texture.py

@@ -0,0 +1,61 @@
+#!/usr/bin/env python3
+# Builds wwwroot/vendor/globe/earth-blue-marble.jpg from NASA's Blue Marble Next Generation (topography and bathymetry,
+# December 2004, public domain), downloaded once:
+#   https://eoimages.gsfc.nasa.gov/images/imagerecords/73000/73909/world.topo.bathy.200412.3x5400x2700.jpg
+# usage: tools/globe/prepare-day-texture.py <that jpg> <output jpg>      (needs Pillow; not part of the build)
+#
+# NASA's mosaic does not wrap cleanly: towards the poles a strip at its east edge is lighter than the map on both sides
+# of the 180° meridian, which the globe draws as a line from each pole. On every row where the map agrees on both sides
+# of the seam but the strip around it does not, the strip is replaced by a blend between the two sides. A real feature
+# on the meridian (Fiji, the Ross Ice Shelf) makes the two sides differ, so its rows are left alone. Within half a degree
+# of each pole a row is a circle a few kilometres across, so it gets one colour, its median, which removes the strip
+# there too. Then the map is scaled to 4096×2048 (the globe's texture size) with Lanczos.
+import sys
+from PIL import Image
+
+BAND = 16      # pixels on each side of the seam that may be repaired
+ANCHOR = 4     # pixels averaged just outside the band, on each side
+AGREE = 24     # the two sides agree when their mean channels differ by less than this
+ODD = 30       # a band pixel is an artefact when it differs from the blend by more than this
+POLAR = 0.5    # degrees from a pole within which a row is one colour
+
+
+def mean(pixels):
+    return tuple(sum(p[i] for p in pixels) / len(pixels) for i in range(3))
+
+
+def distance(a, b):
+    return sum(abs(x - y) for x, y in zip(a, b)) / 3
+
+
+source, target = sys.argv[1], sys.argv[2]
+image = Image.open(source).convert("RGB")
+width, height = image.size
+pixels = image.load()
+repaired = 0
+for y in range(height):
+    west = mean([pixels[width - BAND - 1 - i, y] for i in range(ANCHOR)])
+    east = mean([pixels[BAND + i, y] for i in range(ANCHOR)])
+    if distance(west, east) >= AGREE:
+        continue
+    columns = [width - BAND + i for i in range(BAND)] + list(range(BAND))
+    blend = [tuple(w + (e - w) * (i + 1) / (2 * BAND + 1) for w, e in zip(west, east)) for i in range(2 * BAND)]
+    if max(distance(pixels[x, y], b) for x, b in zip(columns, blend)) <= ODD:
+        continue
+    for x, b in zip(columns, blend):
+        pixels[x, y] = tuple(round(c) for c in b)
+    repaired += 1
+
+polar = 0
+for y in range(height):
+    latitude = 90 - (y + 0.5) * 180 / height
+    if abs(latitude) < 90 - POLAR:
+        continue
+    row = [pixels[x, y] for x in range(width)]
+    median = tuple(sorted(p[i] for p in row)[len(row) // 2] for i in range(3))
+    for x in range(width):
+        pixels[x, y] = median
+    polar += 1
+
+image.resize((4096, 2048), Image.LANCZOS).save(target, "JPEG", quality=90, optimize=True, progressive=True)
+print(f"{repaired} of {height} rows repaired at the seam, {polar} polar rows made one colour; wrote {target}")