Ask what is supported
Answer from stored records, return available provenance, and signal whether a query has sufficient symbolic support.
An answer needs a basis.
A gap deserves a name.
The symbolic reasoning core is how Lois reasons over what is already known. It queries recorded knowledge, explains paths through typed relationships, and checks conflicts within known constraints. We expose those capabilities to the surrounding runtime so other components can ask for evidence.
Implemented symbolic core; the complete scientific research workflow remains a research direction.
Knowledge you can query,
explain, and check.
A graph stores connections. A running reasoning system performs defined operations over them. Lois exposes queries, relationship explanations, and contradiction checks as tools for the surrounding application. Each operation stays bounded by the graph and rules available to it.
Follow the explanation
Expose a chain of relationships between known concepts. The meaning of that chain still depends on the semantics of each relation.
Check a known conflict
Compare a candidate claim against constraints the system can express. Where the knowledge is missing, return unknown rather than imply verification.
Three outcomes. Each one useful.
INTERACTIVE ARCHITECTURE ILLUSTRATION · NOT A LIVE LOIS QUERY
Concept Arecorded linkConcept Brecorded linkConcept C
A path through recorded knowledge
The graph contains a path from concept A through concept B to concept C. An explanation can expose that path and the records behind its relationships.
Finding a path does not make every relation transitive, or establish a causal mechanism.
Candidate claimconflicts withKnown constraintrequires reviewEvidence retained
A conflict worth stopping for
A candidate claim conflicts with an encoded exclusivity constraint. Within that known scope, the symbolic check can stop it from being treated as a supported answer.
Only conflicts expressible in the available graph and rules can be checked. Different experimental contexts still need domain review.
Research questionasks forMissing relationevidence gapUnknown
Unknown is a useful result
The relationship needed to answer is absent. The system can report the gap instead of presenting a plausible connection as established knowledge.
For a researcher, an explicit gap can help define the next question. It is not an experimental finding.
Put a claim back
inside its evidence and conditions.
The kind of understanding we want Lois to support is practical: identify what a claim depends on, locate an unsupported relationship, and distinguish an explanation from a hypothesis awaiting a test. That is a concrete foundation for research assistance.
Research memory that can be revisited
The scientific extension aims to let a team return to earlier reasoning, inspect the evidence available at the time, and reassess it after new results. Persistence, versioning, and domain evaluation are parts of that validation path.
Language makes it readable.
Evidence determines its support.
The hybrid architecture separates fluent expression from the basis of a claim. A language model can help a researcher understand a result; sources and relationships determine what supports it. More persuasive wording does not increase the strength of the evidence.
Continue exploringScience runtime