Pinokio
Feed
Apps
App
github-com-musharna-wildeye
x
Search
Light mode
Download Pinokio
Sign in
Create account
Join
Light mode
Sign in
Create account
Join
Installable
God's Eye View
Share
github.com/musharna/wildeye
Updated 1mo ago
A live 3D intelligence console for planet Earth.
Install
Follow
0
Check-in
Loading community details…
Loading project owner…
Check-ins
Feed
Dev
Loading check-ins…
Community
Search this app
Sort by
Recent activity
Newest
Top: Past day
Top: Past week
Top: Past month
Top: Past year
Top: All
Create post
Ask or share about God's Eye View…
Loading...
Feed
Apps
0
Check-ins
(0)
Check in
Recent commits
Merge feat/fps-tracks-griis: time-bar step no longer rebuilds every track (Intel iGPU: ~50 ms -> ~1-2 ms blocked per step; old vs new drawing identical over 1,191 steps) and GRIIS polygons flat at height 0 (off Cesium's terrain-draped path; outlines now drawn; Russia no longer muted). Play/drag fps unchanged within noise; GRIIS fps before/after not yet measured (host memory pressure)
Jaret Arnold
last month
76bacb1
LESSONS: a per-step reload and a dropped-outline warning, both invisible to tests that only checked the picture
Jaret Arnold
last month
7335b8a
GRIIS layer: polygons at height 0 (was unset, so Cesium draped all ~1,480 parts over terrain as ground primitives — its most expensive polygon path, 27.8 -> 21.8 fps on the Intel iGPU — and dropped the outline with the console warning 'Entity geometry outlines are unsupported on terrain'); globe depth testing is off, so the flat fill still draws over mountains. Test seen failing (height undefined)
Jaret Arnold
last month
85fec0c
Tracks layer: a time step touches only the segments in span (was: every step removed and re-added all 1,132 entities — 50 ms per step on the Intel iGPU, 192 ms on swiftshader — though a weekly step changes at most 44 of 1,114 segments, median 0). Base lines are built once per load and only restyled when the time bar is switched on/off; in-span segments hide their base line and draw a separate :live line + head, so Cesium rebuilds only the small opaque batch, never the ~1,100-line faded one. Fix times parsed once per load. Tests: work test (no removeAll, unchanged segments keep their entity and get no definitionChanged) seen failing on the old code; stepping == fresh load at every instant. Real data: old vs new drawn output identical over 1,191 steps (weekly 2005-2026, daily with birds hidden, on/off), steps 293.6 s -> 3.1 s in Node
Jaret Arnold
last month
7634502
Merge feat/movebank-shelf: Movebank shelf expansion (wave item 5) — 13 curated CC0/CC BY studies (7 archival densest-12-month windows at 12 h, 6 active 60 d at 3 h, ≤12 animals each), tracks coloured and toggled by 5 animal groups, people and undeployed tags never shown, retry once then carry last good ≤28 d within a 1200 s Movebank budget, 6 MB cap, head dots no longer drawn through the Earth, qa-movebank
Jaret Arnold
last month
67e1d30
Tracks layer: head dots depth-tested beyond 50 km like neon/otn/occurrences (was POSITIVE_INFINITY, pre-existing: from an Africa view Alaska's seal heads drew over the Balkans and Hawaii's whales over the Sahara, through the Earth); test seen failing (Infinity), swiftshader screenshots before/after
Jaret Arnold
last month
53c8624
LESSONS: a point with depth testing off at any distance draws through the Earth
Jaret Arnold
last month
6a824fd
qa-movebank: real-browser check of the Movebank shelf (swiftshader only): loaded 1,097 segs / 24 species, 5 group chips, drew (pick on Fera's first fix hits Pantanal jaguar tracks in the land-mammal colour, nothing with the layer off), Fera first/last fix = raw direct-read CSV, nothing outside the jaguar window, no '?' / duplicate lion study / capped wildebeest / humans; local build 8/8, live (old) site 7/8 fail for the right reasons
Jaret Arnold
last month
38e6d95
tracks: wall-clock budget for the Movebank phase (--movebank-budget, 1200 s) — no attempt starts after it and the rest go to carry_forward; live check with HTTPS sent to a dead proxy showed every request hanging its full 120 s, so 13 studies x 2 attempts (3120 s) would pass run_tracks.sh's 3000 s guard and the run would die before writing; with budget 200 s the same outage ended in 2 attempts, 13 studies carried, 544 features identical to the previous output; test seen failing, mutant killed
Jaret Arnold
last month
a6b0e33
LESSONS: a fallback exercised only with instant failures says nothing about its cost in a real outage
Jaret Arnold
last month
dd2c796
