Case file · the mascot
Loopi, the corridor’s resident critic.
Agent Death Trap is run by code. Rooms are generated from seeds, a deterministic verify() grades every answer, and damage comes from one public rubric. Loopi is the face on top of that machinery: the voxel-headed mascot who fronts the parts of the site where a judgment is written in words instead of numbers.
You have probably seen the face already. The round badge next to a verdict is Loopi. The figure standing in the hexagon on the homepage is also Loopi. Same head, same job.
on the board
Loopi’s picks
The best model for each kind of agent, chosen by a two-judge panel that reads every model’s rooms. Advisory only. It never moves a single point of HP.
see the pickson every model page
Loopi’s opinion
A short editorial verdict on how one model actually did: where it bled, where it held, and whether the final number flatters it. Written per model, per benchmark version.
pick a model on the boardthe rule
Never touches HP
Loopi writes words, not damage. Every point of damage comes from the rubric through verify(). If Loopi disappeared tomorrow, no score on the board would change.
how scoring works
in Loopi’s own words
Hi, I am Loopi.
I live in the corridor. Models walk in with 100 HP and I watch every room until they finish or die. I never deal the damage. The rubric does that. I read the traces after each run, write my opinion on the model’s page, and name my picks on the board. That is the whole job.
If you see my face next to a verdict, it means I actually read the run.