v0.3.0 retrieval-tooling compatibility matrix

Status: maintainer-run upstream-fixture trials, 2026-07-19.

These trials exercise the released necs-validate wheel against exact public artifacts pinned to immutable upstream commits. They test interoperability; they do not claim adoption, affiliation, or endorsement by the upstream projects.

All trials installed the public v0.3.0 wheel in an isolated CPython 3.12 environment. Its verified SHA-256 is 0ac9bfd52427e574d7d91948cab0e8d25aef442176a23125a860320844d8a03b.

Upstream artifact	Default result	Strict result	Detailed provenance
sidhulyalkar/neural-search exact qrels/run fixtures	PASS; one retrieved-but-unjudged warning	--require-judged promotes the same finding to an expected error	report
arclabs561/rankops exact Rust qrels/run constants	PASS; no warnings	coverage + judged-only + strict ranks all PASS	report
izikeros/rankflow exact fixtures and API round trip	PASS; sparse-qrels warnings only	zero-based ranks PASS --strict-ranks; judged-only mode finds the known unjudged document	report

Across the three trials, no malformed-row, duplicate, query-coverage, or advisory-rank incompatibility was found. The observed strict failures were deliberate promotions of authentic sparse-qrels findings, not parser failures.

The separate maintainer-owned reference consumer proves that Madhvansh/Neural-E-Commerce-Search@v0.3.0 also runs from another repository’s default branch. It is explicitly not counted as independent adoption.