
Automatic glitch detection in gameplay video
Early ReBlink research on using V-JEPA 2 surprise to find temporal glitches in gameplay footage without labeled bug training, with VideoGlitchBench as the next evaluation.
Read moreIdeas, evidence, and field notes
Research, articles, case studies, and coverage from the systems we build and the worlds that inform them.

Early ReBlink research on using V-JEPA 2 surprise to find temporal glitches in gameplay footage without labeled bug training, with VideoGlitchBench as the next evaluation.
Read more
A design exploration of how language could open Total War's campaign and battle controls without lowering its strategic bar, then support policies, standing orders, doctrines, command capacity, and counterplay.
Read more
A design exploration of how language could open Civilization's control surface without lowering its strategic bar, then support policies, governors, commanders, and counterplay.
Read more
A design exploration of how language could open the control surface in Age of Empires IV while preserving difficulty and creating new strategy through policies, standing orders, and costed Overseers.
Read more
Language control can expand who can play a hard game, which devices can support it, and how a studio should assess its potential return.
Read more
ReBlink Matter combines studio material library management with search, generation, refinement, and engine-native shader generation.
Read more
Playthrough captures reasoning traces with the play so a designer or QA lead can tell whether a surprising result is a real rule change, a bad line, unusual agent reasoning, or a contaminated run.
Read more
World-intelligence companies treat game engines as a rich labeled world, but telemetry is not automatically a training set. Train on accepted, attributable transitions from an authoritative engine.
Read more