[Feature]: Prove shared graph semantics across polyglot providers

[Baziar]

Problem statement

Workspai detects many runtimes and qualifies real repositories, but those tests
do not yet prove that every graph provider means the same thing by project,
module, service, API, dependency, proof, and unknown coverage.

Proposed solution

Current foundation

• versioned Model and Knowledge Graph contracts;
• source-backed providers for multiple runtime families;
• isolated and cumulative real-world qualification;
• aggregate-plus-nested qualification on real .NET/Python/Node repositories;
• bounded graph search, evidence, path, and project projections.

Remaining work

Create one small, redistributable fixture workspace with TypeScript, Python, Go,
Rust, Java/Kotlin, .NET, PHP, and C/C++. Add deliberate cross-language API,
schema, build, generated-code, and ambiguous relationships. Golden expectations
describe public graph facts—not provider AST internals.

Acceptance criteria

• Eight runtime families share one versioned fixture manifest.
• Expected entities, relations, proofs, and intentional unknowns are explicit.
• At least three cross-language relations are proven end to end.
• Duplicate logical entities, relation direction, confidence, and coverage are asserted.
• Every expected relation has portable source evidence.
• Repeated runs are semantically identical on Linux, macOS, and Windows.
• Provider-specific additions cannot weaken or rewrite the shared golden floor.
• One offline command emits a machine-readable per-provider report.

Completion evidence

Publish the fixture license inventory, conformance command, all three OS reports,
and a mutation test covering add/edit/delete/rename of source and shared schema.

Current starting points: packages/cli/scripts/real-world-qualification.mjs,
packages/cli/src/workspace-knowledge-graph.ts, and
packages/cli/contracts/workspace-intelligence/workspace-knowledge-graph.v1.json.

Alternatives considered

The following alternatives or adjacent concerns are intentionally outside this issue:

• natural-language ranking quality, tracked by the relevance-corpus issue;
• installing every real toolchain, tracked by the scheduled runtime-matrix issue;
• moving providers into a new package.

Expected impact

Provider work becomes comparable across languages, and every consumer receives
the same minimum entity, relation, proof, and unknown semantics.

[github-actions]

Hey @Baziar — welcome to Workspai! 👋

Thanks for taking the time to open your first issue. Clear reports and focused ideas help us make Workspai more useful for developers and AI tools working across real software workspaces.

A maintainer will review the scope and labels. If you would like to work on this issue yourself, leave a comment and we’ll help you find the relevant contracts, code, and tests.

Glad to have you here.