[hyperpolymath]

Change and reason

Codecov reported 52.38% patch coverage on JoshuaJewell/MetaManifold-WebUI#70, with 30 lines missing. All 30 lines are in src/analysis/differential.jl. They are the adapter that turns a CompositionalDA result into result rows.

The cause is real. CompositionalDA is not in the General registry, so CI cannot install it. In CI, only the refusal path ran.

This change makes the adapter reachable in CI:

• differential_abundance(...; native_backend=nothing) calls a module. The default is the CompositionalDA that the session loaded, or the by-name refusal when the session loaded none. Production behaviour does not change.
• _is_composition_error translates the CompositionError of the backend in use.
• diagnostics.native_method.package reports the name of the module that produced the numbers. A record from the stand-in says NativeDAStub, so it cannot read as a record from the package.
• test/support/native_da_stub.jl is a stand-in with the two entry points, the column names and the verdict strings of the package. Its qvalue is not bh_adjust(pvalue), so a second BH pass is visible. Its structural rows carry NaN, as the package leaves them.
• test/unit/test_differential_native.jl gains four testsets that run on every checkout: the seam, the ALDEx2 wiring, the ANCOM-BC wiring and the refusal order.

The comparisons against the real package stay. They still skip with a message when the package is absent.

This pull request targets the branch of upstream PR 70. A merge here updates that pull request.

Validation

Julia 1.12.5, R 4.3.3 (Ubuntu 24.04 r-base-core), all commands from the repository root.

1. Coverage, with the same flags as CI:

   julia --project=. --code-coverage=user --compiled-modules=no test/runtests.jl differential differential_native

   Result: 1159 pass, 13 skipped, 0 fail. I compared the .cov file with git diff against the merge base with upstream main (89be534). Every instrumented line that this branch adds to src/analysis/differential.jl is covered: 52 of 52.

2. Real package. In a scratch copy I ran Pkg.develop on hyperpolymath/CompositionalDA.jl at 9903e76 (v0.3.0) and ran the same command. Result: 1367 pass, 11 skipped, 0 fail. The two real-package testsets ran and agreed with the adapter.

3. Full unit suite: julia --project=. -t 2 test/runtests.jl. Result: 3618 pass, 25 skipped, 1 error. The error is test_provenance.jl:479: "package 'dada2' is not installed in the R library in use". My sandbox does not have the R packages that renv.lock installs in CI. The test does not touch this change.

4. Planted mutants. I ran 12 deliberate corruptions of the adapter, one at a time, against test/runtests.jl differential_native. The baseline passed. The suite killed 12 of 12:

   mutant	result
   BH applied a second time	killed (8 failures)
   estimate taken from the wrong column	killed (4)
   log(10) in place of log(2)	killed (3)
   structural decisions ignored	killed (1 failure, 1 error)
   contrast direction reversed	killed (2)
   seam bypassed (package checked in place of the backend)	killed (6 failures, 2 errors)
   one retained taxon accepted	killed (5)
   CompositionError not translated	killed (2)
   package name hardcoded in diagnostics	killed (2)
   non-finite statistic published	killed (1 failure, 2 errors)
   aldex_effect taken from the wrong column	killed (4)
   structural notes swapped	killed (2)

   The runner is not part of this change. It counts a kill only when the suite reaches its Test Summary and reports a failure.

Not run: the integration tests, the server tests and the frontend tests. No workflow runs on this pull request: ci.yml on this branch starts only for pull requests into main. The full CI and Codecov run on the upstream pull request, so the Codecov number is not confirmed until PR 70 runs again.

Scope

• No change to Project.toml, Manifest.toml, renv.lock, codecov.yml or any workflow.
• The stand-in tests the wiring of the adapter. It does not test the estimators. The real-package testsets do that, where the package is installed.
• An earlier draft of this work also proposed a patch-coverage gate, a sensitivity test and three estate workflows for the upstream repository. I left them out. They are separate decisions for the upstream owner.

Contribution checks

• I followed CONTRIBUTING.md.
• This branch starts at current upstream main and contains no merges from fork main. (It starts at the head of PR 70, be7792f.)
• Every commit is signed and shows as Verified. (The managed signing identity decides this. See the commit badge.)
• CI has run and passes. (No workflow runs here. See Validation.)
• Existing licensing, citation and acknowledgement text is unchanged.

🤖 Generated with Claude Code

https://claude.ai/code/session_0136P7ZoTPXfxg1C85rVNx78

[coderabbitai]

Important

Review skipped

Auto reviews are disabled on base/target branches other than the default branch.

Please check the settings in the CodeRabbit UI or the .coderabbit.yaml file in this repository. To trigger a single review, invoke the @coderabbitai review command.

Run configuration

• Source mode: Auto
• Repository: Unavailable
• Branch: Unavailable
• File: No configuration file from this repository applied
• Commit: Unavailable
• Layers: Organization UI
• Review profile: ASSERTIVE
• Plan: Advanced
• Run ID: 7cf1c31a-7c07-4e2b-bc32-d6b5161867e7

You can disable this status message by setting the reviews.review_status to false in the CodeRabbit configuration file.

Use the checkbox below for a quick retry:

• Trigger review

• Autofix · Keep fixing CodeRabbit findings and required CI, and resolving merge conflicts

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

Share

• X
• Mastodon
• Reddit
• LinkedIn

[sonarqubecloud]

Quality Gate passed

Issues
0 New issues
0 Accepted issues

Measures
0 Security Hotspots
0.0% Coverage on New Code
0.0% Duplication on New Code

See analysis details on SonarQube Cloud

[chatgpt-codex-connector]

You have reached your Codex usage limits for code reviews. You can see your limits in the Codex usage dashboard.