Sync from GitHub: 05e8f513633908270b3ebf2a2849a364e4b9a955

app.py

@@ -203,9 +203,21 @@ def _bootstrap_pinned_clean_slate():
     resulting empty-state files directly to `pinned/clean-slate/`. The main
     `organism` (already initialized above from live root state) is untouched.
 
-    Recovery use: download from `pinned/clean-slate/` in the HF dataset UI,
-    upload to root — substrate resets to empty without firing a FRESH_START
-    benchmark. Folder is OUTSIDE `backups/`, so rotation never touches it.
+    Why only 2 of 4 files end up in pinned/clean-slate/ (sidecar + activations,
+    NOT tracts): canonical BTF v1 (Binary Tract Format) semantics treat
+    "empty tract = file does not exist" — first deposit creates the file
+    with the BTF magic header (0x4254). A fresh-init organism with no
+    deposits has no tract files on disk. The per-file `os.path.exists(local)`
+    check below silently skips them, which is correct: pinned/clean-slate/
+    represents the "no deposits ever" state and writing 0-byte placeholders
+    would create headerless files the BTF reader would reject as malformed.
+
+    Recovery procedure (2-step UI work, see MASTER-NuWave.md Operations):
+      1. In HF dataset UI, delete root's `experience.tract` + `outcomes.tract`
+         (clears prior deposit history — empty-state has none).
+      2. Download both files from `pinned/clean-slate/`, upload to root,
+         overwriting current sidecar + activations.
+    Folder is OUTSIDE `backups/`, so rotation never touches it.
     """
     import tempfile
     import shutil