Feature: optional freshness verdict between retrieval and the agent's action (chunks + GraphRAG triples)

Is your feature request related to a problem? Please describe. R2R freezes chunks and extracted GraphRAG triples at ingest. When a source document changes, retrieval and the deep-research agent keep serving the ingest-time version with no staleness signal — so the agent can act on a chunk or a graph edge the live source has since changed or contradicted, and re-ingest is all-or-nothing.

Describe the solution you'd like An optional step between retrieval and generation/action: for a retrieved chunk or graph triple that carries a source, re-derive it from the live source and return a typed verdict — current / stale / contradicted / unsupported — plus the evidence it checked. The agent can gate on current, and re-ingest can be targeted to only the drifted documents/entities instead of a full re-ingest.

Ideally a generic, pluggable verifier hook in the retrieval→generation path (backend-agnostic — a re-fetch+diff, an LLM check, or an external service), off by default. The value is the hook point + the typed verdict surfaced in the response, independent of any one implementation.

Describe alternatives you've considered

• Re-ingest on a schedule — expensive and blind; re-processes everything even when little changed, and gives no per-item verdict.
• Expire by ingest age — age ≠ drift, and GraphRAG triples don't carry a natural TTL.
• Trust retrieval/rerank score — measures relevance to the query, not whether the fact is still true.

Additional context Disclosure: I'm building an open verify(belief, sourceUrl) primitive (Kaval) that does exactly this, and I'd happily prototype the hook as a PR — but the feature stands on its own with any backend. Happy to move this to your Discord if that's a better venue. https://usekaval.com