Gap
crates/gbrg-napi/src/lib.rs returns hardcoded ProofArtifact stub JSON — blast_radius yields "derivation":"stub: gbrg-napi not yet wired to a frozen index", generated:false, zeros; graph_status returns "skeleton". So the consumable (N-API) surface of GBRG is a skeleton even though gbrg-core is real.
Impact (downstream, blocked on this)
The fix (gbrg-core already exposes what's needed)
Wire the napi bodies to the real reads:
- resolve the IRI:
gbrg_core::cell_iri_to_node_id(&cell_id) -> NodeId
- query a frozen
GraphIndex: dependents_count, reverse_dependents, transitive_dependents, code_dependents_count, test_coverage_reach, reachable_set (containment), and build the result with build_proof_artifact / blast_radius_score.
- Missing piece: a process-held frozen
GraphIndex. build_index(cells, edges) exists (whatif.rs) but is private; expose a public freeze/build_index_from_source(root) (parse via gbrg-parser → SemanticCells+edges → build_index) and a napi freeze_index(root) that stores it in a global (OnceCell/Mutex), which blast_radius(cell_id) then queries.
Acceptance
blast_radius returns a real ProofArtifact (generated:true, real dependents_count/blast_radius) when an index is frozen, and an honest INCONCLUSIVE (not a fake 0) when none is.cargo build -p gbrg-napi green; a round-trip test (freeze a small source tree → blast_radius(known cell) → non-zero dependents).
A partial wiring PR (real IRI resolution + honest-INCONCLUSIVE-vs-real-query branch) is going up now; this issue tracks the frozen-index plumbing.
Gap
crates/gbrg-napi/src/lib.rsreturns hardcoded ProofArtifact stub JSON —blast_radiusyields"derivation":"stub: gbrg-napi not yet wired to a frozen index",generated:false, zeros;graph_statusreturns"skeleton". So the consumable (N-API) surface of GBRG is a skeleton even thoughgbrg-coreis real.Impact (downstream, blocked on this)
via:"regex-stopgap"edges for the same reason.The fix (gbrg-core already exposes what's needed)
Wire the napi bodies to the real reads:
gbrg_core::cell_iri_to_node_id(&cell_id) -> NodeIdGraphIndex:dependents_count,reverse_dependents,transitive_dependents,code_dependents_count,test_coverage_reach,reachable_set(containment), and build the result withbuild_proof_artifact/blast_radius_score.GraphIndex.build_index(cells, edges)exists (whatif.rs) but is private; expose a publicfreeze/build_index_from_source(root)(parse viagbrg-parser→ SemanticCells+edges → build_index) and a napifreeze_index(root)that stores it in a global (OnceCell/Mutex), whichblast_radius(cell_id)then queries.Acceptance
blast_radiusreturns a real ProofArtifact (generated:true, realdependents_count/blast_radius) when an index is frozen, and an honest INCONCLUSIVE (not a fake 0) when none is.cargo build -p gbrg-napigreen; a round-trip test (freeze a small source tree → blast_radius(known cell) → non-zero dependents).A partial wiring PR (real IRI resolution + honest-INCONCLUSIVE-vs-real-query branch) is going up now; this issue tracks the frozen-index plumbing.