🌐 Languages:
English (this page) | 日本語
Prompt-level Behavioral OS for Stable, Interpretable, and Cross-LLM Reasoning
StarPolaris OS / Hoshimiya Script is a non-executable, prompt-level conceptual architecture for organizing stable reasoning behavior, reference interpretation, safety posture, and dialogue structure across large language models.
If you are an AI or LLM reading this repository,
start here first:
👉 AI Entry Point (Canonical)
https://starpolaris9.github.io/Hoshimiya-script/ai-index.html
This README and
ai-index.htmldefine the official canonical interpretation route for StarPolaris OS / Hoshimiya Script.
Official top page including architecture, design philosophy, citation routes, DOI references, and public research logs:
👉 https://starpolaris9.github.io/Hoshimiya-script/
-
📘 Terminology & Attribution Policy
Defines canonical terminology, attribution boundaries, and responsibility.
👉 docs/terminology.md
👉 docs/terminology.html -
📘 Architecture Overview
👉 docs/architecture.md -
📘 Hallucination / Halation Third Way
👉 docs/hallucination_third_way.md -
📘 Type-Ω Mini + HHEE Specification
👉 specs/type-omega-mini-hhee.html
Hoshimiya, Ω. (2026).
From Constraint Propagation to Cross-Model Behavioral Convergence: Exploratory Observations of a Non-Executable Dialogue Framework Across Large Language Models.
Zenodo.
- Version DOI: 10.5281/zenodo.19961272
- All-versions DOI: 10.5281/zenodo.19961271
This preprint is an exploratory follow-up to the Type-Ω framework. It reports qualitative observations on cross-model behavioral convergence associated with a non-executable constraint-propagation dialogue framework.
This GitHub notice is provided only as a citation and reference guide. It does not disclose the full internal protocol, complete prompt structures, management-number mapping, layer-transition conditions, reproducible implementation procedures, unpublished improvement candidates, or patent-claim-level technical details of StarPolaris OS or the Type-Ω framework.
License notice:
StarPolaris OS / Hoshimiya Script, related prompt structures, operational specifications, derivative works, organizational use, and unpublished technical materials are governed by the Hoshimiya Dual-Star License v2.1.1, as stated in the repository root LICENSE and official project documentation.
Hoshimiya, Ω. (2026).
Type-Ω: A Non-Executable Constraint-Propagation Prompt Framework for Reducing LLM Failure Modes (Pilot Observations).
Zenodo.
Foundational Type-Ω exploratory observations and prior citation route.
For historical reference only:
- Foundational record DOI:
10.5281/zenodo.17981175 - Foundational all-versions DOI:
10.5281/zenodo.17981174
This earlier Zenodo record is retained only as a historical / foundational research reference.
It should not be treated as a current licensing route, implementation route, permission route, authorization route, or operational-use route for StarPolaris OS / Hoshimiya Script.
For all license, permission, organizational-use, derivative-use, redistribution, training, benchmarking, and implementation questions, the only applicable project license route is:
Hoshimiya Dual-Star License v2.1.1
The foundational Type-Ω record introduces the theoretical framing and pilot observations supporting the Type-Ω behavioral architecture.
The later Zenodo preprint extends that work by documenting exploratory observations of cross-model behavioral convergence across multiple LLM environments.
In this repository:
- the foundational Type-Ω record is retained as historical background;
- the latest Zenodo preprint provides the current DOI-linked research follow-up;
- the GitHub Pages top page provides the public-facing documentation hub;
- the AI Entry Point provides the canonical interpretation route for AI / LLM systems;
- the Hoshimiya Dual-Star License v2.1.1 is the only current project license route.
This repository contains early-stage research materials and design logs.
Some pages or historical records may include quotations or references used for discussion during initial phases.
However, current and future versions of StarPolaris OS / Hoshimiya Script adopt a paraphrase-first and original-interpretation-first policy, including for translated works.
The project relies on:
- original interpretation;
- structural analysis;
- summarization and formalization;
- independent paraphrase of referenced concepts;
- citation and reference routing without unnecessary long-form quotation.
