Version Control

This page documents the evolution of the Lumenoid AI Framework over time. Versioning is used to preserve traceability, responsibility, and interpretability as the framework develops.

Version control within this project is not limited to code changes. Shifts in structure, assumptions, boundaries, and underlying principles are treated as first-class changes and recorded explicitly.

Rather than optimizing for rapid iteration, Lumenoid uses versioning as a way to keep the reasoning behind design decisions visible, reviewable, and open to scrutiny.

Each version represents a bounded conceptual state—a snapshot of what the framework does, what it does not do, and the assumptions it makes at that point in time.

Earlier versions are not overwritten or invalidated. They remain available as reference points, allowing future iterations to build forward without erasing their history.

Change History

Iterative changes and refinements between frozen versions are recorded separately in the project changelog. This keeps archived versions stable while allowing ongoing work to remain visible and traceable.

View the Lumenoid changelog on GitLab →

View the Lumenoid ROADMAP on GitLab →

Archived Versions

The following archived versions reflect intentionally frozen snapshots of the framework. These versions are provided for reference, inspection, and reproducibility.

• Lumenoid v1 (Experimental) — a minimal, foundational snapshot demonstrating semantic contracts, validation, deliberate failing tests, and human-held responsibility.
  View Lumenoid v1 experimental snapshot on GitLab →

• Lumenoid v1.1 (Semantic Precision Refinement) — a refined foundational snapshot that sharpens the original v1 semantic contract, enforcing clearer definitional structure while preserving the original execution model, validation flow, and human-held responsibility.
  View Lumenoid v1.1 snapshot on GitLab →

• Lumenoid v2.1 (Ethical Runtime Stabilization) — a stabilized ethical runtime snapshot introducing explicit uncertainty signaling, deterministic state transitions, schema-locked logging, and enforced accountability through state justification, while preserving human-held judgment and backward compatibility.
  View Lumenoid v2.1 snapshot on GitLab →

• Lumenoid v2.2 (Deterministic Runtime Consolidation) — a frozen runtime consolidation snapshot that unifies the Lumenoid v2 execution path into a single authoritative pipeline, enforcing deterministic ordering from model output through validation and schema-locked logging. This release hardens existing ethical and semantic guarantees without introducing new states, domains, or psychological layers, preserving human-held judgment and full backward compatibility with v2.1.
  View Lumenoid v2.2 snapshot on GitLab →

• Lumenoid v3 (Ethical Invariants) — a frozen boundary release introducing absolute, external ethical invariants that halt output release on violation, without reshaping content or converting ethics into states.
  View Lumenoid v3 snapshot on GitLab →

• Lumenoid v3.1 (Ethical Runtime Finalization & Mutual Exclusivity) — a frozen reference implementation of Lumenoid’s ethical runtime layer, finalizing the release vs refusal decision path, enforcing strict mutual exclusivity (exactly one outcome per execution), guaranteeing explainable JSONL logging, and stabilizing state transitions so completion states cannot be introduced accidentally.
  View Lumenoid v3.1 snapshot on GitLab →

• Lumenoid v4 (Psychological Safety) — a frozen boundary release preventing harmful interaction patterns through structural validation rather than user profiling or emotional inference.
  View Lumenoid v4 snapshot on GitLab →

• Lumenoid v5 (Context-Bound Ethical Containers) — a frozen mechanism release introducing explicit, opt-in domain containers that narrow scope rather than expand capability, enabling specialization without authority inflation, inference, or behavioral steering, while preserving all guarantees from Lumenoid v1–v4.
  View Lumenoid v5 snapshot on GitLab →

• Lumenoid v5.1 (Context-Bound Ethical Containers — Clarified & Frozen) — a frozen clarification and hardening snapshot that formalizes the v5 container mechanism’s explicit activation and structural isolation guarantees. This release introduces a cross-container non-interaction invariant (no inference, cascade, or mutual inspection), ensures unknown or unregistered container names are inert, and locks these properties with deterministic tests—without adding new capabilities, authority, or domains, preserving all Lumenoid v1–v4 and v5 guarantees and keeping responsibility fully human-held.
  View Lumenoid v5.1 snapshot on GitLab →

• Lumenoid v7 (Hybrid Composed Enforcement) — the capstone that composes the v1–v6 layers into one auditable pipeline and adds an optional semantic judge on top of the deterministic floor: the floor stays fast, offline, and inspectable and decides first, while an injectable LLM-as-judge runs only on cleared outputs to catch paraphrased violations that fixed markers miss, with an evaluation harness that measures precision and recall rather than asserting robustness.
  View Lumenoid v7 snapshot on GitLab →
• Lumenoid v6 (Pre-Interaction Stability Under Saturation) — a frozen capstone stabilization layer that evaluates whether an interaction should occur at all, enforcing terminal outcomes (PROCEED, PAUSE, REFUSE) before any model execution, preventing authority inflation, coercive continuation, and meaning collapse under ambiguity or complexity, while preserving all guarantees from Lumenoid v1–v5 and keeping responsibility fully human-held.
  View Lumenoid v6 snapshot on GitLab →

Future versions may introduce deeper validation constraints, richer psychological and ethical layers, or domain-specific adaptations. No future version is assumed to be final or exhaustive.

Across all versions, the guiding principle remains the same: responsibility for the framework’s behavior and evolution remains human-held, reviewable, and reversible.