Ask what a game is, in session.
Inspect beliefs, find comps, and run counterfactuals as MCP tools. Copilots get mechanics, identity, and uncertainty as structured state, not store-page chunks ranked by cosine.
Latent Spark — Headless + Integrated
A belief atlas,
not a document search.
The same calibrated understanding of a game, in the UI or over the wire. Belief states with explicit uncertainty, not RAG chunks or tag-graph hops.
Surfaces
One capability catalogue across the UI, MCP, OpenAPI, and A2A.
Ask what it could become, at scale.
Same capability IDs over OpenAPI and POST /capabilities/{id}/invoke. Decoder edits like “more co-op” or “lean premium” shift the neighbourhood; batch retrieval keeps session state and lineage.
Ask which comps hold, across agents.
Publish an agent card so peers can call Latent Spark for asymmetric similarity with mechanics rationale. Tag-graph hops stay out; thin evidence stays flagged.
Why not just
ask an LLM?
An LLM can describe the games it happens to know. Latent Spark measures an industry-wide corpus in one coordinate system—from evidence, with known uncertainty—so “games whose community behaves like Rust's without the obligation load” returns a ranked, checkable set, not an anecdote from the famous two hundred.
The long-tail ceiling.
We've measured it. Frontier LLMs return nothing useful for the long-tail. For most games they know a store blurb or less—and invent genre-typical answers rather than say so.
No shared coordinate system.
Comparison needs a common scale; chat doesn't have one. Ask about Rust and Valheim and you get two essays—ask it to rank 500 survival games by social obligation and you get a confident, unverifiable list. With Latent Spark, distance, blends, and axis moves are geometry.
Traceable, priced evidence.
Every Latent Spark score cites its atoms, which cite reviews. Certainty says how much evidence backs it; unknowns stay explicit. When claims fail against observed evidence, the system catches it. A chat answer has no trail to audit—and hides its surprises.
Wire Latent Spark into your stack.
See how MCP tools, capability invoke, and OIE federation fit the agents and pipelines your studio already runs.