Public data and model
AAC Children Conversations and TinkyBrain-v6 are available on Hugging Face. The public paper describes an earlier v3 model snapshot. These artifacts are not biomedical validation.
Lock the evidence.
Keep the science open.
A credible research system distinguishes released artifacts, implemented prototype capabilities, and work still to be validated. Preserve the evidence a conclusion depends on. Keep the conclusion open to new results and informed criticism.
What is public. What is built. What is next.
Peel and Lois architecture
The public paper describes typed knowledge, provenance, bounded reasoning, and offline operation. Inspectable reasoning does not guarantee that every source is correct.
Bioscience research workflow
Literature maps, hypothesis review, and lab continuity are proposed applications. No clinical outcomes or laboratory productivity gains are reported.
Preserve the basis.
Leave room for revision.
A research extension should let a colleague answer the questions below. These are evaluation requirements for a scientific pilot, not a claim that a complete research evidence platform has shipped.
Where did the claim come from?
Retain the source, its version, the study conditions, and the relevant evidence so a reviewer can follow the claim back.
Why did the system answer that way?
Expose the relationships, rules, and assumptions used. Distinguish directly recorded information from a derived conclusion.
What changed afterward?
Connect new results to earlier versions and preserve the revision history instead of silently replacing the earlier basis.
What still needs a test?
Keep hypotheses, missing support, and reviewed conclusions distinct. Record which checks would justify moving forward.
Scientific value needs
results someone else can examine.
| Question | How to evaluate it |
|---|---|
| Is the answer supported? | Expert review of agreement between answers and cited records. |
| Is context preserved? | Check the study system, conditions, and semantics of the relations used. |
| When should it abstain? | Evaluate missing evidence and questions outside the graph’s coverage. |
| Are conflicts identified correctly? | Use reviewable conflict cases and measure both missed conflicts and false alarms. |
| Does it improve research work? | Compare against an existing workflow with predefined criteria before claiming time savings or discovery gains. |
Artifacts and research.
Start with the public work and examine the basis of the story. Research papers are preprints and have not been peer-reviewed.
Released
AAC Children Conversations
The public conversation dataset
ReleasedTinkyBrain-v6
Current public AAC model; the paper reports the earlier v3 snapshot
PreprintPeel: symbolic facts and provenance
Public architecture, reported scale, and limitations
PreprintTraversing Data in Symbolic Systems
Typed paths and their formal properties
PreprintRetrieval Is Not Memory
Memory, retention, and knowledge governance
PreprintVerified Before Acting
A research framework for pre-action evaluation
A traceable claim can still
come from a mistaken source.
An incorrect source can be recorded faithfully. Relations and rules cannot cover every question. A path in a graph does not establish a biological mechanism. Peel’s proposition is to make those dependencies visible and inspectable, then subject its scientific applications to rigorous evaluation.
Continue exploringOverview