LICENSE (Hoshimiya Dual-Star License v2.1.1).
This page records one comparative evaluation setup conducted
under controlled prompt conditions using separate LLM instances.
The purpose is to observe whether a prompt-level behavioral specification (Type-Ω)
can preserve consistent reasoning structure and constraint handling
within the same evaluation framework.
Six test prompts were used alongside five predefined behavioral invariants:
- ontological minimality
- contradiction containment
- cross-format invariance
- epistemic transparency
- staged constraint propagation
Within this specific evaluation setup,
the staged constraint propagation indicator
showed an increase from baseline behavior to full adherence.
Within the same setup,
the aggregated achievement score increased from 57.9% to 100%.
These values reflect behavior observed under this configuration only.
They are not intended as general performance claims
nor as benchmarks applicable outside this context.
The results suggest that Type-Ω functions as a prompt-level behavioral specification
rather than as a stylistic or cosmetic prompt pattern.
Based on StarPolaris OS / Hoshimiya Script.
Conceptual research; non-executable; prompt-level specification.
Based on StarPolaris OS / Hoshimiya Script (Hoshimiya Ω).
Conceptual architecture research only. Non-executable. Prompt-level specification. No warranty implied.