Archived MAIC packages leave no maintained CRAN installation path
Source note. This entry derives in part from a source document whose inline citations were ChatGPT interface tokens rather than references, and so resolve to nothing. Claims below are attributed to that document and were checked independently where possible; anything that could not be independently sourced is marked in the verification trail.
This entry describes a limitation of a specific implementation rather than of the field. It is catalogued because the underlying methodological problem is general, and because a reference implementation is where such problems become concrete and checkable.
Statement
Archived and explicitly deprecated MAIC packages are poor foundations for a new reference workflow: CRAN archived maic on 22 July 2025 because email to the maintainer was undeliverable, leaving only versions 0.1.2 to 0.1.4 with the last dated 27 April 2022, and the Roche MAIC README states the package is no longer developed and recommends maicplus. That is a reason to migrate with a documented mapping rather than a reason to call historical analyses unverifiable, since source archives, dependency locks and containers can preserve them. BUGSnet has never been on CRAN, so installation still depends on GitHub plus JAGS, and it carries many long-open issues, several dating from 2019, but it is not without releases: the repository has five version tags through v1.1.2 on 22 June 2024. It has no native PAIC layer, so its reporting-oriented graphics sit outside the population-adjustment workflow.
Why it is open
Published analyses built on archived packages cannot be reinstalled from CRAN and drift out of compatibility as dependencies move, so reproducing them needs a preserved environment most submissions never captured; the record degrades quietly rather than becoming impossible. Migration to a maintained successor is not a drop-in substitution because defaults, diagnostics and outputs differ, and no migration mapping or behavioral comparison has been published. BUGSnet’s weaker signals, meaning no CRAN path and issues open since 2019 despite a 2024 release, make adoption for new work a bet on maintainer capacity. Its strength is exactly the reporting-oriented output the field says it needs, so decay there removes the capability that is scarcest.
What has been tried
| Work | What it contributes |
|---|---|
| heorltd/maic | early MAIC workflow package; CRAN archived it on 22 July 2025 because email to the maintainer was undeliverable, and the archive holds only 0.1.2 to 0.1.4 with the last dated 27 April 2022 |
| Roche/MAIC | Roche MAIC package whose README carries the deprecation notice and the recommendation to use maicplus instead |
| BUGSnet (GitHub) | accessible, feature-rich Bayesian NMA workflow with reporting-oriented graphics and survival examples; never on CRAN, so installation depends on GitHub plus JAGS, and it has no native PAIC layer |
| BUGSnet releases (GitHub) | five tagged releases, v1.0.3 (2019) through v1.1.2 on 22 June 2024, contradicting the entry’s claim that the repository has no releases |
Probable solution or research direction
Publish a migration path and behavioral mapping from the archived and deprecated MAIC packages to maicplus, so historical analyses can be re-expressed rather than only reinstalled, and record the environment (dependency locks or containers) needed to re-run the originals rather than declaring them lost. For BUGSnet, either restore a maintained, versioned release path so its reporting outputs stay reproducible or port those graphics into a maintained package. Check maintenance and installation status before adopting any of these tools for new work, judging by release tags and issue age rather than by a binary maintained-or-not label.
Verification
Verdict. Overstated. 2 independent auditors judged the claim as stated too strong. The BUGSnet GitHub repository shows many long-open issues, several dating from 2019, and no release since v1.1.2 on 22 June 2024; it has never been on CRAN, so installation still depends on GitHub plus JAGS rather than a maintained CRAN path.
The source titled this. “Archived and unmaintained ITC packages break reproducibility” The title says archived packages break reproducibility, which the audit rejected since preserved source, dependency locks and containers keep historical analyses reproducible.
The source said. “Legacy MAIC tooling is no longer viable: the maic package was archived on CRAN in July 2025 and Roche states its MAIC package is no longer developed, recommending maicplus instead, so those installs belong in reproducibility archaeology rather than in a new reference workflow. BUGSnet, which is feature-rich and reporting-oriented, shows weaker maintenance signals, with long-open issues, no releases and a GitHub plus JAGS installation path rather than a maintained CRAN path.”
What was wrong with it. Both auditors rejected the claim that BUGSnet has no releases: the repository carries five tagged releases through v1.1.2 on 22 June 2024, and GPT-5.6 Sol confirmed Version: 1.1.2 in the source at audrey-b/BUGSnet@46102ef. That matters because the whole maintenance-decay argument rests on those signals; GPT-5.6 Sol also rejected the stronger reproducibility claim, since archived software remains reproducible with preserved environments, and the claim was unattributed in the source and is now sourced to the CRAN archive notice and the Roche README.
- Literature and prior-art check
- open / overstated. Three of four claim clusters verify exactly. CRAN states maic was ‘Archived on 2025-07-22 as email to the maintainer is undeliverable’, so July 2025 is right, and the archive holds only 0.1.2-0.1.4 (last 2022-04-27). The Roche/MAIC README carries the deprecation and the maicplus recommendation verbatim. BUGSnet is genuinely absent from CRAN (404), so the GitHub plus JAGS install path holds, and long-open issues are real: over 30 open issues, many filed in 2019. But ‘no releases’ is false. The repository has five tagged releases, the most recent BUGSnet v1.1.2 on 2024-06-22, which is a weaker maintenance signal than active packages but not the absence of releases. Since the whole BUGSnet-decay argument rests on maintenance signals, getting one of the two signals wrong matters.
- GPT-5.6 Sol, technical and source-code lens
- open / overstated. C1 is an empty placeholder. Archival and explicit deprecation are sound reasons not to choose the legacy MAIC packages for a new reference workflow, and a migration mapping remains useful. They increase reproducibility fragility but do not make historical analyses inherently unverifiable when source archives, dependency locks, or containers exist. BUGSnet upstream uses a GitHub and JAGS installation path, includes extensive NMA graphics and survival material, and has no native PAIC layer. The categorical claim of no releases is not supported by its source history: it has five version tags and a v1.1.2 DESCRIPTION. Offline evidence cannot determine whether separate GitHub Release objects exist or establish comparative maintainer capacity.
- Decision path
-
R5:default-open → R8:overstated-majority(n=2)
Source
Derived from A:L11, A:L110, A:L111, A:L129, C:L2255 of the reviewed corpus.