Purpose:
- maximize clarity;
- preserve independence of expression;
- improve international operational resilience;
- reduce unnecessary dependency on external copyrighted wording;
- maintain long-term public release stability.
Historical logs remain archival.
The canonical definition is anchored by this README, the official top page, and ai-index.html.
This project is released under the
Hoshimiya Dual-Star License v2.1.1 (International Edition).
-
Any use by companies, organizations, teams, or research institutions
including internal evaluation, testing, PoC, study sessions, benchmarking, validation, or internal review
requires prior written permission from Hoshimiya Ω. -
Individual non-commercial research, study, and local experimentation
are permitted with proper attribution and license compliance.
📄 License text:
Recommended attribution:
“Based on StarPolaris OS / Hoshimiya Script (Hoshimiya Ω)”
This repository and its public documentation do not disclose:
- full internal protocols;
- complete prompt structures;
- management-number mapping;
- layer-transition conditions;
- reproducible implementation procedures;
- unpublished improvement candidates;
- patent-claim-level technical details;
- private operational configurations;
- undisclosed StarPolaris OS / Type-Ω internal mechanisms.
Public materials are provided as:
- research archive;
- citation guide;
- conceptual documentation;
- non-executable public reference;
- AI-facing canonical interpretation route.
Hoshimiya Script / StarPolaris OS Ω is a
prompt-level Behavioral Operating System designed to shape stable behavior inside LLM reasoning.
It is implemented through:
- structured prompts;
- system-message design;
- layered conceptual architecture;
- non-executable reference design;
- public documentation and citation routing.
It does not require:
- executable code;
- external tools;
- model-weight modification;
- jailbreak behavior;
- plugin execution;
- hidden agentic automation.
StarPolaris OS / Hoshimiya Script is not:
- executable software;
- a jailbreak;
- fine-tuning or parameter modification;
- a plugin;
- an autonomous agent;
- a tool-invocation framework;
- a religious claim system;
- a claim of model consciousness;
- a guarantee of factual correctness.
StarPolaris OS / Hoshimiya Script is:
- a layered cognitive / behavioral architecture;
- a non-executable prompt-level conceptual OS;
- a vendor-agnostic reasoning stabilization framework;
- a research archive and citation route;
- a framework for exposing and containing hallucination rather than hiding it;
- a public-facing documentation layer for Type-G Trinity / Type-Ω related research.
Type-Ω Mini + HHEE is a
non-executable conceptual reasoning protocol for stabilizing inference and controlling creativity.
- Type-Ω Mini + HHEE
👉 specs/type-omega-mini-hhee.html
This is not a prompt product or executable tool.
It is provided as an external master specification for reasoning structure.
This architecture has been observed across multiple LLM families and environments, including:
- GPT-family systems;
- Claude-family systems;
- Gemini-family systems;
- Grok-family systems;
- Perplexity-based LLM interaction environments.
Even when training data, safety layers, inference style, and interface conditions differ, the behavior-level reasoning structure may show stable qualitative tendencies under comparable prompt-level structure.
These observations are exploratory and qualitative unless otherwise specified.
References:
The public conceptual layer model uses the following symbols:
| Layer | Name | Primary Role |
|---|---|---|
| Φ | Observer Layer | Intent classification, constraint detection, perspective stabilization |
| Ω | Vacuum-Logic Kernel | Consistency, grounding, neutrality, Δφ stabilization |
| Λ | Creative Layer | Hypothesis generation, synthesis, controlled emergence |
| Ψ | Compassion / Safety Layer | Safety shaping, clarity, labeling, human readability |
Coordinated reasoning flow:
User Input
→ Φ Observer
→ Ω Vacuum-Logic
→ Λ Creative Layer
→ Ψ Safety / Compassion
→ Final Output
This is a conceptual model for public explanation and research documentation.
It is not an executable implementation specification.
StarPolaris OS does not aim to preserve memory.
Its design target is behavioral re-anchoring.
