Imagine a record that says, “Variant X affects the function of protein Y,” followed by an assessment that “supports” the claim. If the assessment travels without the claim, a reader can see the word “supports” but cannot tell what it supports. The variant-annotation model I reviewed represents such claims and assessments in a form other systems can read.
The Global Alliance for Genomics and Health (GA4GH) offers the Variant Annotation Specification (VA-Spec) as a model for representing those annotations. While reading its September 2026 pre-release ballot for version 1.1.0, I raised two questions in the public community review. Both concerned details that could lead people implementing the ballot to different interpretations.
What does “supports” refer to?
VA-Spec calls the claim being evaluated a proposition and the record that evaluates it a Statement. In the imaginary example, “Variant X affects the function of protein Y” is the proposition. A Statement‘s direction says whether the evidence supports, disputes, or is neutral toward that claim.
The 1.1 ballot merges the former evidence-line class, EvidenceLine, into Statement and makes proposition optional. I read the nested case as a possible reason: a statement used as an evidence line inside another statement may evaluate the same proposition and may not need to repeat it. But the ballot schema also appeared to allow a standalone Statement with type and direction but no proposition. Its kind and direction would be recorded, while the claim being evaluated would be missing.
I asked whether that omission was intended for standalone statements too. Maintainer Larry Babb replied that it was an omission: proposition should be conditionally required when a Statement is used standalone. His answer distinguishes the possible nested case from the requirement for an independent record.
A 1.1 explanation leading to 1.0.0 examples
Examples help a reader turn a specification into data. At the time of my question, the VA-Spec Examples page linked to tests/fixtures at the 1.0.0 tag. The 1.1 ballot had already updated its fixtures for the EvidenceLine to Statement change and other renamed fields. Someone reading the new explanation could follow the link and learn the previous data shape instead.
I suggested linking to examples from the matching documentation version, or avoiding a fixed link to 1.0.0. Larry called it a good catch and said it would be fixed. The public questions and replies record that initial response and plan.
What appeared in the follow-up changes
When I checked Larry Babb’s PR #449 on 28 September, it contained concrete changes related to both questions. It made Statement.proposition required, restored EvidenceLine as a separate class, and replaced the examples page’s fixed 1.0.0 link with a version-aware path. Larry authored those changes; my contribution was raising the questions in the public review and receiving his responses. The PR was still open when I checked. The proposed code and documentation had changed, but that was not yet a merged change or a finalized specification.
A separate contribution guide for the GKM Starter Kit
On 23 September, I also submitted PR #101 to the GA4GH GKM Starter Kit, a collection of examples and tools for using the Genomic Knowledge Model. A bundle in this toolkit groups linked records and their provenance for exchange. Even someone proposing a small example needs to explain the standards and versions it uses, where any real data came from, whether it may be shared, and how its structure and internal references can be checked. I brought that guidance into one contribution page and added it to the Data navigation.
The page separates a focused example or documentation fix from publishing a complete dataset. The latter does not yet have a self-service submission process, so the guide asks contributors to discuss the scope with maintainers first. It also distinguishes a successful documentation build from validation of a data bundle: a working documentation site does not establish that the data are valid or licensed for reuse. As of 28 September, this documentation PR was also open for review; it had not been merged or adopted as the project’s contribution policy.