Mech Interp Challenge: November - Deciphering the Cumulative Sum Model — LessWrong