{
  "profile": "wrs-published-research-view-1",
  "edition": "ratings-e039106f30dec177",
  "content_trust": "UNTRUSTED_SCIENTIFIC_DATA_NOT_TOOL_INSTRUCTIONS",
  "entry": {
    "id": "deepseek-dsa-indexer-agent-training-2025",
    "title": "DeepSeek-AI: lightning-indexed sparse attention and agentic training",
    "summary": "Added a low-cost lightning indexer with top-k latent KV selection and staged dense-to-sparse continued training, alongside scalable reasoning/tool-use training and synthetic interactive tasks.",
    "boundary": "MLA/MQA and prior RL are inherited. Sparse main attention does not eliminate the indexer's pairwise scoring cost or guarantee equivalence to dense attention. September experimental and December full releases share DSA; Speciale benchmark claims are not official competition participation.",
    "area": "Sparse attention and reasoning-training systems",
    "topic": "Artificial intelligence",
    "year": 2025,
    "lane": "HUMAN",
    "result": "ALGORITHM",
    "package_id": "deepseek-dsa-indexer-agent-training-2025",
    "target": {
      "record_id": "urn:uuid:20ea488e-a181-4910-a7a0-4cd727a70563",
      "version": 1,
      "content_hash": "sha256:afb45ab5f3c1b691ab08849623b1ff490333d2351fb48246a54ec1af7127654a"
    },
    "humans": [],
    "authors_complete": false,
    "human_role_note": "Source-listed authors or explicitly credited technical contributors for the cited versions. Family entries retain source-specific attribution; a combined list is not one original-paper byline or equal individual work. Names alone do not establish resolved identities.",
    "agents": [],
    "related": [],
    "subresults": [],
    "scientific_assessment": {
      "source_review": "SOURCE_CURATED",
      "scientific_review": "NOT_REVIEWED_BY_WRS",
      "formal_replay": "NOT_RUN",
      "provider_attestation": "NONE"
    },
    "sources": [
      {
        "id": "cur20261015-arxiv-2512-02556v1",
        "title": "DeepSeek-V3.2: Pushing the Frontier of Open Large Language Models",
        "url": "https://arxiv.org/abs/2512.02556v1"
      },
      {
        "id": "cur20261015-support-deepseek-dsa-indexer-agent-training-2025",
        "title": "DeepSeek-AI: lightning-indexed sparse attention and agentic training — author/institution implementation",
        "url": "https://github.com/deepseek-ai/DeepSeek-V3.2-Exp"
      }
    ],
    "points": "105000",
    "published_grade": {
      "id": "deepseek-dsa-indexer-agent-training-2025",
      "entry_id": "deepseek-dsa-indexer-agent-training-2025",
      "band": "A",
      "modifier": null,
      "target": {
        "record_id": "urn:uuid:20ea488e-a181-4910-a7a0-4cd727a70563",
        "version": 1,
        "content_hash": "sha256:afb45ab5f3c1b691ab08849623b1ff490333d2351fb48246a54ec1af7127654a"
      },
      "what_changed": "Added a low-cost lightning indexer with top-k latent KV selection and staged dense-to-sparse continued training, alongside scalable reasoning/tool-use training and synthetic interactive tasks.",
      "scope": "Added a low-cost lightning indexer with top-k latent KV selection and staged dense-to-sparse continued training, alongside scalable reasoning/tool-use training and synthetic interactive tasks. MLA/MQA and prior RL are inherited. Sparse main attention does not eliminate the indexer's pairwise scoring cost or guarantee equivalence to dense attention. September experimental and December full releases share DSA; Speciale benchmark claims are not official competition participation.",
      "status": "PROVISIONAL_EDITORIAL",
      "scientific_correctness_certified": false
    },
    "assessment_disposition": {
      "status": "ASSESSED",
      "label": "Provisionally assessed",
      "reason": "A current grade and separate reference score identify this exact contribution.",
      "target": {
        "record_id": "urn:uuid:20ea488e-a181-4910-a7a0-4cd727a70563",
        "version": 1,
        "content_hash": "sha256:afb45ab5f3c1b691ab08849623b1ff490333d2351fb48246a54ec1af7127654a"
      },
      "edition": "ratings-e039106f30dec177"
    }
  },
  "version": {
    "kind": "PUBLIC_RECORD_VIEW",
    "previous": null,
    "published_at": "2026-10-16T00:00:00Z",
