In pure GR an infalling observer will sail past the EH and not notice anything unusual since spacetime is locally Minkowski (ignoring tidal forces, which is valid e.g. for humans falling into supermassive BHs). If the (GR+QM) firewall hypothesis is correct (a big if), an infalling observer will instead be promptly incinerated within a Planck's length of the EH. The intuition one builds from a pure GR understanding of BHs may be dramatically wrong, not just at the singularity, but all the way out at the EH.
> If the (GR+QM) firewall hypothesis is correct (a big if)
A big if indeed, but if that hypothesis is correct, then the GR solution that applies is no longer the standard black hole solution. The "firewall" is not vacuum--more precisely, it does not have a vanishing stress-energy tensor. Which means "the intuition one builds from a pure GR understanding" for the "firewall" case will need to be a pure GR understanding of a different solution from the standard BH, and of course such an understanding can be perfectly correct.
In other words, if you're going to talk about a "firewall" solution, then saying "well, GR doesn't model that correctly because it's not a standard GR black hole" is simply wrong. GR can model lots of other things besides standard (vacuum solution) black holes. You just have to use the correct GR model for the actual stress-energy tensor that is present. Of course statements about a standard vacuum black hole will not be correct for a different non-vacuum solution; but that is not contradicting anything I said, because the post I was responding to was assuming a standard vacuum black hole, and my statement was correct for that case.
The real question is whether such a "firewall" model, with a nonzero stress-energy tensor, would even have an event horizon. As far as I know nobody has actually answered that question; the treatments I have seen have simply assumed that there is one without taking into account the fact that the "firewall" stress-energy tensor is non-vanishing. If there is an event horizon in such a model, then my statement would still be correct for that model, since my statement was based on general properties of event horizons.
