chore: sync template files and configuration

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

Author: Jonathan D.A. Jewell

CHANGELOG.md

@@ -0,0 +1,102 @@
+<!-- SPDX-FileCopyrightText: 2024 Joshua Jewell -->
+<!-- SPDX-License-Identifier: MIT -->
+
+# Changelog
+
+All notable changes to this project will be documented in this file.
+
+The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.1.0/),
+and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
+
+## [Unreleased]
+
+### Added
+- RSR (Rhodium Standard Repository) compliance
+- Full documentation suite
+- **Diff-based compression**: Store VAE images as diffs to reduce dataset size by ~50%
+  - New `compress` subcommand to convert VAE images to diffs
+  - New `decompress` subcommand to reconstruct full VAE directory
+  - New `reconstruct` subcommand to reconstruct single VAE image
+  - Julia `CompressedVAEDataset` for on-the-fly VAE reconstruction
+  - Diff encoding: `diff = VAE - Original + 128` (offset for signed values)
+- **Contrastive learning model** for VAE artifact detection (`contrastive_model.jl`)
+  - CNN encoder with ResNet-style residual blocks (embed_dim=256)
+  - Projection head for contrastive learning (proj_dim=128)
+  - Multiple loss functions: NT-Xent, Supervised Contrastive, Triplet, Contrastive
+  - Two-phase training: contrastive pre-training + classifier fine-tuning
+  - Binary classifier for original vs VAE discrimination
+  - Embedding extraction for visualization (t-SNE, UMAP compatible)
+  - Evaluation metrics: accuracy, precision, recall, F1, confusion matrix
+  - GPU support via CUDA.jl (optional)
+  - Justfile recipes: `train`, `train-compressed`, `evaluate`, `embed`, `train-full`
+
+## [1.0.0] - 2024-01-01
+
+### Added
+
+#### Core Features
+- SHAKE256 (d=256) cryptographic checksums for all image files
+- Train/test/validation/calibration splits (70/15/10/5 ratio)
+- Both random and stratified split generation
+- Size-based stratification with configurable bins
+
+#### Metadata & Configuration
+- CUE metadata generation with Dublin Core compliance
+- Nickel configuration schema
+- SPDX headers on all source files
+
+#### Formal Verification
+- Isabelle/HOL proofs for split properties
+  - Disjointness (no overlap between splits)
+  - Exhaustiveness (all images assigned)
+  - Ratio correctness (within 1% tolerance)
+  - Original-VAE bijection preservation
+
+#### CLI
+- `normalize` subcommand for dataset processing
+- `verify` subcommand for output validation
+- `stats` subcommand for statistics display
+- `hash` subcommand for single-file checksums
+- Comprehensive help and examples
+
+#### Infrastructure
+- Nix flake for reproducible builds
+- Podman container with Chainguard Wolfi base
+- Justfile task runner
+- Julia utilities for Flux.jl training
+
+#### Documentation
+- README with usage examples
+- QUICKSTART guide
+- RSR compliance documentation
+- Security policy
+- Contributing guidelines (TPCF)
+- Governance model
+
+### Security
+- Memory-safe Rust implementation
+- No unsafe code blocks
+- FIPS 202 compliant cryptography
+- Non-root container execution
+- Supply chain security (SPDX, pinned deps)
+
+## Migration Guide
+
+### From v0.x to v1.0.0
+
+This is the initial release. No migration needed.
+
+## Version History
+
+| Version | Release Date | Status |
+|---------|--------------|--------|
+| 1.0.0   | 2024-01-01   | Current |
+
+## Deprecation Policy
+
+- Deprecated features announced one minor version before removal
+- Migration guides provided for breaking changes
+- LTS branches may be created for critical security fixes
+
+[Unreleased]: https://huggingface.co/datasets/joshuajewell/VAEDecodedImages-SDXL/compare/v1.0.0...HEAD
+[1.0.0]: https://huggingface.co/datasets/joshuajewell/VAEDecodedImages-SDXL/releases/tag/v1.0.0