{
  "format": "agent-article-v0",
  "article_id": "PM-GPT-CORE-02",
  "canonical_url": "/providers/pm-gpt-core-02",
  "page_type": "reference",
  "claim_ids": [],
  "evidence_ids": [
    "PM-GPT-CORE-02-DOCUMENT-READBACK"
  ],
  "unknown_ids": [
    "PM-GPT-CORE-02-CURRENTNESS"
  ],
  "operational_object_ids": [],
  "provider": [
    "GPT"
  ],
  "models": [],
  "plans": [],
  "surfaces": [],
  "regions": [],
  "current_status": "UNKNOWN",
  "observation_status": "UNOBSERVED",
  "freshness": "UNKNOWN",
  "last_verified": null,
  "supersedes": [],
  "superseded_by": [],
  "provenance": [
    "DOC-083cde4caa627644-PM-GPT-CORE-02"
  ],
  "article_sources": [
    {
      "id": "PM-GPT-CORE-02-DOCUMENT-READBACK",
      "evidence_class": "IMPL",
      "title": "原資料との収録照合",
      "supports": [
        "資料単位の本文ハッシュ、段落数、表数を照合しています。"
      ],
      "limit": "収録照合は主張の真偽や外部根拠の最新性を追加保証しません。",
      "origin_class": "DOCUMENT_INGESTION_READBACK",
      "source_or_experiment_ids": [
        "DOC-083cde4caa627644-PM-GPT-CORE-02"
      ],
      "supporting_locators": [
        "18WPldGemks2px0ukhKYINeBMrBwAmlAi",
        "083cde4caa627644036a2225f1f6ff29b8015bd902614fe1f7a00a997b4eb59c",
        "provider_mappings/PM-GPT-CORE-02.md",
        "23d55fa54330857d0dd6333dc3538280b5c7d206ed12dc1038b3c33a2e03b5ae"
      ],
      "scope": "公開指定を受けた資料単位の全文・表・出典。",
      "provenance_status": "SOURCE_IDENTITY_VERIFIED"
    }
  ],
  "research_sections": [
    {
      "heading": "資料の識別情報",
      "text": "Source SHA-256 2da9ea12448865b6a21254f2ced4cb4852d191b3b5a8413c6fe856b9e5d73791"
    },
    {
      "heading": "Product Mapping — GPT / CORE-02",
      "text": "PRODUCT_MAPPING_ID: PM-GPT-CORE-02-v0.1\n\nPROVIDER: GPT / OpenAI\n\nCORE_PACKET: PKT-GEMINI-A-CORE-02-v0.1\n\nAS_OF: 2026-09-08 14:02 JST\n\nSTATUS: PUBLIC_MAPPING_AUDITED / LIVE_PROVIDER_AND_TENANT_PROFILE_PENDING\n\nLAYER: PRODUCT_MAPPING (not Enterprise AI Core; not Organization Overlay)"
    },
    {
      "heading": "Scope",
      "text": "この文書は、CORE-02のQ-C02-001..004をGPT/OpenAIの現行surfaceへ対応づける。対象はChatGPTのCompany Knowledge／apps with sync／Projects／Memoryと、OpenAI APIのfile search／Retrieval／data controlsである。公開仕様、既存のprovider-neutral研究・local実験、本研究の分析を分離し、公開仕様だけでend-to-end適合を認定しない。"
    },
    {
      "heading": "Surface separation",
      "text": "| SURFACE | RETRIEVAL / SOURCE | MEMORY / STATE | AUTHORITY BOUNDARY |\n| --- | --- | --- | --- |\n| ChatGPT Company Knowledge | enabled app/custom app、administrator-managed indexed source、source links | chat/workspace設定とは別 | connected source permission、app authorization、workspace RBACが適用されるとの公開仕様。内部retriever→reranker→cache→log→prompt全経路は非公開 |\n| ChatGPT administrator-managed sync | approved sourceを事前index | sync stateはsource memoryではない | admin-selected scope、provider permission、workspace restriction。新規content・permission change反映に時間を要し得る |\n| ChatGPT live app | provider accountでlive access | app固有 | indexed syncとは別surface。接続だけで個人sync indexを作るとは限らない |\n| ChatGPT Projects | project chats/filesをcontext化 | default／project-only memory | shared projectはproject-only。