Have you tried a hard square curve, with just two change points and a p of 0.99 if inside and 0.01 if outside? Two switches should do it. From there you could try adding one sigmoid at a time and hopefully see what is breaking in the full model.
Have you tried a hard square curve, with just two change points and a p of 0.99 if inside and 0.01 if outside? Two switches should do it. From there you could try adding one sigmoid at a time and hopefully see what is breaking in the full model.