Solstice research is not one thing. Some of it is hard runtime work, some of it is applied evidence systems, and some of it is category-defining material around replayable state and memory economics. This page separates the key surfaces so the site stops forcing all of that into one link.
These are the three public research surfaces worth showing directly. They make different claims and should not be collapsed into one vague "innovation" bucket.
The TSC series is the clearest expression of Solstice runtime research. It covers the compression thesis, benchmark results, scaling behavior, and the economics of keeping model state viable in production systems. This is where the deeper technical story belongs when the audience cares about infrastructure and not just product packaging.
DeltaStore is where the replay and continuity thesis becomes concrete. It is less about publishing a formal paper and more about making the argument that AI systems should be reconstructable, branchable, and auditable by default. This is the runtime-adjacent research surface that supports the broader Delta story.
Evidence Engine is the research side of the Solstice reporting stack. It takes evidence gathering, claim testing, and synthesis seriously instead of treating "research" as summary generation. This is the best public surface for showing that Solstice can do disciplined evidence work, not just broad AI narration.
Proteus is the governed-projection surface. It turns live evidence into typed enterprise objects — runbooks, incidents, support resolutions — where every field carries provenance, a calibrated confidence band, tracked gaps, and a policy decision that gates action. This is where Solstice makes the argument that AI outputs should behave like first-class records, not free-text answers.