Kabbalistic Core — public research edition

Kabbalistic Core is a deterministic, inspectable cognitive graph organized by a Kabbalistic topology. It turns an intention into typed perspectives, directed peer observations, an explicit integration step, a bounded symbolic return, and a traceable response. A local interface can optionally use a compatible Qwen3 8B model for bounded proposals while retaining a visibly labeled deterministic fallback.

This repository is a clean public-edition export. It contains no private Git history, private corpus, personal archive, cloud-document locator, model weight, continuity database, authority checkpoint, or real session trace. The bundled corpus and evaluation inputs are project-authored synthetic fixtures.

What it does

• Routes one intention through an explicit Tree-of-Life graph.
• Maintains three distinct proposal roles: Form, Flow, and Accord.
• Requires all three first-pass views before six directed peer observations.
• Integrates at Tiferet without allowing a majority to override hard bounds.
• Records permissions, transitions, evidence bindings, state hashes, and stop reasons instead of presenting hidden reasoning.
• Offers three project-owned functional kernels: clear_sight, liberation, and regeneration.
• Retrieves exact excerpts from a versioned, hash-bound synthetic corpus.
• Compares the graph with a budget-matched neutral three-perspective pipeline.
• Includes a bounded encrypted continuity experiment with explicit proposal, consent, commit, correction, and causal comparison records.

What it does not claim

This is not a consciousness detector, a copied person, a general-purpose agent, an autonomous authority, or proof that software exhausts Kabbalah or mystical experience. Symbolic intensity is not evidence. Retrieved text cannot grant tools or permissions. A fluent response is not proof of correctness, identity, growth, or personhood.

The continuity code demonstrates narrow record-to-response causality with synthetic inputs. It is not a general relationship-memory product. Forgetting, production key management, multi-user isolation, remote deployment, and a complete threat model are outside this preview.

Repository map

src/kabbalistic_core/   graph, engine, trace, retrieval, evaluation, continuity
tests/                  Python contract and causal-slice tests
ui/                     local inspection interface and rendered HTML tests
evaluation/             synthetic comparison definitions and local runners
examples/corpus_sources project-authored source documents for the public corpus
docs/                   architecture, boundaries, evaluation, and operation notes

Quick start: deterministic CLI

Requirements: Python 3.12 or newer.

python -m venv .venv
.\.venv\Scripts\Activate.ps1
python -m pip install -e .
python -m unittest discover -s tests -v
kabbalistic-core --intention "Choose one reversible next step" --symbol "threshold" --kernels clear_sight,regeneration --no-memory-proposal

The CLI writes paired trace artifacts to runs/ by default. That directory is ignored because a real intention or feedback record may be private. Use --output-dir to select another local-only location.

On macOS or Linux, activate the virtual environment with source .venv/bin/activate and use the same Python and CLI commands.

Local inspection interface

Requirements: Python 3.12+, Node.js 22.13+, and pnpm 10.

python -m pip install -e .
Set-Location ui
pnpm install --frozen-lockfile
Set-Location ..
.\start-seed.ps1 -CheckOnly
.\start-seed.ps1

The launcher binds the backend and interface to loopback only. If a compatible local Qwen3 8B model service and model file are available, the launcher can use them. Otherwise the application exposes the deterministic fallback as such. It does not download dependencies or model weights at runtime.

For a portable two-terminal setup on another operating system:

Terminal 1: python -m kabbalistic_core.poc_server
Terminal 2: pnpm --dir ui run dev

Then open http://127.0.0.1:3000 locally.

Public synthetic corpus

The default corpus is classified public-demo-safe with publication consent recorded. It contains twelve embedded excerpts drawn from the three Markdown files in examples/corpus_sources/. Document and excerpt SHA-256 values make the exact input field inspectable. All three documents were written for this public edition; they are not excerpts from a private archive.

To substitute a corpus, preserve the schema and declared hashes. Public runtime mode refuses a corpus unless its classification is public-demo-safe and its publication consent is recorded. A label is not a rights review: only add material you own or are independently authorized to redistribute.

Evaluation harness

The evaluation module supports hash-bound paired outputs, deterministic balanced blinding, complete score sheets, sealed scores, and a preregistered decision rule that preserves null results.

The included JSON definitions are project-authored synthetic public fixtures. They exercise the public evaluation API and protocol evolution without publishing or summarizing any nonpublic research run. Generate new local results from the public fixtures instead of treating them as historical evidence.

Continuity experiment

The continuity runner creates disposable local stores from synthetic inputs and tests one consented episode followed by a bounded correction. The data key is external to both the SQLite content store and the authority checkpoint.

Read docs/CONTINUITY.md before running it. Never commit a database, checkpoint, master key, real relationship history, or generated evidence containing user input. The ignore rules are a backstop, not permission to handle sensitive data carelessly.

Verification

python -m unittest discover -s tests -v
pnpm --dir ui run lint
pnpm --dir ui run test

The interface test builds the application and exercises the server-rendered HTML. GitHub Actions runs the same Python and interface checks on pushes and pull requests.

Security and publication boundary

Read SECURITY.md and docs/PUBLICATION_BOUNDARY.md before publishing a fork. The development server is local-only and not suitable for internet exposure. The repository deliberately ignores private content, databases, checkpoints, environment files, keys, logs, exports, build products, and model weights.

License

This preview is not open source yet. The temporary LICENSE reserves all rights until the repository owner makes a deliberate software and content license choice. Third-party dependencies remain under their own licenses.