A little light. A ribbon of possibilities.
Follow a little light.
Leave room to grow.
Wisp is gathering glowing seed pods in a quiet garden. Its ribbon grows with every discovery. Suggest a route, leave room to turn, and watch a little plan unfold.
Select a square, then a digit. Entries need a verified single in a row, column, or box.
Your next move
Choose any available move. Tapping a place on the board narrows the list.
A DAY TO COME BACK TO
The same name begins the same challenge. Your decisions tell the rest.
THE DECISION DAYBOOK
A peek inside that little brain PYTOOLZ · WSRF ↗
This is the fitted model’s actual view of the current decision. It stays optional, and you can pause to explore it.
What it favors
These numbers estimate the training objective for an action. They are not probabilities. The bars show relative preference among currently available actions.
Your nudge, compared
Early and current models
What this move looks like to the model
Follow one tree’s actual branches
This is one tree’s path. The final estimate averages every tree in the active zone.
Imagine this move
Training happens in Python with pytoolz/WSRF. Your browser runs the exported forest; it does not retrain while you play. World rules and verified puzzle deductions remain explicit.
Training recipe & measured results · Full evaluation record · Python training source
Revisit a decision. Try another possibility.
Scrub through the board before each move. Continuing from a past decision replaces the later moves in your current game.