Knowledge with context
What does a record say? In which context? What does it connect to? Peel keeps these distinctions visible. The cube illustrates why similar words on different paths can carry different meanings.
Give knowledge structure.
Give reasoning a boundary.
Peel brings knowledge, reasoning, language, and execution into a system with distinct responsibilities. That separation makes it possible to inspect where an answer came from and understand where its authority ends.
From a record to an inspectable answer.
A view of what the system contains and the responsibilities each part carries.
01
Records
Keep a statement in its context.
02
Relationships
Give connections a type and direction.
03
Reasoning
Follow permitted paths and explain them.
04
Expression
Make an evidenced answer useful to people.
Each layer has a job to do.
Relationships with meaning
Part of, requires, and related to mean different things. Named relationships let a system distinguish those connections. An association link, by itself, does not establish causation.
The Lois symbolic core
Query stored knowledge, explain relationship paths, and check conflicts within the encoded constraints. Its reach is bounded by the knowledge and rules actually present.
Language and the runtime
Language models can help express an answer; Perslis connects tools. The architectural goal is to keep generated language and actions within evidence boundaries. Complete scientific workflows still require evaluation.
A connection needs
the conditions that make it meaningful.
In research, an observed change belongs with the system studied, the conditions, and the method. A scientific knowledge layer needs those qualifiers to determine whether two findings agree, disagree, or address different questions.
From architecture to domain validation
The public Peel data demonstrates conversation structure. Taking these principles into bioscience requires domain vocabularies, evidence standards, and expert evaluation. That extension is a research direction.
Keep reasoning close
to the researcher.
The public prototype’s symbolic query path operates locally and offline. The associated small-model work targets on-device use too. For a research team, that opens a practical direction: retain a defined knowledge snapshot and inspect the answers it supports without depending on remote inference.
One system, distinct public artifacts
The conversation dataset makes context and knowledge structure visible. TinkyBrain represents the associated on-device language-model work. The symbolic core and language model carry different responsibilities; privacy, access controls, and reliability in a lab deployment still need separate evaluation.
Repeatable reasoning.
Sources that remain open to challenge.
With the same graph, rules, and query conditions, symbolic reasoning can produce a repeatable result. Correctness still depends on the sources, the coverage of the graph, and the suitability of its relationships. Inspectability makes those dependencies available for criticism and correction.
The symbolic floor, at work.
A short, unedited capture of the typed layer that authors facts — and returns unknown when the evidence is missing. Recorded on a research prototype.
The symbolic floor
The typed layer that authors facts and returns unknown when evidence is missing.
Continue exploringLois