← LookyBox Arcade

Looky Look Lab — which is better: A or B?

Both pictures are the same scene, drawn by the real Looky engine (the games' own renderer), with one setting changed: a look setting, or in Walk mode a movement setting. Pick the one you like; press About the same if you can't tell. Keys: ← A · → B · ↓ same. Picks are saved on this computer; Export writes a results file Claude can apply to the games.
Drag a picture to spin it, wheel to zoom. Both pictures move together.
A · moving
A
B · press Tab
B
Score 0 ApprenticeYour picks 0Consistency – (hidden "same picture" checks)Ready to apply – settings with 6+ picks and a clear winnersaving on this computer

What it has learned so far

Score = how often an option won, adjusted for what it was up against (a Bradley–Terry model fitted to every pick). Green = current favourite. "Now" marks what the games use today. A favourite means little before about 6 comparisons. Movement settings are only tested in Walk mode.

Current prescription for the engine and the games

The exact constants the games would use if we applied today's favourites. Settings without enough picks keep their current value. tools/apply-look.py results.json --apply writes the shader ones into the engine; the rest are listed for the game builds.


Analysis: upgrade or keep?


One line per setting. UPGRADE = enough picks and the favourite differs from what the games use now. KEEP = enough picks and today's value wins. MORE PICKS = not decided yet. The win counts are plain tallies of every pick.




Needs fixing


Two sources: the automatic geometry audit of every monster (parts found by the leg/arm rule, loose blocks that touch nothing, size) and the flags players raise under a picture (missing part, broken movement, floating blocks, too dark, colours off, clipping).


Rig fix for —:Move a slider, watch the monster re-rig (press Show rig to see the joints). Fixes are saved and sent to the lab.