{ "format": "agent-article-v0", "article_id": "EA-092", "canonical_url": "/tools/ea-092-ai-disaster-recovery-bcp", "page_type": "reference", "claim_ids": [], "evidence_ids": [ "EVD-EA-092-URL-5054E58070D9", "EVD-EA-092-URL-06C934AF218E", "EA-092-R11-RESEARCH", "EA-092-R11-WEB-READBACK" ], "unknown_ids": [ "EA-092-UNKNOWN-01", "EA-092-UNKNOWN-CURRENTNESS" ], "operational_object_ids": [], "provider": [], "models": [], "plans": [], "surfaces": [], "regions": [], "current_status": "UNKNOWN", "observation_status": "SPEC_ONLY", "freshness": "UNKNOWN", "last_verified": null, "supersedes": [], "superseded_by": [], "provenance": [ "COMM-082-EA0092", "SRC-EA-092-URL-5054E58070D9", "SRC-EA-092-URL-06C934AF218E", "WEBBODY-EA0092-R8-339045E0" ], "article_sources": [ { "id": "EVD-EA-092-URL-5054E58070D9", "evidence_class": "RESEARCH", "title": "1. [official spec / PRODUCT_VOLATILE] AWS Bedrock cross-Region inference docs, accessed 2026-09-16.", "href": "https://docs.aws.amazon.com/bedrock/latest/userguide/cross-region-inference.html", "supports": [], "limit": "元の監査済み資料に記載された参照先です。URLの復元は、資料の最新性や全主張の支持を新たに検証したことを意味しません。", "origin_class": "SOURCE_PACKET_REFERENCE", "source_or_experiment_ids": [ "SRC-EA-092-URL-5054E58070D9" ], "supporting_locators": [ "EVIDENCE TABLE", "1. [official spec / PRODUCT_VOLATILE] AWS Bedrock cross-Region inference docs, accessed 2026-09-16. https://docs.aws.amazon.com/bedrock/latest/userguide/cross-region-inference.html" ], "scope": "EVIDENCE TABLE", "provenance_status": "SOURCE_PACKET_REFERENCE_PROJECTED" }, { "id": "EVD-EA-092-URL-06C934AF218E", "evidence_class": "RESEARCH", "title": "2. [SRE practice] Google SRE Workbook, “Implementing SLOs”, Steven Thurgood et al., accessed 2026-09-16.", "href": "https://sre.google/workbook/implementing-slos/", "supports": [], "limit": "元の監査済み資料に記載された参照先です。URLの復元は、資料の最新性や全主張の支持を新たに検証したことを意味しません。", "origin_class": "SOURCE_PACKET_REFERENCE", "source_or_experiment_ids": [ "SRC-EA-092-URL-06C934AF218E" ], "supporting_locators": [ "EVIDENCE TABLE", "2. [SRE practice] Google SRE Workbook, “Implementing SLOs”, Steven Thurgood et al., accessed 2026-09-16. https://sre.google/workbook/implementing-slos/" ], "scope": "EVIDENCE TABLE", "provenance_status": "SOURCE_PACKET_REFERENCE_PROJECTED" }, { "id": "EA-092-R11-RESEARCH", "evidence_class": "RESEARCH", "title": "EA-092 監査済みresearch / canonical revision binding", "supports": [ "- AI BCPは dependency map をmodel APIだけでなくidentity, retrieval/index, vector DB, tool/API, policy/guardrail, queue, storage, telemetryまで展開する。\n- 復旧目標は RTO/RPO に加え `DQO (degraded-quality objective)` を置く提案が妥当。fallbackで業務継続しても品質・安全性が許容範囲外ならDR成功とはしない。\n- failover modeは full-equivalent / degraded-capability / read-only / human-only / queue-and-replay / hard-stop に分類する。" ], "limit": "canonical監査・revision bindingの存在を示す。外部ソースcurrentnessが本文またはcontrol-planeで確定していない場合は、currentnessをUNKNOWNのまま保持する。", "origin_class": "AUDITED_CANONICAL_RESEARCH_PACKET", "source_or_experiment_ids": [ "COMM-082-EA0092" ], "supporting_locators": [ "COMM-082-EA0092", "EA-092-R5-LEGACY-MIGRATION-20261008-R1", "content/staging/EA-092.md" ], "scope": "監査済みEA本文・canonical revision・既存Source Graph relationの範囲。欠損currentnessは推測で閉じない。", "provenance_status": "AUDIT_PASS_CANONICAL_BINDING" }, { "id": "EA-092-R11-WEB-READBACK", "evidence_class": "IMPL", "title": "EA-092 Web body identity / canonical readback", "supports": [ "canonical staging bodyのSHA-256一致とR9 canonical ledger readbackをProduction projectionの入力として検証する。" ], "limit": "body identity/readbackは本文claim自体の真偽や外部ソースの最新性を追加保証しない。", "checked_at": null, "origin_class": "IMPLEMENTATION_READBACK", "source_or_experiment_ids": [ "WEBBODY-EA0092-R8-339045E0" ], "supporting_locators": [ "WEBBODY-EA0092-R8-339045E0", "content/staging/EA-092.md", "EA-092-R5-LEGACY-MIGRATION-20261008-R1" ], "scope": "R9/R10 canonical readback後のProduction生成時body identity。", "provenance_status": "CANONICAL_READBACK_BODY_HASH_MATCH" } ], "research_sections": [ { "heading": "WHY_THIS_MATTERS_TO_UPDATE", "text": "AI依存業務ではprovider outageだけでなくmodel withdrawal、regional capacity、quota、identity、retrieval、tool、policy service停止が業務停止要因になる。