{
  "schema_version": 1,
  "builder_plan": {
    "scenario": "A self-hosted model uses more compute or memory than your workload can justify.",
    "change": "Check for a usable pruned checkpoint and runtime support before attempting model surgery.",
    "baseline": "The original unpruned model at the same precision, context, and load.",
    "steps": [
      "Locate the paper's implementation and check its license, model support, data needs, and hardware requirements. Stop here if you cannot obtain the required artifacts.",
      "Save a small set of representative product tasks and expected outcomes. Keep evaluation tasks out of any training or tuning, and choose quality and cost limits before running.",
      "Run the baseline and the proposed change on identical inputs. Keep other settings fixed and repeat paired runs to expose variation.",
      "Save per-task outputs, failures, timings, and all usage costs. Review regressions before expanding the test."
    ],
    "measure": "Task quality, actual GPU memory, serving throughput, and preparation cost.",
    "decision": "Keep the change only if it meets your preselected quality and latency limits and improves the metric you care about after setup and operating costs. Otherwise retain the baseline. This is an exploration plan, not a validated recommendation."
  },
  "builder_review": null,
  "arxiv_id": "2610.23571",
  "title": "BMuon: Budgeting Muon's Spectral Geometry for Structured Compressibility",
  "summary": "BMuon replaces the Muon optimizer's spectral-norm geometry with a shared channel-allocation budget so that models trained with it retain accuracy under structured MLP pruning, with allocations also serving as a pruning criterion. It costs a modified optimizer with a minorization-maximization surrogate step, applies only when the reader trains models from scratch, and adds implementation complexity over standard Muon or AdamW. It can fail if the budget is set too tight, degrading baseline training quality, and its benefits are irrelevant to readers who only fine-tune or consume pretrained models.",
  "rationale": "The technique is only actionable during full model pretraining, which exceeds the reader's infrastructure, so it is relevant to track but not adoptable now.",
  "technique": "Budgeted spectral optimizer for pruning-robust training",
  "family": "esparsidade_e_poda",
  "recommendation": "observar",
  "published": "2026-10-16",
  "as_of": "2026-10-19",
  "editorial_status": "indexed",
  "source_url": "https://arxiv.org/abs/2610.23571",
  "independent_implementations": 0,
  "total_implementations": 0,
  "stars": 0,
  "citations": null,
  "claims": [
    {
      "claim_id": "claim-01",
      "statement": "BMuon replaces the Muon optimizer's spectral-norm geometry with a shared channel-allocation budget so that models trained with it retain accuracy under structured MLP pruning, with allocations also serving as a pruning criterion. It costs a modified optimizer with a minorization-maximization surrogate step, applies only when the reader trains models from scratch, and adds implementation complexity over standard Muon or AdamW. It can fail if the budget is set too tight, degrading baseline training quality, and its benefits are irrelevant to readers who only fine-tune or consume pretrained models.",
      "basis": "inferred",
      "result": "",
      "baseline": "",
      "conditions": "",
      "source_url": "https://arxiv.org/abs/2610.23571",
      "source_page": null,
      "source_excerpt": ""
    },
    {
      "claim_id": "claim-02",
      "statement": "BMuon improves over Muon under gradual structured MLP pruning on ViT-S/16 ImageNet-1K and under post-training MLP channel pruning on a 1B-parameter LLaMA-style model; no multiplicative factor is reported in the abstract.",
      "basis": "inferred",
      "result": "",
      "baseline": "",
      "conditions": "",
      "source_url": "https://arxiv.org/abs/2610.23571",
      "source_page": null,
      "source_excerpt": ""
    }
  ],
  "exposure_map": [
    {
      "dimension": "quality",
      "basis": "not_evaluated",
      "finding": "",
      "interpretation": "",
      "limitation": "",
      "next_check": "",
      "source_url": "",
      "source_page": null,
      "source_excerpt": ""
    },
    {
      "dimension": "compute",
      "basis": "not_evaluated",
      "finding": "",
      "interpretation": "",
      "limitation": "",
      "next_check": "",
      "source_url": "",
      "source_page": null,
      "source_excerpt": ""
    },
    {
      "dimension": "latency",
      "basis": "not_evaluated",
      "finding": "",
      "interpretation": "",
      "limitation": "",
      "next_check": "",
      "source_url": "",
      "source_page": null,
      "source_excerpt": ""
    },
    {
      "dimension": "operations",
      "basis": "not_evaluated",
      "finding": "",
      "interpretation": "",
      "limitation": "",
      "next_check": "",
      "source_url": "",
      "source_page": null,
      "source_excerpt": ""
    },
    {
      "dimension": "compatibility",
      "basis": "not_evaluated",
      "finding": "",
      "interpretation": "",
      "limitation": "",
      "next_check": "",
      "source_url": "",
      "source_page": null,
      "source_excerpt": ""
    },
    {
      "dimension": "security",
      "basis": "not_evaluated",
      "finding": "",
      "interpretation": "",
      "limitation": "",
      "next_check": "",
      "source_url": "",
      "source_page": null,
      "source_excerpt": ""
    },
    {
      "dimension": "data_and_training",
      "basis": "not_evaluated",
      "finding": "",
      "interpretation": "",
      "limitation": "",
      "next_check": "",
      "source_url": "",
      "source_page": null,
      "source_excerpt": ""
    },
    {
      "dimension": "reproducibility",
      "basis": "not_evaluated",
      "finding": "",
      "interpretation": "",
      "limitation": "",
      "next_check": "",
      "source_url": "",
      "source_page": null,
      "source_excerpt": ""
    }
  ],
  "risks": [],
  "minimum_test": [],
  "open_questions": [],
  "independent_tests": [],
  "lab_reproductions": [],
  "report_available": false,
  "validation_tier": "unknown",
  "evidence_tier": "unknown",
  "infrastructure_basis": "unknown",
  "equations": [],
  "equations_status": "not_fetched",
  "core_kind": "",
  "equations_fetched_at": "",
  "equations_source_url": "",
  "equations_source_sha256": "",
  "exposures_reviewed_at": "",
  "exposures_source_sha256": ""
}