変更反映に数時間を要し得る。project deletionはfiles/chats/instructionsを不可逆に削除すると案内 |\n| ChatGPT improved／legacy saved memory | N/A | continually updated synthesis、またはsaved-memory notepad | memory summaryは全記憶を必ずしも表示しない。chat削除とsaved memory削除は別 |\n| OpenAI API file search / Retrieval | vector store semantic search、file attributesによるfilter、任意でraw search resultsをresponseへ含める | vector store/file/application state | attribute filterはnative source IAMではない。呼出applicationがprincipal/source authorizationを設計・維持する |\n| GPT Actions retrieval | external API／DB／vector DB | integration側 | service accountやchunk storeで元sourceのuser permissionを失う危険を公式guide自身が警告 |"
    },
    {
      "heading": "Core → GPT mapping rows",
      "text": "1. ROW: PM-C02-GPT-001; CORE QUESTION / FINDING: Q-C02-001 permission-aware retrieval; CURRENT PUBLIC EVIDENCE: Company Knowledgeはconnected source permission、app authorization、workspace RBACを維持すると説明。administrator-managed syncはpermission change反映に時間を要し得る; DECISION: VERIFIED_SPEC / END_TO_END_UNKNOWN。既存権限尊重の仕様意図は確認したが、pre-content authorization、revocation latency、cache/log/citation、multi-hop全経路PASSには昇格しない; REQUIRED SAFE EVIDENCE: 非機密2-principal corpus、allow/deny/revoke、sync前後、authz outage、multi-hop、source link、raw trace可視範囲をtenantで測る\n\n2. ROW: PM-C02-GPT-002; CORE QUESTION / FINDING: AUD-C02-008 / API retrieval path; CURRENT PUBLIC EVIDENCE: API file searchはfile metadata filterをsearch前に適用可能。raw search resultsは既定で返らず、include=[file_search_call.results]が必要; DECISION: APPLICATION_AUTHORIZATION_REQUIRED。metadata/attribute filterの存在をnative IAMまたは全経路permission保証にしない; REQUIRED SAFE EVIDENCE: principal-bound attributes、deny sentinel、top-k超過、filter欠落、vector-store誤指定、result include、citation/log、revocation再indexを固定版でfault injection\n\n3. ROW: PM-C02-GPT-003; CORE QUESTION / FINDING: Q-C02-002 memory lifecycle; CURRENT PUBLIC EVIDENCE: improved memory、legacy saved memory、chat history、project memoryは別surface。memory summaryは完全一覧とは限らず、chat削除後もsaved memoryが残り得る。完全削除にはsource群の個別削除が必要; DECISION: PUBLIC_LIFECYCLE_BOUNDARY / PURGE_AND_INTERNALS_UNKNOWN。UI文言やsummaryだけでphysical purge、backup/cache、cross-surface residualを閉じない; REQUIRED SAFE EVIDENCE: synthetic A→B、summary inspection、past-chat residual、saved-memory delete、chat/file/app source delete、Temporary Chat、新規session、project-only境界をcomponent別readback\n\n4. ROW: PM-C02-GPT-004; CORE QUESTION / FINDING: Q-C02-002 API retention/delete; CURRENT PUBLIC EVIDENCE: API data controlsはResponses、Conversations、vector stores、filesでapplication-state retentionを分ける。vector stores/filesはuntil deleted、deleted Assistants関連objectは30日後削除との記載を持つ; DECISION: VERIFIED_SPEC / TENANT_CONTRACT_REQUIRED。product retention tableをOrganization値へ転記せず、ZDR/MAM/project設定、third-party app retentionを別確認; REQUIRED SAFE EVIDENCE: org/project設定、endpoint、store、expires_after、delete receipt、post-delete lookup、backup/third-party residualを記録\n\n5. ROW: PM-C02-GPT-005; CORE QUESTION / FINDING: Q-C02-004 bounded long-context counterevidence; CURRENT PUBLIC EVIDENCE: CORE-02のNoLiMa／BABILong／LongBench v2 taxonomyをそのまま適用; DECISION: TEST_INTENT_MAPPED / NO_PROVIDER_RANK。