[Lee-Juntong]

The native Windows decryptor checks a few decrypted header bytes on page 1 but does not authenticate pages. Altered ciphertext in a later page can therefore be accepted and atomically replace a previously good snapshot.

Derive the SQLCipher v4 MAC key and verify every stored page's HMAC-SHA512, including the page number and IV, before producing plaintext or replacing the destination. Reject mismatched authentication without modifying existing output. Preserve the existing 4096-byte page / 80-byte reserve profile.

Validation: nine synthetic encrypted-page tests cover a correct key, wrong key, truncation, ciphertext/IV/MAC corruption, page-number mismatch, and preservation of previous output. The corruption/page-number/output-preservation regressions fail on main; all added tests pass with this change. Live account-wide native key extraction is still unverified.

Local validation uses Windows/Python 3.12 and synthetic fixtures. The consistency-checker file and two fingerprint-dependent existing tests were excluded by the local execution policy; no account databases, chat contents or keys are included.

Local branch suite: 90 passed, two tests deselected, consistency-checker file excluded.

[freanocank]

Thanks for the fix and regression tests. Adopted through #15 with your original commits preserved. The combined WAL regression also proves that valid WAL checksums cannot bypass page authentication, and rejection preserves the previous output.

The combined suite passes all four macOS/Windows × Python 3.10/3.13 CI jobs; Windows reports 209 passed, and macOS reports 208 passed with the Windows-only test skipped.