Document LFTC dataset and checkpoint snapshot

README.md

@@ -19,7 +19,7 @@ This is the large-asset companion to [NLP-Course-Project-LFTC](https://github.co
 Code, reports, figures, PowerPoint, Excel, metrics and per-example predictions are in GitHub.
 
 The local snapshot contains **2,044 asset files**, **56,530,278,162 bytes**.
-Files under `artifacts/` retain their original project-relative paths, including datasets, prepared inputs, pretrained models, final/local checkpoints, feature bundles, and historical candidate checkpoints.
+Files under `artifacts/` retain their original project-relative paths, including datasets, prepared inputs, pretrained models, final/local checkpoints and feature bundles. Historical `.cache/` candidate checkpoints are stored under `artifacts/legacy-candidates/`, since the Hub reserves `.cache/` paths.
 [artifact-manifest.json](artifact-manifest.json) records paths, sizes, SHA256 values and verification status.
 
 ## Restore
@@ -28,7 +28,7 @@ Files under `artifacts/` retain their original project-relative paths, including
 hf download Eyanhe/LFTC_Dataset_And_Checkpoints --repo-type dataset --local-dir ../lftc-assets
 ```
 
-Merge the contents of the downloaded `artifacts/` directory into the GitHub project root, preserving all relative paths, including hidden candidate directories.
+Merge the downloaded files into the GitHub project root according to the manifest's original `path` field. In particular, restore `artifacts/legacy-candidates/` to the original `.cache/` directory; other `artifacts/` paths map directly to the project root.
 The snapshot covers files actually present on the local workstation; it does not claim that every checkpoint produced on the training cluster has been downloaded.
 
 ## Provenance and scope