use data from BrainAccess HALO + Using an Eye Tracker for more computer data
There are a lot of good reasons why Chess is the perfect arena to test the framework of Self Configuration. For one, it consists of short games with a discrete outcome, but thanks to chess analysis bots we can evaluate every single move one makes and how it affects the game, meaning the differentiable optimization metric isnt just the game outcome but the moves we make.
It is also the perfect place to test ideas of differentiable content consumption ( Maximizing Activation through Differentiable Content Consumption ) as most if not all chess learning is going to come from online resources like youtube videos.
- we can use ActivityWatch to measure/evaluate the content consumption we do to optimize for chess, and then the outcome it has on subsequent games.
- Eye Tracking
Optimization objective is scalar so we really can just have a very simple RL loop here. Some system configures the content I consume, the practice I do, therapies ( either Neuropharmacology or tDCS, or both via Pharmacological tDCS lol ), evaluated on real online games.
We can enrich optimization not just by scalar outcome, but by moves played. Use stockfish or another engine to evaluate every single move I make, now we have many many data points and we can numerically represent how good of a move it was.
- might also be a good application of autoresearch, specifically with things hard to differentiate like tDCS. An agent autoresearching tDCS for chess may be able to leverage benchmarks from game performance via stockfish and general ActivityWatch focus data.
- We could do BrainAccess HALO? I dont know if it will leverage much as our EEG is quite weak but also cannot even be worn during the game if we are doing tDCS. Of course we could always do tDCS/eeg outside of the main game loop(instead focusing on the learning phase), however that is to be decided in the future.
- 09-07-2026 12:28 something that could also be done is to curate content consumption by specific errors i make in positional chess. Lichess might be a better platform for this, especially with their unlimited puzzles and puzzle matching algorithm.
Anki and Chess.com Learning Loop
Anki should be the selective retention and scheduled-recall layer inside this system, not the chess environment or the optimization target. It is best suited to a small number of opening decisions, theoretical endgames, recurring strategic structures, and expensive mistakes extracted from my own games.
ActivityWatch can align games and learning sessions with consumed material.
Eye tracking or BrainAccess HALO can later add process measurements, but
their signals should remain explanatory covariates until they predict held-out
or future-game improvement.
Open configuration gates:
- primary time control used for the optimization benchmark;
- current White and Black repertoire;
- board orientation/coordinates preference;
- final Stockfish depth/time budget for routine analysis.
tDCS
For playing and learning chess, the goal shifts from text comprehension to spatial awareness, pattern recognition, and rapid skill acquisition. [1]
The ideal configuration for this is the Left Parietal / Right Shoulder montage.
- Anode (Red / +): Place on P3 (Left Parietal Cortex).
- Cathode (Black / -): Place on the Right Shoulder (extracephalic placement) or Fp2 (Right Forehead).
- Why it works: The parietal lobe processes spatial relationships and object manipulation, which are critical for visualizing board states and calculating future chess moves.
- Current: 1.5 mA to 2.0 mA for 20 minutes.