Clarify updated-rerun provenance

README.md

@@ -101,7 +101,7 @@ paper-final trainable checkpoints against
 The private dataset repository
 [`KylianSu/btecf-dr-handoff`](https://huggingface.co/datasets/KylianSu/btecf-dr-handoff)
 preserves the recomputed AutoMorph masks, Bézier/CLIP features, fixed splits,
-Track 1/2 outputs, and both reconstructed-mask n=500 experiments. An authorized
+Track 1/2 outputs, and both updated n=500 reruns. An authorized
 researcher can restore and checksum all of them with:
 
 ```bash
@@ -109,11 +109,12 @@ hf auth login
 bash scripts/15_download_handoff_assets.sh
 ```
 
-The historical paper n=500 pixel outputs were not retained. The n=500 files in
-the private handoff are a later reconstruction from recomputed masks; they must
-not be described as bit-identical to the submitted-paper run. The historical
-n=50 files in this repository remain the evidence for the submitted Table 4/5
-statistics.
+The current n=500 files are a reproducible post-submission rerun using the
+audited AutoMorph preprocessing. They preserve the fixed-start design and main
+tortuosity conclusion, with small numerical variation from the updated masks.
+They should be cited as the updated rerun rather than as a byte-identical copy
+of the submission snapshot. The archived n=50 files remain the direct evidence
+for the submitted Table 4/5 statistics.
 
 ## Other artifacts you still need (not in this repo)