Function effects_assertion_diagnostics 

pub fn effects_assertion_diagnostics(
    file: FileId,
    hir: &HirFile,
    index: &SymbolIndex,
    scope: &ImportScope,
    rows: &BTreeMap<DefinitionId, EffectRow>,
) -> Vec<Diagnostic>

Check every knot/stitch’s #@effects(…) assertion in hir against rows — that def’s inferred EffectRow, however the caller computed it: the whole-project pure crate::effects_project for the analyzer’s monolithic path, or, for the salsa-memoized production path, a small map built from individual per-def effects(def) queries (only for the defs that actually carry an assertion, preserving the advisory/lazy invariant — an unannotated project never triggers effect inference at all).

A def whose own id can’t be resolved, or whose row is missing from rows, produces no diagnostic here — both are the caller’s contract to uphold (every assertion-carrying def gets an entry), not a case this function can distinguish from “not computed yet”.

scope is hir’s own ImportScope (issue #881, the T2 follow-up to M-2d/#790): a reads/writes/calls clause name is resolved through the exact same import-scoped [lookup_by_name] the reference resolver uses, so a #@effects assertion in a file that imports one of several same-name cross-module cells binds to that importer’s cell — never a flat first-inserted winner that could silently name a different module’s definition than the one the body’s own inferred row actually touches.