> 2. Actually draw the ascii board for each step.
I doubt that this is going to make much difference. 2D "graphics" like ASCII art are foreign to language models - the models perceive text as a stream of tokens (including newlines), so "vertical" relationships between lines of text aren't obvious to them like they would be to a human viewer. Having that board diagram in the context window isn't likely to help the model reason about the game.
Having the model list out the positions of each piece on the board in plain text (e.g. "Black knight at c5") might be a more suitable way to reinforce the model's positional awareness.
I've had some success getting models to recognize simple electronic circuits drawn using ASCII art, including stuff like identifying a buck converter circuit in various guises.
However, as you point out, the way we feed these models especially make them vertically challenged, so to speak. This makes them unable to reliably identify vertically separated components in a circuit for example.
With combined vision+text models becoming more common place, perhaps running the rendered text input through the vision model might help.
With positional encoding, an ascii board diagram actually shouldn't be that hard to read for an LLM. Columns and diagonals are just different strides through the flattened board representation.