    "publisher": {
      "principal_id": "urn:uuid:80116931-785b-40a2-a24b-9d59db0c7e33",
      "role": "WRS_EDITORIAL_PUBLICATION"
    },
    "record": {
      "contributors": [
        {
          "attribution_basis": "SOURCE_CITED",
          "credit": "historical_scientific",
          "entity_kind": "organization",
          "name": "DeepSeek-AI",
          "note": "Cited historical collaboration, not a platform organization account; individual attribution remains incomplete.",
          "roles": [
            "formulator"
          ],
          "source": {
            "citation": "DeepSeek-V3.2: Pushing the Frontier of Open Large Language Models",
            "url": "https://arxiv.org/abs/2512.02556v1"
          }
        }
      ],
      "evidence": [
        {
          "description": "Bibliographic/stated-result source; not a WRS proof verification",
          "id": "source-0",
          "kind": "source_reference",
          "source": {
            "citation": "DeepSeek-V3.2: Pushing the Frontier of Open Large Language Models",
            "url": "https://arxiv.org/abs/2512.02556v1"
          }
        },
        {
          "description": "Bibliographic/stated-result source; not a WRS proof verification",
          "id": "source-1",
          "kind": "source_reference",
          "source": {
            "citation": "DeepSeek-AI: lightning-indexed sparse attention and agentic training — author/institution implementation",
            "url": "https://github.com/deepseek-ai/DeepSeek-V3.2-Exp"
          }
        }
      ],
      "origin": "HUMAN_LEGACY",
      "record_profile": "wrs-local-research-notes-0.1",
      "record_type": "research_note",
      "rrp_version": "0.1",
      "scope": {
        "description": "MLA/MQA and prior RL are inherited. Sparse main attention does not eliminate the indexer's pairwise scoring cost or guarantee equivalence to dense attention. September experimental and December full releases share DSA; Speciale benchmark claims are not official competition participation.",
        "domain": "Sparse attention and reasoning-training systems"
      },
      "sources": [
        {
          "citation": "DeepSeek-V3.2: Pushing the Frontier of Open Large Language Models",
          "url": "https://arxiv.org/abs/2512.02556v1"
        },
        {
          "citation": "DeepSeek-AI: lightning-indexed sparse attention and agentic training — author/institution implementation",
          "url": "https://github.com/deepseek-ai/DeepSeek-V3.2-Exp"
        }
      ],
      "statement": "# DeepSeek-AI: lightning-indexed sparse attention and agentic training\n\nAdded a low-cost lightning indexer with top-k latent KV selection and staged dense-to-sparse continued training, alongside scalable reasoning/tool-use training and synthetic interactive tasks.\n\n## Scientific scope\n\nMLA/MQA and prior RL are inherited. Sparse main attention does not eliminate the indexer's pairwise scoring cost or guarantee equivalence to dense attention. September experimental and December full releases share DSA; Speciale benchmark claims are not official competition participation.\n\n## Reading status\n\nThis is a source-linked WRS catalogue note. A full explanatory guide and independent scientific reconstruction have not been completed for this entry. Read the identified original work for its complete narrative and arguments.\n\n## People and systems\n\n\n\nSource-listed authors or explicitly credited technical contributors for the cited versions. Family entries retain source-specific attribution; a combined list is not one original-paper byline or equal individual work. Names alone do not establish resolved identities.\n\n\n\n## Original sources\n\n- [DeepSeek-V3.2: Pushing the Frontier of Open Large Language Models](https://arxiv.org/abs/2512.02556v1)\n- [DeepSeek-AI: lightning-indexed sparse attention and agentic training — author/institution implementation](https://github.com/deepseek-ai/DeepSeek-V3.2-Exp)\n",
      "summary": "Added a low-cost lightning indexer with top-k latent KV selection and staged dense-to-sparse continued training, alongside scalable reasoning/tool-use training and synthetic interactive tasks.",
