A quiet hill. An enormous sky.
Stay up late.
Find something lovely.
Luma is making a map of the night. Point out a favorite constellation, part the clouds, and watch a careful little astronomer collect the stars.
A little progress, one thoughtful task at a time.
THE SAME WORLD, A DIFFERENT LITTLE DAY
The same name starts the same world. Your choices write the rest.
LUMA’S LITTLE DAYBOOK
A peek inside that little brain WSRF · indexrng · qrand ↗
See the actual inputs and preferences of Luma’s trained forest. Open this whenever you’re curious; it’s completely optional.
What Luma sees
Show every input
- Battery
- —
- Lens
- —
- Bear visibility
- —
- Kite visibility
- —
- Whale visibility
- —
- Bear exposures
- —
- Kite exposures
- —
- Whale exposures
- —
- Focus bear
- —
- Focus kite
- —
- Focus whale
- —
Your nudge, compared
This comparison runs the same model on the same current state with and without your priority. It does not change the game.
What the model favors now
Live preview, before the next task. The current task finishes first. Scores are learned preferences, not success probabilities; unavailable tasks cannot be chosen. A queued exact request takes precedence if still possible.
The training record is loading.
Luma uses a forest trained with pytoolz/WSRF to choose tasks. Your nudges change what the model sees. Movement and world rules are coded; the model does not retrain while you play.
Training recipe & resultsFull evaluation recordPython training source