{
  "schema_version": 1,
  "title": "Reproduction audit: Hybrid Sequence Models",
  "emoji": "\ud83d\udd2c",
  "proposed_target_slug": "repro-hybrid-seq-82ejxjzg6r",
  "publication_status": "local-only; no Space created or modified",
  "paper": {
    "arxiv_id": "2603.08859v1",
    "openreview_id": "82EJxJzG6r"
  },
  "updated_at": "2026-07-28T00:00:00+00:00",
  "root": {
    "slug": "index",
    "title": "Reproduction audit",
    "file": "pages/index.md",
    "children": [
      {
        "slug": "executive-summary",
        "title": "Executive summary",
        "file": "pages/executive-summary/page.md",
        "children": []
      },
      {
        "slug": "claim-1-theorem-3-3-literal-bound",
        "title": "Claim 1: Theorem 3.3 proves that any k-layer state-space model solving the function-composition tasks under injectivity conditions must have total log state-space size scaling as \u03a9(m\u00b7log|V| \u2212 q\u00b7log|Y|), linear in the hidden dimension m (Theorem 3.3).",
        "file": "pages/claim-1-theorem-3-3-literal-bound/page.md",
        "children": []
      },
      {
        "slug": "claim-2-theorem-3-7-window-bound",
        "title": "Claim 2: Theorem 3.7 proves that sliding-window Transformers solving the same tasks under a local-sensitivity condition require total window size scaling with the context-dependency range R (Theorem 3.7).",
        "file": "pages/claim-2-theorem-3-7-window-bound/page.md",
        "children": []
      },
      {
        "slug": "claim-3-theorem-4-3-selective-copy",
        "title": "Claim 3: Theorem 4.3 constructs a two-layer hybrid (Mamba + attention) model that solves the selective copying task using embedding dimension O(max(log|V|, log L)) and working memory \u00d5(N), versus \u03a9(L) required by pure Transformers (Theorem 4.3).",
        "file": "pages/claim-3-theorem-4-3-selective-copy/page.md",
        "children": []
      },
      {
        "slug": "claim-4-theorem-4-6-associative-recall",
        "title": "Claim 4: Theorem 4.6 constructs a three-layer hybrid model that achieves 99% accuracy on the associative recall task using embedding dimension O(max(log|V|, log L)) and window size \u00d5(|V|) (Theorem 4.6).",
        "file": "pages/claim-4-theorem-4-6-associative-recall/page.md",
        "children": []
      },
      {
        "slug": "claim-5-figure-4-selective-copy-learning",
        "title": "Claim 5: On the selective copying task, the learned hybrid model reaches perfect accuracy with roughly 2,000 parameters while pure Transformer/SSM models need roughly 12,000 parameters to match it, a 6x parameter gap (Figure 4).",
        "file": "pages/claim-5-figure-4-selective-copy-learning/page.md",
        "children": []
      },
      {
        "slug": "claim-6-figures-5-6-associative-recall-learning",
        "title": "Claim 6: On multi-key associative recall, the hybrid model reaches 60% accuracy using 6x fewer parameters than pure Transformers, which plateau near 40% accuracy on single-key associative recall (Figures 5-6).",
        "file": "pages/claim-6-figures-5-6-associative-recall-learning/page.md",
        "children": []
      }
    ]
  },
  "routes_built": {
    "claim_pages": [
      "pages/claim-1-theorem-3-3-literal-bound/page.md",
      "pages/claim-2-theorem-3-7-window-bound/page.md",
      "pages/claim-3-theorem-4-3-selective-copy/page.md",
      "pages/claim-4-theorem-4-6-associative-recall/page.md",
      "pages/claim-5-figure-4-selective-copy-learning/page.md",
      "pages/claim-6-figures-5-6-associative-recall-learning/page.md"
    ]
  }
}