25 lines
1.6 KiB
Markdown
25 lines
1.6 KiB
Markdown
|
# Published reference versions
|
||
|
|
|
||
|
|
[releases.json](releases.json) selects the current qualified preview (`0.1`) and pins
|
||
|
|
its packaging commit and manifest digest. Consumers record that version, commit and
|
||
|
|
digest in their use receipt. Candidate prose and other corpus inputs are resolved at
|
||
|
|
the manifest's `corpus_revision`; packaging documents and evidence locators are
|
||
|
|
resolved at the packaging revision. Relative links viewed on a moving branch can
|
||
|
|
show newer research: use a checkout or commit-specific source URL for a pinned read.
|
||
|
|
|
||
|
|
Run `python3 scripts/reference_preview.py --check --version 0.1` to validate a
|
||
|
|
published version, or `make reference-check` for the current pointer. The local Git
|
||
|
|
object database must contain both pinned revisions; a shallow clone must fetch that
|
||
|
|
history first. Missing objects fail validation rather than substituting live files.
|
||
|
|
|
||
|
|
`make evolution-check classification-check` separately validates current research.
|
||
|
|
A legitimate candidate/profile revision must pass those checks but must neither
|
||
|
|
rewrite nor invalidate the older export. Published manifest, development cases and
|
||
|
|
release documents are checked against their original packaging commit. Generated
|
||
|
|
outputs must match the pinned inputs. The registry is a reviewed publication index,
|
||
|
|
not a signature or an independent trust root; changing a release pin requires review.
|
||
|
|
|
||
|
|
A future publication gets a new directory and registry entry. Move `current` only
|
||
|
|
once that publication is verified. The frozen preview remains a research hypothesis
|
||
|
|
catalog, not canonical admission or a held-out retrieval evaluation.
|