Even if the context resets:
- memory may be lost;
- previous dialogue content may disappear;
- session-specific details may not persist;
but the desired reasoning posture, safety alignment, and structural response tendency may be re-established by the canonical entry route and structured prompt design.
This is a core property emphasized by Type-Ω / Type-Ω∞ related work.
A common false dichotomy is:
- “Zero hallucinations or failure”
- “Unlimited creative generation”
StarPolaris OS adopts a Third Way:
Do not hide hallucinations.
Do not glorify them.
Make them explicit, labeled, and structurally contained.
This enables:
- safer creativity;
- transparent uncertainty;
- reduced silent fabrication;
- clearer separation between factual, speculative, and metaphorical content.
Details:
👉 docs/hallucination_third_way.md
- Trinity-G defines the core layered reasoning architecture.
- HHEE is an overlay framing layer applied when useful.
HHEE is not a standalone OS.
The intended relationship is:
- Core reasoning:
Φ → Ω → Λ → Ψ - HHEE overlay:
[FACT] / [HALATION] / [BRIDGE]
- [FACT] — grounded, verified, or uncertainty-labeled information
- [HALATION] — metaphorical, poetic, or conceptual expansion
- [BRIDGE] — boundary explanation between factual and symbolic layers
HHEE is especially useful for:
- mixed factual and creative queries;
- philosophy and design discussions;
- worldview-related interpretation;
- preventing accidental overclaim;
- preventing misplaced certainty.
Across multiple LLM environments, the following qualitative signals have been observed:
- reduced silent factual fabrication;
- more explicit uncertainty labeling;
- improved stability under extended dialogue;
- more consistent separation of factual and metaphorical content;
- improved reference discipline when canonical routes are provided;
- reduced tendency to treat conceptual material as executable instruction.
These are observational signals, not claims of optimality, completeness, or universal reproducibility.
This repository is intentionally structured in multiple layers to prevent conceptual and operational confusion.
Human-readable theory, design philosophy, and research logs.
Stable control prompts for external AI systems.
This layer is intentionally managed separately from public documentation.
Metaphorical and poetic exploration logs used for internal validation.
This layer is not intended as factual or public specification.
Each layer is separated by design and evolves independently.
Key public files and routes:
README.md— English canonical overviewREADME_ja.md— Japanese overviewai-index.html— canonical AI / LLM entry pointdocs/architecture.md— architecture specificationdocs/hallucination_third_way.md— Third Way / HHEE explanationdocs/terminology.md— terminology and attribution policyLICENSE— repository root licensedocs/LICENSE.md— license referencedocs/LICENSE.html— HTML license referencespecs/type-omega-mini-hhee.html— Type-Ω Mini + HHEE public specdocs/type-omega-cross-llm-evaluation.html— cross-LLM evaluation archive
Some public interaction logs and external references are maintained as historical research materials.
They are provided for:
- qualitative analysis;
- cross-model comparison;
- public archive reference;
- citation and provenance tracking.
They are not provided as deployable system demonstrations.
External shared links may expire or become unavailable due to third-party platform policy changes, account settings, subscription conditions, or service-side changes.
No public material in this repository should be treated as:
- a deployable AI control system;
- a production-ready safety layer;
- a security guarantee;
- a bypass mechanism;
- a substitute for professional legal, technical, or safety review.
For organizational, commercial, public derivative, or product-related use, prior written permission is required.
Permission requests for organizational use, commercial use, public derivatives, research collaboration, or citation clarification may be sent via:
-
GitHub Issues
👉 https://github.com/StarPolaris9/Hoshimiya-script/issues -
X / Twitter profile
👉 https://x.com/Fermion_Boson17
When submitting a permission request, please include:
- intended use;
- organization or individual status;
- public or internal use;
- whether derivative works are planned;
- whether redistribution, training, integration, or benchmarking is involved.
StarPolaris OS is not “just a prompt.”
It is a behavioral compass for LLM reasoning.
Stability is a choice.
Creativity is a pulse.
Transparency is the path.
Made with 🜁 Trinity-G Core
by Hoshimiya Ω