Genomic colocation: declare Short Variant as input, search, and category node

[jbrestel]

What this is

The model half of adding Short Variant to the Genomic Colocation combine step.

Depends on VEuPathDB/ApiCommonWebService#22 — merge that one first. This PR declares a search whose plugin behaviour only exists there: without it, VariantsBySpanLogic resolves to a plugin that throws for VariantRecordClass. The two branches share the name feat/variant-colocation.

That companion PR is also substantially larger than its title suggests: it repairs five Oracle-isms that made genomic colocation entirely non-functional on PostgreSQL — not just for the new record type. Read it first for context.

Changes

spanParams.xml — adds VariantRecordClass to both span_a and span_b, the answerParams enumerating which record classes may be colocation inputs.

This is two lines and it is the whole feature in one direction. Without it a Short Variant result cannot be offered as an input at all, so variants → genes is impossible — the search would appear to work one way and be silently unavailable the other. An earlier draft of this design omitted it entirely.

spanQuestions.xml — adds VariantsBySpanLogic, replacing the commented-out SnpsBySpanLogic and SnpsChipsBySpanLogic blocks (plus an orphaned banner comment). No UI change is needed: ApiBinaryOperations.tsx matches colocation searches by convention, searchName.endsWith('BySpanLogic').

Four deliberate choices, flagged so they don't read as oversights:

• includeProjects is copied from the VariantRecordClasses recordClassSet, so the search exists exactly where the record does.
• sorting mirrors the record's own default (chromosome_order_num asc, location asc). Not variant_location asc — that is a textAttribute and not a safe sort key.
• The three dynamic columns are named in attributesList as well as declared in dynamicAttributes, or they do not render.
• No CDS or gene dynamic attributes. The old SNP question carried linkedGeneId and position_in_protein; both dropped on purpose.

individuals.txt — replaces the two SNP BySpanLogic ontology entries with one for VariantsBySpanLogic. Needed because wb model does not regenerate the category OWL; without it the search exists but is uncategorised, with no error anywhere. Note the existing quirk, mirrored rather than fixed: every BySpanLogic entry sits under the DynSpanRecordClass parent regardless of its own record type.

IsolatesBySpanLogic and OrfsBySpanLogic stay commented out.

docs/superpowers/specs/2026-08-15-variant-genomic-colocation-design.md — the design record. Its §6 is written for reviewers and covers the full Oracle-ism table plus the one decision in the companion PR that most deserves pushback: which schema owns the span temp tables. getDefaultSchema() returns the login schema on Oracle and a hardcoded public on PostgreSQL, while an unqualified CREATE TABLE follows search_path — so create and drop diverged, and a correct colocation surfaced as an error while leaking a table per run. The chosen fix and two rejected alternatives are documented there.

The doc also corrects its own earlier claim that the damage was two lines, and supersedes a prior draft that proposed a model-declared span-source contract (rejected: unvalidated one-row-per-record promise that Transcript and DynSpan could not use anyway).

Build note

Requires wb full — Java changed in ApiCommonWebService and individuals.txt changed here, so neither wb model nor wb ontology alone is sufficient.

Test plan

Verified live on a PlasmoDB dev instance against genomicsdb_071n:

• VariantsBySpanLogic registered on record type variation (/service/record-types/variation)
• Present in the category tree as a sibling of GenesBySpanLogic and DynSpansBySpanLogic
• Gene → Variant
• Variant → Gene (the direction the span_a/span_b change enables)
• Variant → Genomic Span
• Gene → Gene unaffected, including a 10001 bp upstream offset
• Strand selector on a variant comparison returns results rather than zeroing out
• UniDB not tested — RecordsBySpanLogic declares wsColumn project_id unconditionally while VariantRecordClass excludes it from the PK there. Pre-existing and shared with DynSpanRecordClass; the shared query is deliberately untouched.

🤖 Generated with Claude Code

[steve-fischer-200]

see comment about include/exclude

also the superpowers files seem like a lot of noise