[raattgift]

Observers outside a black hole (BH) are free to disagree about the location and shape of the horizon. However, there is no observer which is free to say that there is no horizon [1]. This latter point is why the horizon is a physical feature of the universe containing the BH. The former point is why infallers can cross the horizon according to them, while some outside observers will never actually see the crossing [2].

This is not an analogy as much as an example of an external-vs-internal observer problem. When you close the (opaque, insulated) door of your fridge, observers inside will see the (filament of the incandescent) light significantly dim, and if the door stays closed long enough, will see the light thermalize with rest of the internal volume. Someone standing outside the fridge might not even see the initial dimming; indeed, that observer may only ever see the light as "on" (rather than "heating from cold" or "cooling from hot").

[1] We could talk about naked singularities a bit: this usually means that there is at least one outside-the-black-hole observer for which the shape of the horizon is such that the centre of mass-energy of the BH is outside the horizon, rather than an observer for which there is no horizon at all. However, even these scantily clad BHs don't arise in realistic universes described by General Relativity. Fully naked singularites (where at least one observer exists which does not see any horizon at all) require an alternative theory of gravitation, or conditions extremely unlike those anywhere in our universe.

[2] Consider the observation of a supermassive black hole at the edge of the observable universe. From our view here around Earth, we see a race between a very bright star about to cross the black hole's event horizon and the black hole about to cross our Hubble horizon. Observatory A sees the star vanishing behind the horizon just in time; Observatory B sees the BH cross out of observability before the star vanishes behind the BH horizon. A and B have (very slightly) different Hubble horizons focused on them [3], and also with a (n also slightly) different radial distance to the BH horizon. "B" can never directly see the same coincidence of events that "A" sees; should "B" deny the infalling?

[3] Maybe this is illustrative of observer-centred observables? Glories (an optical phenomenon similar to rainbows) are so observer-specific that you and your handheld camera will have different ones (and each of your eyes will have different ones). As noted in the "From the air" subsection, we can tell what seat a photographer of a glory from a plane must have been sitting in. https://www.atoptics.co.uk/droplets/gloim1.htm Likewise, we can determine the location in spacetime of an observer of a star-into-black-hole event from that observer's detailed description.