System 1 and System 2 on a game: the Pokémon world map

A showcase of solvi.core.dispatch with a searching slow path and solvi.core.knowledge.worldmap; the player and the replay viewer are in spaces/pokemon/.

examples/23_pokemon_world_map.py puts the pieces together on the world map of Pokémon Red: 190 places and 447 exits recorded from a real playthrough (place names and exits as a player sees them — "Leave north", "Door at (12,11)" — no ROM bytes, no graphics), and the game's first fifteen goals, each naming the place it needs and never the way there. One decision is "which exit do I take here?"; its input is a plain view — the exits on offer as text, the route System 1 remembers, the plans the world map gives — so every decision replays.

• System 1 is a catalog of two rules (the remembered route's exit; the only exit there is) asked for a span of the exits on offer, so an answer that is not on the screen is refused. It abstains when it remembers no route, when the remembered exit is not on offer, or when the last step surprised it.
• System 2 is SearchPath(system2, space=..., into="plan", search={"objective": "value"}): the known way to the goal when the player's WorldMap has one, else every unexplored exit within reach, scored by expected value, behind a hard check that the plan's first step is on offer. An LLM can add a hint as an alternative producer of one fact (off by default); an invalid reply is no hint, never a guess.
• The dispatcher stores every decision; consolidation is the application's code: after each goal it compiles System 1's routes from the map's confirmed claims (WorldMap.distances(..., confirmed_only=True)), so what System 2 found by deliberating becomes what System 1 does at once. The dispatcher itself does not teach System 1.

run	moves	System 1	System 2	slow decisions after a surprise
1, empty memory	183	36	147	—
2, run 1's memory	62 (the fewest possible)	60	2	2 (Professor Oak's walk to the lab, the ship leaving)

System 1 takes about 0.3 ms per decision and System 2 about 2 ms on a laptop CPU. All 245 stored decisions replay without the game, both world maps' journals verify, and System.report over each run's store gives the same counts. --live plays the runs again (the same decisions), --rom PATH first checks every recorded exit against your own ROM's map tables. What it does not show: the gain is the memory carried to a second run over the same world — a first exploration is not shorter; walking, battles and menus inside a place are outside the showcase; and a place counts as reachable from the stage at which the recorded player first reached it. The replay viewer is a static Space in spaces/pokemon/.