fractalsense-ai/lumina-os
A zero-trust, deterministic orchestration layer that secures AI reasoning behind immutable Domain Physics — giving the LLM exactly one job: high-weight reasoning.
What's novel
A zero-trust, deterministic orchestration layer that secures AI reasoning behind immutable Domain Physics — giving the LLM exactly one job: high-weight reasoning.
Code Analysis
9 files read · 3 roundsA deterministic orchestration framework that wraps LLM calls in layered security contracts — assembling prompts from domain physics + state + actor evidence, verifying all outputs through invariant checking and schema validation, and maintaining hash-chained audit logs with human-in-the-loop escalat
Strengths
Genuinely sophisticated architecture: the three-collaborator PPA orchestrator (Resolver/Drafter/Writer), the immutable state machine, the hash-chained system log with auto-freeze on escalation, and the dynamic model-pack loading with provenance hashing are all real, non-trivial implementations. The separation between the deterministic engine and domain-specific model-packs is cleanly executed through the runtime loader's dynamic callable loading and the persona builder's role-narrowed prompt ass
Weaknesses
The invariant checker's expression language is limited to simple field comparisons (no nested expressions, no logical AND/OR), which constrains domain-pack expressiveness. Some modules are quite large (runtime_loader at 31kb, ppa_orchestrator at 23kb) and the backward-compatibility delegate methods in the orchestrator add maintenance surface. The SLM layer, while well-structured, is essentially a thin HTTP wrapper around Ollama/OpenAI/Anthropic with prompt engineering rather than novel ML logic.
Score Breakdown
Signal breakdown
Innovation
Craft
Traction
Scope
Evidence
Commits
178
Contributors
2
Files
328
Active weeks
3
Repository
Language
Python
Stars
1
Forks
0
License
Apache-2.0