The paladin calls for the horse. It magically appears.
From Gygax's DMG, p 18:
It will magically appear, but not in actual physical form. The paladin will magically "see" his or her faithful destrier in whatever location it is currently in . . .
The horse
already exists when the paladin calls for it. It appears to him/her in the sense of
being seen by him/her. It is not spontaneously called into existence.
In a typical game, the GM will look at the encounter table and see that Lich is a possible result. By looking at this table and agreeing to roll on it, he has thus accepted as an established fact that there is a Lich around here... somewhere.
Two things.
First, the GM need not have accepted any such fact. If the lich result never comes up on the table in the whole course of the campaign, the GM is not committed to it being true that an undiscovered lich existed in the campaign world. In my experience (playing, reading around magazines and message boards, etc) it is a small minority of GMs who treat the random encounter table as representing gameworld actualities rather than gameworld possibiites.
Second, and more importantly,
the GM will not have worked out any details of the lich until it comes up on the table.
That is the point at which the GM works out its backstory, fits it into the gameworld history etc. In other words, an out-of-game event - the roll on the table - leads the GM to introduce new material into the gameworld past. I don't understand why you do not respond to this point. Even if - contrary to the rules - you treat the paladin's horse as a case of spontaneous creation, no one treats the random encounter result as a spontaneous creation. These entities, together with their histories and backstories, get incorporated into the unfolding fiction via an act of authorship on the part of the GM.
There is no meta-game event, because whether or not the party encounters the Lich is not a factor in the Lich existing, or in determining anything about its prior history.
Just as [MENTION=22779]Hussar[/MENTION] is curious what you think people mean when they refer to "player authorship", so I am increasingly curious what you think anyone means when they talk about the metagame shaping the authorship of the fiction.
In this particular case, the rolling of an encounter is a metagame event - it happens in the real world, using dice and a table, in accordance with a wandering monster procedure spelled out for the GM in a rulebook.
And as a result of following that procedure, the GM now has to engage in an act of authorship - namely, detailing the lich's origins and backstory - which (i) hadn't previously been undertaken, and (ii) adds new content into the shared fiction - much like a new Harry Potte novel adds new content into that fiction, even if some of that content deals with events that, within the fiction, happened before the events of an earlier novel. The time-sequence of authorship doesn't always correspond to the time-sequence of ingame events; hence prequel are possible.
And not that none of this is driven by consideations of ingame causality. The GM is not inferring to the lich, or its origins and purposes, simply by extrapolation via ingame causal laws from a known prior state of the ingame world. S/he is doing all this because s/he is falling the content-creation rules of the game, which include rand
The DM should have determined this before rolling. It is probably the same one, given how few Liches "live" in any given area.
The random encounters tables in the back of Gygax's DMG have hundreds of entries. No GM ever determined all this stuff before rolling. Nor does any DMG I'm familiar with advise that s/he should. As I've mentioned upthread, part of a GM's skill is being able to manage backstory, including backstory introduction resulting from such devices as random encounters, in a manner that preserves the consistency of the gameworld.