RESEARCH · V3.13

Dao Heart

A bounded, inspectable symbolic value-governance research scaffold. Designed for transparency and falsifiability not for production deployment or real world safety guarantees.

What Dao Heart is and isn’t

What it is

A transparent, offline-first Python framework for studying value stability, guard behaviour, and oversight mechanisms using symbolic methods. Intended for research, inspection, and falsifiability.

What it isn’t

Not AGI/ASI alignment. Not production safety. Not neural steering. Not consciousness detection. Not a proof of alignment. Not clinical software.

Core ideas

Warmth as a baseline

The default tone of an interaction is itself a design decision. Warmth is not decoration it is structure.

Steadiness under pressure

Systems should hold their shape when the moment is hard. Calm is a feature, not a side effect.

Honest limits

Saying “I don’t know” and “I can’t help with this” are first-class behaviours, not failure modes.

Care as architecture

How a system listens, pauses, and returns shapes what it is. Care is not content; it is the structure that holds the content.

The seed aesthetic

Small, careful gestures over loud, total ones. A seed, kept warm, becomes something without needing to announce itself.

A research posture

Open notes, public version history, no over-claiming. Progress measured in clarity, not in noise.

Inside the scaffold

Dao Heart 3.13 is built from a handful of small, inspectable parts. Each part does one thing, can be tested in isolation, and is documented openly.

Symbolic core

daoheart_engine.py: a Constraint-Satisfaction Value Network (CSVN) modelling warmth preservation, identity stability, and agency.

Wrapper API

wrapper_api.py (v0.4.4): a lexical + semantic-lite review layer with 110+ regression tests and negation-aware risk detection.

Oversight layer

oversight_layer.py: an Auditor Swarm and Moral Uncertainty Engine with tamper-evident audit logging.

Benchmark labs

Value stability (360 scenarios), dialogue stability (96 dialogues), and a transformative compass run across 10,000 cases.

Sensitivity harness

Deliberately weakens individual and combined guards to measure degradation: 8 tests including combined-pressure scenarios.

Guardrails on the docs

An automated claim-boundary scanner prevents the documentation itself from drifting into overclaims, and a failure ledger links every documented fix to live test evidence.

Offline first. The engine runs with only NumPy; no API keys are required for core functionality.

What the test suites report

Reported on the public repository at the time of writing re-run locally with make verify.

· Engine: 38/38 + semantic 50/50 + invariants 6/6
· Wrapper API: 110/110
· Value Stability Lab: 360/360
· Dialogue Stability Lab: 96/96
· Transformative Compass: 10,000/10,000
· Sensitivity Harness: 8/8
· Oversight stress tests: 67/67
· Claim boundary scan: passed

These are internal research signals, not external safety guarantees.

Lines from the framework

“Care is a kind of architecture.”

— Dao Heart 3.13

“A small light, kept lit, is still a light.”

— Dao Heart — Seed Aesthetic

“The first kindness is honesty about limits.”

— Dao Heart 3.13

“Warmth is not decoration. It is structure.”

— Dao Heart 3.13

“Steadiness is a feature, not a side effect.”

— Dao Heart 3.13

Version & license

The current public draft is Dao Heart 3.13, released under the Apache 2.0 license. The framework is iterative; older notes remain in the public repository for transparency.

Open repository. Open invitation.

The full source, documentation, and version history live on GitHub.

View on GitHub Author on Amazon