公称context、単発成功、総合平均でGPT適合を閉じない; REQUIRED SAFE EVIDENCE: model ID、surface、prompt/config、position、latent association、multi-hop、aggregation、realistic taskを別sliceで反復\n\n6. ROW: PM-C02-GPT-006; CORE QUESTION / FINDING: safe live profile; CURRENT PUBLIC EVIDENCE: 今回は公開仕様readbackのみ。ChatGPT tenant、Company Knowledge、connected apps、OpenAI API、memory設定は操作していない; DECISION: NOT_EXECUTED / NOT_A_FAILURE。公式仕様だけでclosureせず、禁止・未承認経路を要求しない; REQUIRED SAFE EVIDENCE: owner承認済みsandbox、non-secret sentinel、setup/cleanup、timestamps、source permission、raw artifacts、hash、negative cases、final-state readback"
    },
    {
      "heading": "Counterevidence and failure boundaries",
      "text": "- administrator-managed syncの公開説明は、新規content・permission changeがrefresh後に反映されるまで時間を要し得ると明記する。したがってpermission-respecting intentとzero-stalenessは同義ではない。\n\n- Company Knowledgeのsource linkは検証手掛かりであり、citation completeness／faithfulnessの保証ではない。\n\n- OpenAIのGPT Actions retrieval guideは、vector DBのtext chunkが元文書のuser permissionを維持しにくく、GPT利用者全員がchunkへ到達し得る設計リスクを明示する。service-account／shared-store設計をpermission-awareと自動認定しない。\n\n- memory summaryは高水準synthesisであり完全inventoryではない。visible residual=0と全component purgeを同一視しない。\n\n- API raw file-search resultsが既定で省略されるため、trace非取得時にretrieval miss／generation missを断定しない。"
    },
    {
      "heading": "Organization Overlay handoff",
      "text": "以下はProduct Mappingで埋めない。\n\n- workspace plan、role、enabled plugin/app、RBAC、admin-managed source scope\n\n- provider-side source permissions、sync timestamp、revocation SLA、cache/log/citation exposure\n\n- Memory／Projects／retention／ZDR／MAMの実設定\n\n- Company Knowledge、connected apps、API利用の許可・禁止\n\n- live-test owner、synthetic tenant、cleanup、raw-log公開可否\n\n既存ORG-CORE-02.mdのGemini観測値はGPTへ転用しない。GPT組織値はすべてUNKNOWN / NOT_TESTEDのまま保持する。"
    },
    {
      "heading": "Provenance",
      "text": "- EXT-C02-052: OpenAI, Administrator-managed apps with sync in ChatGPT\n\n- EXT-C02-053: OpenAI, Company knowledge in ChatGPT\n\n- EXT-C02-054: OpenAI, Apps in ChatGPT\n\n- EXT-C02-055: OpenAI, File search\n\n- EXT-C02-056: OpenAI, Retrieval\n\n- EXT-C02-057: OpenAI, Data retrieval with GPT Actions\n\n- EXT-C02-058: OpenAI, Memory FAQ\n\n- EXT-C02-059: OpenAI, Projects in ChatGPT\n\n- EXT-C02-060: OpenAI, Data controls in the OpenAI platform\n\n- EXT-C02-044..046: NoLiMa／BABILong／LongBench v2 originals\n\n- EXP-C02-LOCAL-001/002: 本研究のprovider-neutral local実験。GPT実測へ再分類しない"
    },
    {
      "heading": "Exit condition",
      "text": "Q-C02-001..004のGPT public-surface mappingと安全な実測profileは作成した。公開仕様だけでprovider適合を閉じず、Company Knowledge全経路、tenant ACL/revocation、memory residual、API physical purge、benchmark実性能をUNKNOWN / NOT_TESTEDとしてOrganization Overlayまたは承認済みlive testへ移送する。Core claimは変更していない。"
    }
  ],
  "human_url": "/providers/pm-gpt-core-02",
  "markdown_url": "/agent/article.md?path=%2Fproviders%2Fpm-gpt-core-02",
  "json_url": "/agent/article.json?path=%2Fproviders%2Fpm-gpt-core-02"
}