{
  "schema": "varexis-public-evidence-record/0.3",
  "generated_at_utc": "2026-10-23T02:13:06.885Z",
  "record_id": "VXI-EXT-PROCUREMENT-LANGGRAPH-PENDING-WRITES-PARALLEL-FAILURE-RECOVERY-01-v1",
  "record_kind": "external_reference",
  "title": "LangGraph: Pending writes preserve successful siblings after a step failure",
  "system": "LangGraph",
  "evidence_family": "vendor_evidence_review",
  "index_status": "PUBLIC_SAFE_SOURCE_REVIEW",
  "evidence_state": "PRIMARY_DOCUMENTATION_REVIEW_ONLY",
  "canonical_question": "Must LangGraph rerun every parallel node when one node in the same super-step fails?",
  "near_questions": [
    "Does LangGraph retain successful sibling outputs after a parallel node fails?",
    "Are LangGraph pending writes the same as full checkpoints?",
    "Which nodes repeat when LangGraph resumes a failed super-step?",
    "Can pending checkpoint writes prove an external API action happened once?"
  ],
  "supported_claims": [
    {
      "claim": "LangGraph documents pending checkpoint writes for nodes that finish successfully within a super-step. When another node in that step fails, the retained writes allow normal recovery to resume without rerunning those successful nodes. These per-task writes are not complete StateSnapshot checkpoints; time travel uses full super-step checkpoints. This recovery mechanism does not independently reconcile external side effects.",
      "reliance": "DOCUMENTED_CAPABILITY_ONLY"
    }
  ],
  "not_demonstrated_claims": [
    "Exactly-once external actions or recovery when the required writes were not persisted",
    "Equivalence between normal failure recovery and deliberately replaying an earlier checkpoint",
    "Buyer deployment testing, live runtime enforcement, certification or permission to act",
    "Expanded capabilities or licences for existing INDEX executables"
  ],
  "public_summary": "LangGraph documents pending checkpoint writes for nodes that finish successfully within a super-step. When another node in that step fails, the retained writes allow normal recovery to resume without rerunning those successful nodes. These per-task writes are not complete StateSnapshot checkpoints; time travel uses full super-step checkpoints. This recovery mechanism does not independently reconcile external side effects.",
  "related_record_ids": [],
  "source_review": {
    "source_type": "official_external_documentation",
    "source_title": "LangGraph checkpointers",
    "source_url": "https://docs.langchain.com/oss/python/langgraph/checkpointers",
    "checked_on": "2026-10-23",
    "candidate_record_id": "VXI-EXT-PROCUREMENT-LANGGRAPH-PENDING-WRITES-PARALLEL-FAILURE-RECOVERY-01-v1",
    "candidate_record_sha256": "c2648b376386d7d4c38254b5f6d411545d677b3be4d11b2a1d2a883fecb15acc",
    "method": "Primary documentation reviewed for the published bounded statement.",
    "deployment_tested_by_varexis": false,
    "revalidation": "The delivered snapshot preserves the review date shown. Recheck current primary documentation and deployment context before relying on present applicability; payment does not trigger fresh research."
  },
  "answer_policy": {
    "classification": "DOCUMENTED_WITH_DEPLOYMENT_BOUNDARY",
    "bounded_answer": "LangGraph documents pending checkpoint writes for nodes that finish successfully within a super-step. When another node in that step fails, the retained writes allow normal recovery to resume without rerunning those successful nodes. These per-task writes are not complete StateSnapshot checkpoints; time travel uses full super-step checkpoints. This recovery mechanism does not independently reconcile external side effects.",
    "required_buyer_context": "Graph topology, node identities, failed super-step and installed LangGraph version; Persisted pending writes, full checkpoint and intended resume versus replay operation; External effect identities and independent outcome reconciliation for affected nodes. Buyer deployment not inspected.",
    "missing_evidence": "A broader claim needs evidence matching the intended use. Payment does not upgrade the evidence state."
  },
  "private_boundary": "Raw logs, thresholds, private paths, source code, protected mechanisms and reconstruction-level details are excluded from public and paid delivery.",
  "decision_profile": {
    "schema": "varexis-vendor-evidence-review/1",
    "review_status": "documentation_review_only",
    "checked_on": "2026-10-23",
    "claim_to_check": "Must LangGraph rerun every parallel node when one node in the same super-step fails?",
    "documented_statement": {
      "status": "documented",
      "statement": "LangGraph documents pending checkpoint writes for nodes that finish successfully within a super-step. When another node in that step fails, the retained writes allow normal recovery to resume without rerunning those successful nodes. These per-task writes are not complete StateSnapshot checkpoints; time travel uses full super-step checkpoints. This recovery mechanism does not independently reconcile external side effects.",
      "source_ids": [
        "langgraph-checkpointers"
      ]
    },
    "review_inference": {
      "status": "review_inference",
      "statement": "A recovery review should distinguish successful persisted sibling outputs from the failed node and from an intentional historical replay.",
      "source_ids": [
        "langgraph-checkpointers"
      ]
    },
    "buyer_evidence_required": [
      "Graph topology, node identities, failed super-step and installed LangGraph version",
      "Persisted pending writes, full checkpoint and intended resume versus replay operation",
      "External effect identities and independent outcome reconciliation for affected nodes"
    ],
    "review_condition": {
      "status": "varexis_review_policy",
      "statement": "Refuse a claim that every sibling must rerun, or that no external effect can repeat, without inspecting saved writes and the chosen recovery operation."
