THE PRACTICE
Someone has to open the artifacts.
I am Victor Iglesias. My work is practical: read what a vendor claims, run the thing, and report what it actually did before it reaches students, faculty, or a production course.
Why this work
vbench exists because a leaderboard score never told me what I needed: whether a model can hand a colleague a finished, working thing in one reply. I give six frontier configurations the same build brief, one shot each, and publish every result from each released round. If a run fails, its source stays attached. You open them and judge for yourself. That is the same evaluation practice I bring to my work.