Well, last night's session was a very close call indeed, in part because the PCs' actions allowed the adversaries to prepare for their entrance into the final room by pre-casting several key spells. More importantly, it allowed two separate encounter groups to combine into one, which many folks warn against, given the tight math of challenges in the Building Encounters guidelines. This said, I prefer things to develop logically, and the arrival of reinforcements from the roof of the tower was the only logical way to have the scenario work out. I used some threat mitigation strategies to make things a little easier on the PCs, and I suspect it was the only thing standing between success and a TPK.
There were 5 12th-level PCs (fighter, rogue, druid, cleric & wizard). The initial threat was the leader of the Aspis Consortium expedition, a 14th-level assassin/cleric of Norgerber, a 12th-level Omox demon (nasty guy!), a 9th-level occult sorcerer and two 9th-level specters (dominated by the leader). That totals to 165xp, a severe to extreme threat encounter (150-200 for a party of five).
The second wave included a 9th-level Vrock demon, a 10th-level occult sorcerer and 2 8th-level mercenaries. Only 85xp on their own, a low to moderate threat encounter, but combined with the previous one, a whopping 250xp, well over the extreme threat encounter ceiling of 200xp. So I decided that the second wave would arrive only after a couple rounds. The Vrock demon arrived at its initiative on round 2, flying in through the crumbled gap in the top right corner of the tower, and the 2 mercenaries with their sorcerer would arrive pelting down the corner staircase on round 4.
This sort of threat mitigation approach is key in evening the odds, and changing a deathtrap into something that proved to be just barely manageable. The druid was down as far as "Dying 3" (the equivalent of being deep in negative hit points), the rogue was twice down to single-digit hit points, and even the fighter with 200 hit points was down to less than a quarter of that total. The wizard was blinded around round 6, and only the cleric was really holding his own, though he was also down below half on two separate occasions.
The Omox demon and Vrock demon were destroyed, as were the two specters, the two mercenaries and one of the occult sorcers, but the leader and the other occult sorcerer dimension doored away once it was clear they no longer had the advantage. That's another threat mitigation strategy - I really dislike the notion that everybody and everything fights to the death. It doesn't make sense to me, when they could just as easily retreat, regroup with some nearby allies, and return to mop up the floor with the PCs a few minutes later. Fortunately, the players realized their time was limited, they grabbed what loot they could, managed to spot a second force of a dozen men racing towards the tower, and made good their escape before having to face those forces, using a Living Landslide (a mid-level Earth elemental) to collapse part of their escape tunnel and prevent direct pursuit.
Now, some folks would say that I broke several rules in PF2 encounter design:
1) don't go beyond an "extreme" threat encounter
2) don't combine multiple encounters into one
3) don't allow adversaries to pre-buff before the encounter begins.
I did all those things, because it was logical, to my DM way of thinking, to do so. But I also used a few threat mitigation approaches that also felt logical and warranted. That's where your DM intuition and experience comes in.
There has been a lot of talk in this thread about how the relative threat level math in PF2 is far tighter than in other games. It's true. It would be disingenuous of me to claim otherwise. But there are always ways to change or finesse the playing field, so to speak, and those ways tend to make an RPG session far more interesting than a constant kill or be killed approach.