Context. stargraph's differentiator is deterministic, provenance-typed, replayable governance via CLIPS rules rather than LLM routers. MemGraphRAG's Conflict-Detection agent is an LLM scanning for conflicts — exactly the non-deterministic dependency stargraph exists to replace. Encoding ontology-constraint / logical-inconsistency checks as a signable Bosun rule pack delivers the consistency benefit while preserving auditability (a genuine improvement over the paper's design for the regulated/air-gapped market). Pairs with the conflict-driven-deltas issue (rules detect → deltas resolve).
Proposal. Author a kg_conflicts Bosun pack.
- Own deftemplate carrying triple slots (subject/predicate/object/confidence) + provenance.
- defrules asserting
bosun.violation(kind 'logical-inconsistency') when two facts share (subject, predicate) but differ on object.
- Reuse the
fathom-matches user-function pattern (safety_pii/rules.clp:34) for regex/granularity checks. Sign + version per manifest.
Where.
- New pack alongside
src/stargraph/bosun/safety_pii/ (rules.clp + manifest.yaml); scaffold via harbor:new-pack; register via register_packs() hook (plugin/hookspecs.py:78).
CRITICAL slot-shape mismatch (must define own template). The EXISTING stargraph.evidence deftemplate (safety_pii/rules.clp:22-25) carries slots (run_id, step, text) — NOT subject/predicate/object. kg_promotion asserts subject/predicate/object/source to that same template name (kg_promotion.py:150-169) via assert_with_provenance, so today those slots would not match safety_pii rule patterns. The conflict pack must declare its own evidence-fact deftemplate.
Acceptance criteria.
Caveats. Temporal rules over t_valid/t_invalid have no data yet; CLIPS over the full fact layer faces the O(n²) cost the paper flags.
Source: arXiv:2606.00610v1 — "MemGraphRAG: Memory-based Multi-Agent System for Graph Retrieval-Augmented Generation". Distilled from arXiv-research/2606.00610v1/analysis.md; file refs verified against current main by the analysis pass.
Context. stargraph's differentiator is deterministic, provenance-typed, replayable governance via CLIPS rules rather than LLM routers. MemGraphRAG's Conflict-Detection agent is an LLM scanning for conflicts — exactly the non-deterministic dependency stargraph exists to replace. Encoding ontology-constraint / logical-inconsistency checks as a signable Bosun rule pack delivers the consistency benefit while preserving auditability (a genuine improvement over the paper's design for the regulated/air-gapped market). Pairs with the conflict-driven-deltas issue (rules detect → deltas resolve).
Proposal. Author a
kg_conflictsBosun pack.bosun.violation(kind 'logical-inconsistency')when two facts share (subject, predicate) but differ on object.fathom-matchesuser-function pattern (safety_pii/rules.clp:34) for regex/granularity checks. Sign + version per manifest.Where.
src/stargraph/bosun/safety_pii/(rules.clp + manifest.yaml); scaffold viaharbor:new-pack; register viaregister_packs()hook (plugin/hookspecs.py:78).CRITICAL slot-shape mismatch (must define own template). The EXISTING
stargraph.evidencedeftemplate (safety_pii/rules.clp:22-25) carries slots(run_id, step, text)— NOT subject/predicate/object.kg_promotionasserts subject/predicate/object/source to that same template name (kg_promotion.py:150-169) viaassert_with_provenance, so today those slots would not match safety_pii rule patterns. The conflict pack must declare its own evidence-fact deftemplate.Acceptance criteria.
bosun.violationon same-(subject,predicate) divergent-object facts.Caveats. Temporal rules over
t_valid/t_invalidhave no data yet; CLIPS over the full fact layer faces the O(n²) cost the paper flags.Source: arXiv:2606.00610v1 — "MemGraphRAG: Memory-based Multi-Agent System for Graph Retrieval-Augmented Generation". Distilled from
arXiv-research/2606.00610v1/analysis.md; file refs verified against currentmainby the analysis pass.