Back to Agents That Account for Themselves
working paper2026Research paper plus theorem document

Graph-Augmented Recursive Language Models for Personal Knowledge Systems

The cognitive twin work treats a long interaction history as a cognitive fingerprint: a record of how someone asks, builds, revises, and reasons. The research question is how to distill that record into a model and retrieval system without flattening away personal structure.

Paper workspace

Live draft structure

working-draft

Artifacts

Draft PDF

Cognitive twin draft PDF. Public page should avoid exposing private training data.

Open artifact

Editable source

Draft PDF exists, but release requires a privacy pass over corpus details and training claims.

Source anchors

cognitive-twin-research-paper.tex

cognitive-twin-theorems.tex

Comp-Core/packages/cognitive-twin/paper/cog-rlm-paper.md

Method tags

cognitive twinpersonal knowledgegraph augmented modeling

Ingest intersections

cognitive-twinpersonalizationsftgraphprivacy

Status

Paper plus theorems document drafted; SFT pipeline operational.

Key claims

01

A person's interaction history contains trainable cognitive structure.

02

Theorems and pipeline documents separate the conceptual claim from the SFT machinery.

03

Privacy and release boundaries matter more here than in ordinary papers.

Public reading note

Public abstract is safe; full corpus and training details remain private.

Standard skeleton

What this paper must keep proving

Schema

problem

Personal AI loses structure when conversation history is treated as raw text rather than a graph of recurring cognition.

method

Distill interaction history into a graph-augmented model and companion theorems.

implementation

Conversation extraction, graph construction, SFT pair generation, theorem document.

data

Private interaction corpus. Public work should only expose aggregated methods and privacy-safe findings.

evaluation

Behavioral fidelity, retrieval grounding, privacy boundaries, and personal reasoning reconstruction.

references

Personalization, cognitive modeling, graph-augmented LMs, SFT/DPO training.

openQuestions

Which results can be made public without leaking private corpus structure.

Checkpoints and references

Proof chain

paperpending

Claim checkpoint

central-claim slot

Every central claim must point to a proof anchor or remain labeled as speculative.

implementationpending

Implementation checkpoint

implementation-map slot

Every method should identify the code path, harness, schema, or protocol that embodies it.

experimentpending

Evidence checkpoint

evidence-manifest slot

Every reported result should point to run IDs, packet IDs, data snapshots, commits, or review artifacts.

external-referencepending

Reference checkpoint

references slot

Every external claim should resolve to a cited paper, benchmark, standard, or documented prior system.

paperpending

Release checkpoint

release-gate slot

Every PDF needs a named condition before it can move from draft to citation-ready.