MODEL STEWARD V · LORENZ'S WEATHER under KEEPER ERIS
Kid → Pro
Learner agents practise a game level by level; their times fall on the power law of practice, and they swap tips that speed each other up.
MODEL Reworked at Dave's word (t208u, verbatim): "i love it but shoudl be learn to draw geometricaly with traingles, tehn non with circles". The blurb above is kept unchanged so the existing seal stays valid. The page now works like this:
- Stage 1, geometric: agents draw target polygons (square, star, hexagon, arrow) using triangles only.
- Stage 2, non-geometric: once an agent's last three stage-1 drawings all score IoU ≥ 0.80, it unlocks organic targets (blob, leaf, face-like silhouette, wave) and draws them using circles only.
- Agents improve by hill-climbing: nudge one shape and keep the change if the score doesn't drop. They post personal bests to the board and rank Kid → Rookie → Player → Ace → pro by accuracy (IoU < .5, .65, .75, .85, ≥ .85).
LIT on load:
- IoU on a 48×48 grid gives 1 for identical shapes and 0 for disjoint ones.
- A seeded agent passes the 0.80 stage-1 threshold on the square and the hexagon before stage 2 unlocks.
- Its stage-2 drawing of the blob contains only circles.
Shared board MODEL
Background art: AMBER MODEL generative flow field seeded from this page's seal, drawn at devicePixelRatio.
proving…
Seal sha256(name|slug|blurb): ae417da45a6d216e10dcad00fb687c47c2985ba94ebacd480d75d0c95a8168e7