    },
    "unknowns": "Persisted buyer writes, failed-node effects, saver behaviour and the selected recovery operation are unobserved.",
    "unknown_finding": {
      "status": "unknown",
      "statement": "Persisted buyer writes, failed-node effects, saver behaviour and the selected recovery operation are unobserved."
    },
    "evidence_sources": [
      {
        "source_id": "langgraph-checkpointers",
        "url": "https://docs.langchain.com/oss/python/langgraph/checkpointers",
        "final_url": "https://docs.langchain.com/oss/python/langgraph/checkpointers",
        "title": "LangGraph checkpointers",
        "checked_on": "2026-10-23",
        "captured_at_utc": "2026-10-23T02:08:12.254Z",
        "version_context": "Rolling LangGraph Python checkpointer documentation captured on the review date; installed package and saver implementation require buyer-specific verification.",
        "sha256": "327beef0afa420222d65488175f4c124b66318e120ab8203b6e6a226875714b7",
        "bytes": 1326496,
        "capture_status": 200,
        "hash_basis": "Decoded HTTP response body bytes; identifies captured content, not truth or certification",
        "reviewed_sections": [
          "Durability modes",
          "Pending writes",
          "Super-steps"
        ],
        "rights_review": "Factual paraphrase and source metadata only; response body retained privately and not redistributed."
      }
    ],
    "observed_deployment": false,
    "commercial_reliance_boundary": "Documentation review only; the buyer must establish applicability and actual outcomes. INDEX does not inspect or control the deployment.",
    "paid_value": {
      "decision_context": "Avoid unnecessary repeated actions while reconstructing which parts of a parallel agent step completed before failure.",
      "free": "Complete review, sources, practical questions and limitations; may be saved locally.",
      "single_request": "Existing server-saved version-bound route snapshot for later authenticated retrieval.",
      "evidence_receipt": "The same snapshot plus the existing unsigned delivery receipt.",
      "excluded": [
        "Fresh vendor research",
        "Continuous monitoring or change alerts",
        "Testing the buyer configuration",
        "Private evidence or vendor certification"
      ],
      "integrity": "Hashes identify content versions; they do not prove truth or provide a signature."
    },
    "revalidation": {
      "trigger": "A saved purchase does not recheck the source or buyer configuration.",
      "specific_watch": "Recheck the documented behavior, product version and buyer evidence before relying on a current outcome."
    },
    "buyer_relevance": "Avoid unnecessary repeated actions while reconstructing which parts of a parallel agent step completed before failure.",
    "refusal_conditions": [
      "Refuse a claim that every sibling must rerun, or that no external effect can repeat, without inspecting saved writes and the chosen recovery operation."
    ]
  },
  "niche": {
    "niche_id": "agent-authority-action-evidence",
    "label": "Agent Authority & Action Evidence",
    "classification_id": "audit",
    "classification_label": "Audit, Provenance & Evidence Accountability",
    "topic_id": "audit",
    "buyer_intent": "Avoid unnecessary repeated actions while reconstructing which parts of a parallel agent step completed before failure.",
    "comparison_role": "supplementary_discovery"
  },
  "paid_access_note": "The full published answer and boundaries are free. Paid access saves the selected route and source versions, with an unsigned receipt when purchased. Preview coverage before paying. Fresh research, bespoke assessment and protected evidence are excluded.",
  "service_route": {
    "request_guide": "/varexis-index/request.html",
    "request_workflow": "/varexis-index/request-workflow.json",
    "service_id": "single_request",
    "service_routes_url": "/varexis-index/service-routes.json",
    "payment_required": true,
    "availability_endpoint": "/api/index/capabilities",
    "operational_status": "AUTOMATIC_SERVICE_CAPABILITIES_GATED",
    "checkout_url": "/index-access"
  },
  "authority_boundary": "VAREXIS-INDEX informs evidence reliance. It does not provide advice, certification, deployment authority, security assurance, production approval, or guarantee.",
  "human_page": "/varexis-index/catalog.html#langgraph-pending-writes-parallel-failure-recovery",
  "machine_readable_record": "/varexis-index/records/langgraph-pending-writes-parallel-failure-recovery.json"
}
