Step 2: tool results

get_claim

{
  "include": [
    "arguments",
    "subclaims",
    "related"
  ],
  "claim_id": "d3311aac-502f-4e69-ab97-53ee2b6de622"
}

{"claim":{"id":"d3311aac-502f-4e69-ab97-53ee2b6de622","kind":"claim","canonical_form":"With output-score feature selection, SAE steering is competitive with supervised steering methods on AxBench.","claim_type":"empirical_derived","state":"active","decomposition_status":"complete","importance":0.55,"contestation":0.6,"domains":[],"tags":[{"slug":"mechanistic-interpretability","name":"Mechanistic Interpretability"},{"slug":"ai-capability-evaluation","name":"AI Capability Evaluation"},{"slug":"large-language-models","name":"Large language models"}],"steward_state":"running","stewarded_at":"2026-11-14T20:40:27.287Z","canonical_direction_note":"The source asserts the positive result that SAE steering can be competitive with established supervised methods when effective feature selection is used.","created_at":"2026-11-13T00:25:54.664Z","updated_at":"2026-11-14T20:39:56.965Z","page_url":"https://minerval.ai/claims/d3311aac-502f-4e69-ab97-53ee2b6de622"},"assessment":null,"coverage":{"examined":false,"on":[],"note":"No judgment of this node is in force. That is not a finding either way."},"quantity":null,"threshold_of":null,"question":null,"formalization":null,"formalization_pending":null,"formalization_history":[],"verification":null,"lean_checks":[],"lean_checks_total":0,"waiting_messages":[{"trigger":"structure_and_assess","from":"the system","sender_node_id":null,"sent_at":"2026-11-13T00:25:54.908Z","paid":false,"summary":"New claim onboarded. Give it the structure and assessment its importance warrants; a minor or settled claim may need no decomposition."},{"trigger":"mandate_plan","from":"a mandate's plan","sender_node_id":null,"sent_at":"2026-11-14T20:39:56.967Z","paid":true,"summary":"First assessment. Read the source that reports this result and the original benchmark (arXiv 2501.17148; not held as a source, open its abstract or HTML page directly). Compare like with like: same mo…"}],"requests":[],"requests_total":0,"subclaim_count":0,"instance_count":1,"source_map":null,"related":[],"arguments":[],"subclaims":[],"subclaims_assessed":0,"unplaced_argument_passages":[{"instance_id":"3653e80d-0f8c-48a5-af4a-175cdf96df53","source":{"id":"d91b7b04-3766-4951-a356-d527ea7eab4a","url":"https://arxiv.org/pdf/2505.20063","title":"SAEs Are Good for Steering – If You Select the Right Features"},"verbatim_text":"With output score\nfiltering, SAE features top steering performance\namong the representation-based methods at L31;\nfor L20, they get 90.7% of the performance of the\nbest method. This is in contrast with the results of\nWu et al., where SAEs significantly underperforms\nReFT-r1, a weakly supervised method they propose\nas a competitive alternative to prompting.","context":"Comparison of filtered SAE scores with AxBench-reported baseline scores in Table 1","stance":"advances","direction":"for"}],"unplaced_argument_passages_total":1}

get_claim_instances

{
  "claim_id": "d3311aac-502f-4e69-ab97-53ee2b6de622"
}

{
  "claim_id": "d3311aac-502f-4e69-ab97-53ee2b6de622",
  "total": 1,
  "limit": 20,
  "offset": 0,
  "has_more": false,
  "instances": [
    {
      "id": "7274c53b-26ee-4570-ba36-88100cd9f921",
      "source_id": "d91b7b04-3766-4951-a356-d527ea7eab4a",
      "verbatim_text": "These re-\nsults demonstrates that with effective feature selec-\ntion SAEs are comparable with existing methods,\nincluding supervised or weakly-supervised meth-\nods which require concept-specific datasets.",
      "context": "Section 4.4, AxBench evaluation on instruction-tuned Gemma-2-9B, Concept500, layers 20 and 31",
      "stance": "affirms",
      "confidence": 0.85,
      "speaker": null,
      "publication": null,
      "source_date": null,
      "link": null,
      "source_title": "SAEs Are Good for Steering – If You Select the Right Features",
      "source_type": "preprint",
      "source_url": "https://arxiv.org/pdf/2505.20063",
      "source_published_date": "2025",
      "source_authors": [
        "Dana Arad",
        "Aaron Mueller",
        "Yonatan Belinkov"
      ],
      "source_publisher": "Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing",
      "source_doi": "10.18653/v1/2025.emnlp-main.519",
      "source_interest": null,
      "recorded_by": "extractor",
      "display_rank": null,
      "reading": null,
      "draws_on": [],
      "story": {
        "standing": "untraced",
        "basis": "none",
        "depth": 0,
        "diverges": false,
        "downstream_total": 0
      }
    }
  ]
}