Cave-ins stun creatures not immune to stunning, and that's sufficient for trapping them in cages, which is what Gremlin traps are based on (a lever on a floor on a support. The Gremlin pulls the lever, which removes the support which causes it to fall onto traps one level further down surrounding the support). I've had Gremlin escape by (presumably) manage to move away from the floor before the cave-in (and not be flung down by the collapse), though. My latest design has a two tile floor collapse, rather than a single tile one.
Maybe it's one of the cases where activation of effector (support) depends on the order of building, with one tick difference, which could be enough to make the critter move away. In case of my kobold, there were two sparring messages before the first announcement (about something collapsing) and the second one (the kobold hidden away), and all were in a block of sparring messages (so many messages in a short time).
I run the experiment again, and this time kobold came from the west, and the cage and minecart were thrown differently. Here it is, the number is the tick number:
1 kobold enters the baiting tile (from the west this time, opposite the support)
2 kobold steals the bait, support disappear, floors left hanging
3 hanging floors collapse (disappear, replaced by blocks), clouds created,
collapse announcement4 clouds spread, kobold still in the baiting tile
.
7 cloud spread, kobold still in the baiting tile, not yet engulfed by cloud
8 cloud spread to the baiting tile and the kobold
9 kobold unaffected yet, blocks (from ex-floors) fall to the ground, but still in motion
.
12 kobold moves east (towards the original support), is
caught in the cage, its unconscious counter rises to 455 (from zero), cage not moving yet,
minecart becomes airborne and is in the same tile as the kobold,
kobold hidden away announcement13 kobold unconsciousness falls to 454 (will be falling every tick from now on), both cage and minecart moved two tiles south and one tile west from the previous tick
14 cage and minecart still in the same tile as in previous tick, some blocks (from tiles) move away from the "point of impact"
.
17 one block thrown backo into air, the same position as it was in the beginning
18 cage and minecart moved one tile west
19
20 cage and minecart move one tile east (the same as they were in ticks 13-17)
21
22 cage and minecart move one tile east and two tiles north (the same as they were in tick 12)
.
26 cage and minecart move one tile south and two tiles west (they weren't here yet)
27 cage and minecart move one tile south and one tile east (third time here)
28 cage and minecart move one tile south
.
31 cage and minecart move one tile east
32 cage and minecart move one tile east
33 cage and minecart move two tiles north
.
39 cage and minecart move one tile north
40 cage and minecart move one tile west (starting position, as in tick 12)
.
44 cage and minecart move one tile east (as in tick 39)
45
46 cage and minecart move one tile west and one tile north
.
49 cage and minecart move one tile east, and this is their
final position50+ the blocks also came to the rest, and the cloud has later dissipated
Additionally, the clouds in tick 3 are created both under the falling floors (ground level) and in their places (in the air). Clouds were composed of yellow sand (on the ground, probably because the soil is yellow sand) and black sand in air (and one tile doubled by yellow sand cloud, possibly thrown from below). This black sand cloud is interesting, because there is no black sand on the map.
Overall, the kobold was for quite a ride, like in a roller-coaster. I suppose the clouds of dust throw the items around.