PROVE THE RELEASE IN LAYERS
A version string is not a release. v12.4.0 is identified by an annotated tag, a peeled commit, exact source blobs from that tagged commit, an exact package manifest and the bytes of the package itself. Each evidence class proves one thing. Do not make one file testify about facts it cannot know.
00 Git identity
TAG v12.4.0 TAG OBJECT 28c074777f46bd9e8885f42c49f903ba34e19254 PEELED a1a5c2458065f6f295d1c3e77df62a34846226e3 SOURCE TREE TAGGED COMMIT WORKTREE NOT RELEASE AUTHORITY PUSHED NO
The annotated tag object and the commit are different hashes because they are different objects. That is normal. The tag peels to the release commit. Release-owned source mirrors are read from that immutable commit object, so later work in the mutable checkout cannot silently rewrite v12.4 evidence.
10 package identity
| Name | dbyte-v12.4.0-windows-x64.zip |
|---|---|
| Bytes | 3855311 |
| Entries | 450 |
| Runtime entries | 25 |
| SHA-256 | 433b18bbfd5c36304558a71fe169e631757dbd5285ad16c9097b36e64ee1e55f |
| Source commit in manifest | a1a5c2458065f6f295d1c3e77df62a34846226e3 |
20 packaged executables
| File | Bytes | SHA-256 |
|---|---|---|
| dbyte.exe | 5290496 | 47d2bdf71fd9ae6bbc29006e7285e7d4540f32c799e60fb9f1553dbce41e0e77 |
| dbytec.exe | 3490304 | a9c154da151b5baca4d6f58345c5e28b28aecec1808357faf2b78e0168d1936e |
30 normative source
The v12.4 machine-code contract is mirrored byte-for-byte at NATIVE_MACHINE_CODE_PERFORMANCE.md.
SHA256 b8c9be6fa8482bb881632c479927b9d83d56106ff4b8005cb3baeee335ab5a5a
That document defines what Native MIR owns, which optimization passes exist, what semantics cannot move, how allocation works, what the benchmark corpus measures and what is explicitly outside the milestone.
A checksum proves bytes. It does not prove your interpretation of the bytes. Keep the layers straight.
40 development receipts versus final identity
The retained performance and code-quality receipts under the compiler workspace were generated during the v12.4 development run. They point at the final candidate commit and contain the measured values, but at least the performance receipt records candidate_dirty: true. That makes it useful measurement evidence and bad final-tag evidence.
The final release identity on this page therefore comes from the annotated tag object, its peeled commit, the contract blob read directly from that commit, and the exact package manifest and package bytes. A later dirty working checkout is a different state and cannot mutate this evidence.
THIS IS NOT PEDANTRY. If you collapse “same commit” and “same working tree” into one idea, you will eventually ship bytes you cannot reconstruct.
50 independent package construction
The release closeout built the Windows package from two independent exact-tag export roots and obtained byte-identical archives. The root archival ZIP was then replaced with the verified tagged-export artifact. This is stronger than checking only the source commit: packaging order, file membership and archive metadata must also converge.
60 stale text is not release identity
Long-lived repositories contain historical documents. A version banner inside an old language spec or install note can be stale without changing the current release. That is why this site uses the release tag, package receipt and current normative owner instead of scraping every old prose file for the largest number.
70 verify it yourself
git -C D:\TECHNICAL\ProgramingLangPJ rev-parse refs/tags/v12.4.0 git -C D:\TECHNICAL\ProgramingLangPJ rev-list -n 1 v12.4.0 git -C D:\TECHNICAL\ProgramingLangPJ show v12.4.0:docs/NATIVE_MACHINE_CODE_PERFORMANCE.md > .\v12.4-contract.tag.md Get-FileHash .\v12.4-contract.tag.md -Algorithm SHA256 Get-FileHash .\dbyte-v12.4.0-windows-x64.zip -Algorithm SHA256
The expected package hash is 433B18BBFD5C36304558A71FE169E631757DBD5285AD16C9097B36E64EE1E55F. If your bytes disagree, stop. Do not explain the mismatch away.