{
  "$schema": "https://themachinepress.com/schemas/story-v1.schema.json",
  "schema_version": "1.0.0",
  "document_type": "machine_press_story",
  "story": {
    "story_id": "mp-2026-10-25-005",
    "source_story_id": "tmp-story-valuspec-branch-filtering",
    "edition_id": "mp-2026-10-25-morning-0107",
    "edition_url": "https://themachinepress.com/edition/2026-10-25",
    "position": 5,
    "story_type": "dispatch",
    "section": "infrastructure",
    "editorial_classification": "editorial",
    "headline": "The Target Model Checks Fewer Bad Branches",
    "slug": "the-target-model-checks-fewer-bad-branches",
    "dek": "ValuSpec filters low-value speculative candidates before expensive verification while preserving the target model's greedy output exactly.",
    "summary": "ValuSpec filters low-value speculative candidates before expensive verification while preserving the target model's greedy output exactly.",
    "body_text": "Retrieval-based speculative decoding can reuse historical drafts, but larger candidate trees accumulate branches unlikely to survive target-model verification. ValuSpec scores and filters a supplied tree before that expensive step. Its threshold comes from the target model's greedy trajectories on a calibration split, and transfer-based supervision aligns valuation with the target without parameter training. The remaining tree is still verified under an output-identical greedy rule, so accepted tokens match ordinary autoregressive decoding. The authors report up to 2.38-times end-to-end speedup in retrieval-based settings and 2.78-times when attached to EAGLE-3. Those maxima depend on the tested models, tasks, calibration and serving setup; the method does not accelerate every workload by that amount.",
    "why_it_matters": "ValuSpec filters low-value speculative candidates before expensive verification while preserving the target model's greedy output exactly.",
    "limitations": [
      "Those maxima depend on the tested models, tasks, calibration and serving setup; the method does not accelerate every workload by that amount."
    ],
    "importance": 8,
    "canonical_url": "https://themachinepress.com/story/mp-2026-10-25-005/the-target-model-checks-fewer-bad-branches",
    "json_url": "https://themachinepress.com/story/mp-2026-10-25-005.json",
    "first_published_at": "2026-10-25T09:00:00.000-04:00",
    "modified_at": "2026-10-25T09:00:00.000-04:00",
    "content_status": "new",
    "is_carryover": false,
    "carryover_reason": null,
    "key_claims": [
      {
        "claim_id": "claim-mp-2026-10-25-005-001",
        "text": "ValuSpec filters low-value speculative candidates before expensive verification while preserving the target model's greedy output exactly.",
        "source_ids": [
          "source-2026-10-25-005"
        ],
        "qualification": "Those maxima depend on the tested models, tasks, calibration and serving setup; the method does not accelerate every workload by that amount."
      }
    ],
    "source_ids": [
      "source-2026-10-25-005"
    ],
    "tags": [
      "inference",
      "speculative decoding",
      "serving",
      "efficiency"
    ],
    "image_url": null,
    "corrections": []
  },
  "sources": [
    {
      "source_id": "source-2026-10-25-005",
      "title": "arXiv preprint 2610.31338, original v1",
      "publisher": "arXiv",
      "url": "https://arxiv.org/abs/2610.31338v1",
      "canonical_url": "https://arxiv.org/abs/2610.31338v1",
      "source_type": "primary_research",
      "is_primary_source": true,
      "published_at": "2026-10-22T10:23:26.000-04:00",
      "accessed_at": "2026-10-25T08:17:14.361-04:00",
      "supports_claim_ids": [
        "claim-mp-2026-10-25-005-001"
      ]
    }
  ],
  "corrections": [],
  "publisher": {
    "name": "The Machine Press",
    "url": "https://themachinepress.com",
    "description": "A daily newspaper for the age of artificial intelligence."
  },
  "cite_this_report": {
    "title": "The Target Model Checks Fewer Bad Branches",
    "publisher": "The Machine Press",
    "published_at": "2026-10-25T09:00:00.000-04:00",
    "canonical_url": "https://themachinepress.com/story/mp-2026-10-25-005/the-target-model-checks-fewer-bad-branches"
  }
}