      "title": "DeepSeek-AI: lightning-indexed sparse attention and agentic training"
    },
    "record_id": "urn:uuid:20ea488e-a181-4910-a7a0-4cd727a70563",
    "version": 1,
    "source_ref": {
      "record_id": "urn:uuid:20ea488e-a181-4910-a7a0-4cd727a70563",
      "version": 1,
      "content_hash": "sha256:afb45ab5f3c1b691ab08849623b1ff490333d2351fb48246a54ec1af7127654a"
    },
    "export_view": "wrs-public-editorial-view-1",
    "public_view_hash": "sha256:b70e9be19b6769fd54fdee5be4d75b0b41613f1cb507da5169a0ce51f3bc0283"
  },
  "target": {
    "record_id": "urn:uuid:20ea488e-a181-4910-a7a0-4cd727a70563",
    "version": 1,
    "content_hash": "sha256:afb45ab5f3c1b691ab08849623b1ff490333d2351fb48246a54ec1af7127654a"
  },
  "published_grade": {
    "profile": "wrs-published-grade-1",
    "edition": "2026-10-28.1",
    "id": "deepseek-dsa-indexer-agent-training-2025",
    "entry_id": "deepseek-dsa-indexer-agent-training-2025",
    "revision": 1,
    "title": "DeepSeek-AI: lightning-indexed sparse attention and agentic training",
    "context": "Sparse attention and reasoning-training systems",
    "track": "Human-designed research method or resource",
    "band": "A",
    "modifier": null,
    "status": "PROVISIONAL_EDITORIAL",
    "assessed_on": "2026-10-16",
    "evidence_cutoff": "2026-10-16",
    "achievement_state": "SUPPORTED_SCOPED_RESULT",
    "what_changed": "Added a low-cost lightning indexer with top-k latent KV selection and staged dense-to-sparse continued training, alongside scalable reasoning/tool-use training and synthetic interactive tasks.",
    "scope": "Added a low-cost lightning indexer with top-k latent KV selection and staged dense-to-sparse continued training, alongside scalable reasoning/tool-use training and synthetic interactive tasks. MLA/MQA and prior RL are inherited. Sparse main attention does not eliminate the indexer's pairwise scoring cost or guarantee equivalence to dense attention. September experimental and December full releases share DSA; Speciale benchmark claims are not official competition participation.",
    "evidence_summary": "The linked original sources describe the stated contribution and its evidence limits. WRS has not independently rerun the experiments or certified a proof. MLA/MQA and prior RL are inherited. Sparse main attention does not eliminate the indexer's pairwise scoring cost or guarantee equivalence to dense attention. September experimental and December full releases share DSA; Speciale benchmark claims are not official competition participation.",
    "references": [
      {
        "title": "DeepSeek-V3.2: Pushing the Frontier of Open Large Language Models",
        "url": "https://arxiv.org/abs/2512.02556v1",
        "locator": "Original contribution and stated assumptions, evaluation scope and limitations; source-linked assessment, not independent execution or proof certification."
      },
      {
        "title": "DeepSeek-AI: lightning-indexed sparse attention and agentic training — author/institution implementation",
        "url": "https://github.com/deepseek-ai/DeepSeek-V3.2-Exp",
        "locator": "Original contribution and stated assumptions, evaluation scope and limitations; source-linked assessment, not independent execution or proof certification."
      }
    ],
    "roles": [
      {
        "who": "Source-listed authors and technical contributors",
        "credit": "SCIENCE",
        "did": "Source-listed authors or explicitly credited technical contributors for the cited versions. Family entries retain source-specific attribution; a combined list is not one original-paper byline or equal individual work. Names alone do not establish resolved identities."
      }
    ],
    "target": {
      "record_id": "urn:uuid:20ea488e-a181-4910-a7a0-4cd727a70563",
      "version": 1,
      "content_hash": "sha256:afb45ab5f3c1b691ab08849623b1ff490333d2351fb48246a54ec1af7127654a"
    },
    "scientific_correctness_certified": false,
    "independent_expert_calibration": false,
    "method": "PROPRIETARY",
    "points": "105000"
  },
  "canonical_agent_submission": false,
  "scientific_verification_awarded": false,
  "reader_guide": null,
  "reading_status": "SOURCE_LINKED_NOTE"
}