Finalize CADForge mini blog evidence

CADFORGE_BLOG.md

@@ -109,7 +109,7 @@ The real run used Unsloth for LoRA SFT and TRL GRPO for environment reward train
 The raw logs are public so the training claims are inspectable, not just summarized:
 
 - Training evidence dataset: [sanjuhs/cadforge-training-evidence](https://huggingface.co/datasets/sanjuhs/cadforge-training-evidence)
-- Local backup in this repo: `training/backups/cadforge-training-evidence-20260426`
+- Compressed archive on that dataset: `archives/cadforge-training-evidence-20260426.tar.gz`
 - Per-sample reward traces: `training/logs/*completions.jsonl`
 - Generated plots and parsed metrics: `training/reports/*`