Ackredit is published to no channel and can only be installed from a clone

[dprada]

What — Ackredit is published to no channel. It can only be installed from a clone, which makes every other roadmap theme unreachable: a host library cannot adopt what it cannot depend on, and a performance measurement of a checkout is not a measurement of a release.

How — conda search -c uibcdf ackredit finds nothing, and PyPI does not carry any MolSysSuite library. The repository has no conda recipe, no build environment and no publishing workflow, while every sibling has all three.

                                          ackredit   smonitor
devtools/conda-build/meta.yaml            NO         yes
devtools/conda-envs/build_env.yaml        NO         yes
build_and_upload_conda_packages.yaml      NO         yes

Why — Roadmap theme A, and the one that blocks the rest. It is also the theme where a mistake is least recoverable: a published package cannot be withdrawn the way a commit can be amended, and devguide/release_version_policy.md forbids moving or deleting a published tag.

That asymmetry decides the shape. A candidate is built and uploaded to uibcdf/label/staging, installed from there into a clean environment and checked, and only then promoted to the public label. Publishing first and verifying afterwards is the wrong order for a registry, and it is the order a simpler workflow would impose.

The test the conda recipe runs should include Ackredit discovering itself, closed in uibcdf/ackredit#21. It is the check that would have caught the packaged CITATION.cff being absent, and a conda test runs against the built artifact rather than the source tree, which is where that defect hid.

Scope — This issue delivers the machinery and verifies it end to end. It does not publish: that needs the channel token and is a release decision, not an implementation one.

Coordination — uibcdf/molsyssuite#27 standardises staging and promotion for components whose publication order is coupled. Ackredit depends on SMonitor and DepDigest and nothing depends on Ackredit, so there is no cycle to break here; the staging step is adopted for verification rather than for coordination.

Record — devguide/pending_proposals/ackredit_cannot_be_installed.md

[dprada]

Machinery delivered and verified in a9f7300. The issue stays open: publishing is the remaining half.

Done — devtools/conda-build/ with a noarch recipe and build script, devtools/conda-envs/build_env.yaml, and build_and_upload_conda_packages.yaml that stages, verifies and promotes.

Verified rather than asserted — conda build produces noarch/ackredit-0.6.0-py_0.conda and its own test block passes; installing that artifact into a clean conda environment resolves the whole chain from the channel (depdigest 0.10.1, smonitor 0.16.0, pyyaml 6.0.3), and the installed package reports 0.6.0, discovers its own CITATION.cff with both authors, renders BibTeX and exposes ackredit --help.

Rejected — the simplest shape, publishing straight to the public label as argdigest does in 63 lines. It verifies after publishing, which for a registry is the wrong order, and a published package cannot be withdrawn the way a commit can be amended.

Not treated as coupled publication — molsyssuite#27 standardises staging where publication order is circular. Nothing depends on Ackredit, so staging is adopted here for verification rather than coordination; the distinction is recorded so the heavier protocol is not assumed to apply.

Remaining — staging a candidate and promoting it, which needs ANACONDA_UIBCDF_TOKEN and is a release decision; then rewriting docs/content/about/installation.md around the published package, which must not happen before it exists.

Record — devguide/pending_proposals/ackredit_cannot_be_installed.md, kept open rather than archived because the theme is not finished.