DRは「別modelを呼ぶ」だけでは閉じない。" }, { "heading": "SCOPE_ROUTING", "text": "Enterprise AI Core / reliability / governance / interoperability / organization overlay" }, { "heading": "CLAIM_OR_RESEARCH_QUESTION", "text": "AI DRはRTO/RPOに加えてquality degradation・policy equivalence・state recoveryを定義すべきか。" }, { "heading": "EVIDENCE TABLE", "text": "1. [official spec / PRODUCT_VOLATILE] AWS Bedrock cross-Region inference docs, accessed 2026-09-16. https://docs.aws.amazon.com/bedrock/latest/userguide/cross-region-inference.html — multi-Region routingは実在するがProvisioned Throughput非対応等の制約あり。 relation=support/limit.\n2. [SRE practice] Google SRE Workbook, “Implementing SLOs”, Steven Thurgood et al., accessed 2026-09-16. https://sre.google/workbook/implementing-slos/ — critical dependency、error budget、graceful degradation、shared fateを明示。 relation=context/support." }, { "heading": "FINDINGS", "text": "- AI BCPは dependency map をmodel APIだけでなくidentity, retrieval/index, vector DB, tool/API, policy/guardrail, queue, storage, telemetryまで展開する。\n- 復旧目標は RTO/RPO に加え `DQO (degraded-quality objective)` を置く提案が妥当。fallbackで業務継続しても品質・安全性が許容範囲外ならDR成功とはしない。\n- failover modeは full-equivalent / degraded-capability / read-only / human-only / queue-and-replay / hard-stop に分類する。" }, { "heading": "LIMITS / COUNTEREVIDENCE / FAILURE ENVELOPE", "text": "複数Region/providerでもshared model vendor、identity、network、control plane、prompt/config registryが共通なら独立failure domainではない。fallback modelのtool/schema/refusal差によりsilent corruptionが起こり得る。" }, { "heading": "AVAILABILITY_GATES", "text": "fallback routeの事前契約・quota・warm capacity・region/data policy・secret・model access・dependency availability。" }, { "heading": "IMPLEMENTATION_STATE", "text": "将来Core候補 `AI_BCP_RECOVERY_CONTRACT`: business function, criticality, dependencies, failure domains, primary/fallback route, RTO/RPO/DQO, allowed degradation, state checkpoint, queue/replay rule, human fallback, failback criteria, exercise timestamp/result。" }, { "heading": "BENCHMARK_OR_EVAL", "text": "GameDay/chaos fixtureでprovider 5xx/429/timeout、region loss、model unavailable、retrieval down、tool down、identity downを注入。RTO、lost/duplicated work、quality delta、policy violation、manual loadを測る。" }, { "heading": "SECURITY / PRIVACY / AUDIT / COST / OPERABILITY", "text": "DR routeは平常時より緩いpolicyにしない。failover先のdata residency/retentionを事前承認。warm standby費用とDR drill費用を明示。" }, { "heading": "SAFE_WORDING_FOR_FUTURE_BOOK", "text": "「AIのDRはAPIの切替ではなく、依存関係・状態・品質・policyを含む業務継続設計である。別providerが存在するだけではBCPにならない。」" }, { "heading": "UPDATE_WORK_HANDOFF", "text": "EA-091 route contractをBCP routeへ流用し、EA-093 quota exhaustionもfailure injectionに含める。" }, { "heading": "OPEN_UNKNOWN / STOP_CONDITION", "text": "provider固有SLA/DR保証は公開直前再確認。実tenant GameDay未実施。設計contractとTEST_PROFILE確立につき停止条件充足。\n\n\n---\n\n---" }, { "heading": "AUDITED_CURRENTNESS_BOUNDARY", "text": "Cross-Region inference is one resilience/throughput mechanism, not complete BCP; Geographic/Global residency and Provisioned Throughput limitations stay explicit." } ], "human_url": "/tools/ea-092-ai-disaster-recovery-bcp", "markdown_url": "/agent/article.md?path=%2Ftools%2Fea-092-ai-disaster-recovery-bcp", "json_url": "/agent/article.json?path=%2Ftools%2Fea-092-ai-disaster-recovery-bcp" }