Record ModelScope mirror receipts (R11)

acceptance/PR-M1.md

@@ -539,3 +539,25 @@ Post-upload verification, all measured:
   independent process rc=0 with `acquisition_calls=0` (memory replayed),
   and the 28-test suite run from the cloned `tests/` against the installed
   wheel: **28 passed**.
+
+### ModelScope mirror (owner-approved, same day)
+
+Channel matrix note: ModelScope is domestic and runs **direct, no proxy**; the
+token (`ms-…`, short-lived, write role) validated via the SDK's `whoami`
+(account `kialajin`).
+
+- Repo created public: `https://modelscope.cn/models/kialajin/cognitive-learning-lab`.
+- Upload: 14 files **byte-identical** to the HF repo at HEAD `d0756e93…`
+  (10 hash-surface docs incl. `acceptance/PR-M1.md`, manifest, SHA256SUMS,
+  `pyproject.toml`, `requirements-lock.txt`, wheel, sdist) — upload commit
+  `d0aaacbffe…`, 14 committed / 0 failed, wheel stored via LFS.
+- Download-back (`snapshot_download@master`): every `SHA256SUMS` entry whose
+  path exists in the flat remote layout matches byte-for-byte (10/10);
+  wheel/sdist hashes match the current build directly
+  (`64e02c1a…` / `d69ad3ce…`). The 35 "missing" entries are the flat-`src/` /
+  `dist/` layout difference of `SHA256SUMS`, the same structural note as on
+  HF, not a defect — artifact integrity is verified through the artifacts
+  themselves.
+- README's YAML frontmatter (HF card format) is shipped unchanged to keep the
+  two channels byte-identical; ModelScope tolerates it. Model channel
+  status: HF primary, ModelScope mirror; no weights on either.