question about sdf usage

[Korace0v0]

Hi, thanks for your great work. May I ask a simple question?

I find there is slightly different between nr3d and sdfstudio or instant-nsr-pl. During occ grid updating, both sdfstudio and instant-nsr use sdf to get alpha. Then they use alpha threshold to update the binary. But in fact, I do not really understand why they did this, since step_norm does not stand for the sdf change as they do in the rendering stage.

        estimated_next_sdf = sdf + step_norm * 0.5
        estimated_prev_sdf = sdf - step_norm * 0.5

        prev_cdf = torch.sigmoid(estimated_prev_sdf * inv_s)
        next_cdf = torch.sigmoid(estimated_next_sdf * inv_s)

I find that this work seems to use sdf to get density first, then update binary using the density threshold like npg. Is there any difference between using density or alpha to get occ value?
I do think that density is much more reasonable but I really can't understant it.

normalized_logistic_density(sdf, inv_s)

Thanks for your help.