v1.3.2-stable

Release v1.3.2-stable: JOSS Submission Preparation

Sanitized Release | Numerical Constitutionalism | Hardened Documentation

This is a maintenance release focused on hygiene, documentation alignment, and version finalization in preparation for the official JOSS (Journal of Open Source Software) submission. No modifications were made to computational kernels or execution semantics.

🛡️ Numerical Constitutionalism

• Auditability: Formalized the "Numerical Constitution" in documentation, ensuring all optimized kernels are strictly tethered to the bit-exact Scalar C++ Reference Implementation.
• Determinism Boundaries: Refined language regarding intra-platform determinism and bounded cross-architecture drift.

🛠 Maintenance & Hygiene

• Comment Sanitation: Removed redundant AI-style commentary and speculative internal roadmap references from the core engine.
• Documentation Alignment: Standardized terminology across the repository, Wiki, and JOSS manuscript.
• Version Synchronization: Aligned versioning across Python, Swift, and C++ layers.

📦 Distribution

• Prebuilt Binaries: Certified macOS ARM64 and Linux x86_64 binaries updated with v1.3.2-stable metadata.
• Package Integrity: Validated pip-installable wheel and Swift Package Manager compatibility.

Release v1.3.1: Tooling & Distribution

Standalone Introspection | Cross-Platform Packaging | Installation UX

This release marks the completion of the Tooling and Distribution phase. It introduces a comprehensive suite for execution trace analysis, visualization, and cross-platform packaging, significantly reducing the friction for installation and numerical verification.

🔍 Tooling & Introspection

• Trace Analysis: New standalone tools (trace_analyzer.py) to compute execution order, kernel breakdown, and memory footprints.
• Determinism Verification: Introduced trace_diff.py for bitwise comparison of execution traces, enabling automated determinism checks.
• Visual Execution Flow: Support for generating SVG timelines and DOT graphs from any VECTORIA trace.
• Trace CLI: Unified vectoria-trace command-line utility for all introspection tasks.

📦 Distribution & Packaging

• Python Wheel: Official pip-installable package supporting automated dependency management and CLI integration.
• Swift Package Manager: Finalized SPM support with a new VectoriaExample executable target.
• Prebuilt Binaries: Released certified libvectoria.dylib for macOS (ARM64) with cryptographic checksums.
• Portable Native Loading: Dynamic library loading logic updated to support both packaged and developer environments.

🛠 CI/CD & Documentation

• Cross-Platform CI: New validation workflows for Ubuntu and macOS to ensure packaging and tooling integrity.
• Comprehensive Docs: New Installation, Getting Started, and Determinism Verification guides integrated into the Wiki and repository.
• Semantic Integrity: No modifications to core kernels or execution semantics. This release observes the v1.3.0-stable freeze.

🚀 Getting Started

pip install vectoria
vectoria-trace --help

Tooling & Distribution

Tooling & Distribution

Standalone Introspection | Cross-Platform Packaging | Installation UX

This release marks the completion of the Tooling and Distribution phase. It introduces a comprehensive suite for execution trace analysis, visualization, and cross-platform packaging, significantly reducing the friction for installation and numerical verification.

🔍 Tooling & Introspection

• Trace Analysis: New standalone tools (trace_analyzer.py) to compute execution order, kernel breakdown, and memory footprints.
• Determinism Verification: Introduced trace_diff.py for bitwise comparison of execution traces, enabling automated determinism checks.
• Visual Execution Flow: Support for generating SVG timelines and DOT graphs from any VECTORIA trace.
• Trace CLI: Unified vectoria-trace command-line utility for all introspection tasks.

📦 Distribution & Packaging

• Python Wheel: Official pip-installable package supporting automated dependency management and CLI integration.
• Swift Package Manager: Finalized SPM support with a new VectoriaExample executable target.
• Prebuilt Binaries: Released certified libvectoria.dylib for macOS (ARM64) with cryptographic checksums.
• Portable Native Loading: Dynamic library loading logic updated to support both packaged and developer environments.

🛠 CI/CD & Documentation

• Cross-Platform CI: New validation workflows for Ubuntu and macOS to ensure packaging and tooling integrity.
• Comprehensive Docs: New Installation, Getting Started, and Determinism Verification guides integrated into the Wiki and repository.
• Semantic Integrity: No modifications to core kernels or execution semantics. This release observes the v1.3.0-stable freeze.

🚀 Getting Started

pip install vectoria
vectoria-trace --help

Semantic Surface Frozen | Documentation & Reproducibility Consolidated

This is a stabilization release that formalizes the semantic surface of VECTORIA following the completion of the Phase 8 Transformer stack. As of this release, the mathematical definitions and expansion logic for all operations are frozen to ensure a reliable foundation for downstream integration.

🛡️ Semantic Freeze
The expansion logic for high-level operations (LayerNorm, Attention, MHA, EncoderBlock) is locked.
Numerical behavior is governed by the constitutional TRUTH.md.
Any future performance optimizations (e.g., kernel fusion) must prove bitwise identity to the frozen reference expansions.
🌟 Capabilities (v1.3.0)
Complete Transformer Stack: Supports full Transformer Encoder Blocks using purely semantic, composed graphs.
Structural Integrity: Full support for Transpose, Reshape, Concat, and Slice operations with explicit IR nodes.
Numerical Stability: Numerically robust LogSoftmax and StableSoftmax implementations integrated by default.
Auditability: Canonical tracing walkthroughs ensure that the execution of complex blocks is fully transparent.
🛠 Hardening & Documentation
Documentation Unification: Repository-wide audit to ensure consistent terminology and mathematical formatting (LaTeX).
CI Reproducibility: Hardened documentation for CI validation paths and reproducibility checklists.
No Code Changes: This release introduces no new numerical kernels or performance shortcuts, preserving the proven stability of the v1.2.1 and v1.3.0 semantic baseline.
⚠️ Notes
Inference-Only: VECTORIA remains an inference-forward framework.
Single-Threaded: The core engine remains strictly serial to preserve bitwise determinism.