{
  "content_status": "PASS_WITH_AGGREGATE_CONFLICT_AND_MECHANISM_LIMITS",
  "risk_banner": "Useful historical-covariate linear baseline; Traffic DLinear row/average conflict prevents accepting that claimed average plugin gain. Which printed values match the real runs remains unknown.",
  "independent_original_pdf_visual_pages": [
    4,
    8,
    12,
    13
  ],
  "independent_original_pdf_text_pages": [
    4,
    5,
    6,
    7,
    8,
    9,
    10,
    11,
    12,
    13
  ],
  "confirmed_paper_findings": [
    {
      "id": "aggregate_conflict",
      "locations": [
        "PDF p12 Table11",
        "p8 Table5",
        "p6 Table2"
      ],
      "finding": "DLinear Traffic original row MSE .268/.302/.298/.340 has arithmetic mean .302, not printed .323. MAE .351/.387/.384/.416 gives .3845, not .404. Plugin row means .308 and .39125. Thus if the four rows are correct, average changes worsen rather than improve. The summary repeats the conflicting aggregate.",
      "scope": "Only internal inconsistency is proven. Actual run results could correspond to rows, aggregates, or neither; not determined. E appended signed scope clarification without score changes."
    },
    {
      "id": "equivalence_scope",
      "locations": [
        "PDF p4 Eqs7–11",
        "pp11–13 AppendixB Eqs14–18"
      ],
      "finding": "Reparameterization K'=(1-alpha)K+alphaS correctly represents residual fusion when the kernel/padding can implement identity and bias is correspondingly scaled. This is function-family equivalence, not equality under identical unchanged kernel values, nor proof of superior expressivity. Linear projection after position embedding turns fixed learned position terms into an output offset.",
      "scope": "Optimization, initialization and regularization may differ; no implemented model failure shown."
    },
    {
      "id": "conditional_performance",
      "locations": [
        "PDF p6 Table2",
        "p7 Table3",
        "p8 Table5",
        "p12 Table11"
      ],
      "finding": "ECL and Traffic many-to-one average MSE .323/.152 beat TimeXer .327/.156, but ECL720 .403 loses to .365. EPF MAE .266 is worse than TimeXer .265. DLinear ECL MAE .457 to .463 worsens. Traffic96 DLinear MSE .268 to .338 worsens. Multivariate Traffic .483 is worse than iTransformer .428.",
      "scope": "No universal plugin improvement or uniform SOTA; no seed uncertainty accompanies these comparisons."
    },
    {
      "id": "historical_covariates_and_masking",
      "locations": [
        "PDF p4 model input equations",
        "p8 Table6",
        "p5 setup"
      ],
      "finding": "The stated model consumes historical covariates, not future known covariates. Table6 NP is labelled96-to-96 while main EPF setup is168-to-24. Timing of zero/random masks relative to RevIN statistics is not specified sufficiently to assess the all-endogenous-masked claim.",
      "scope": "Potential protocol ambiguity only, not demonstrated test leakage."
    },
    {
      "id": "reproducibility_and_scaling",
      "locations": [
        "PDF p11 AppendixA",
        "p4 Sec3.6"
      ],
      "finding": "RTX4090, Adam, learning-rate range1e-5 to1e-3, batch8 to64 and10epochs are supplied. Original-source baseline settings are stated, but exact comparable search budgets and seed uncertainty are not fully resolved. Multivariate N-input/N-output kernel3 has3N^2 weights, which is2229132 atN862 before other components.",
      "scope": "Many-to-one efficiency does not establish negligible cost at every multivariate dimension. No full implementation or data archive verification conducted."
    }
  ],
  "readme_source": {
    "url": "https://github.com/mumiao2000/CrossLinear/blob/main/README.md",
    "git_blob_sha": "31d276de7510b04c0585b135f2dbf68f0350d161",
    "independently_read": true,
    "observations": "Python3.8.20, requirements, dataset Drive entry, and separate exogenous_forecast and multivariate_forecast scripts are documented. Commands were not executed; external dataset/Zenodo archive not verified."
  },
  "known_limits": [
    "All five original freeze hashes verified before new reviews were read. Full historical reviews remain missing; legacy summaries contribute zero votes.",
    "Independent reading order is supported by written attestations, not independent start-time telemetry.",
    "Central equations, key tables and factual risks checked; not every sentence or all cited predecessors were independently audited.",
    "No model experiments or training executed. Tentative intervals are judgment ranges, not confidence intervals."
  ],
  "errata_to_apply": [
    {
      "reviewer": "E",
      "path": "/workspace/shared/candidate_review_batch001/reviewer_e/ATT-0086/CLARIFICATION_20261108.json",
      "companion_files": [
        "/workspace/shared/candidate_review_batch001/reviewer_e/ATT-0086/CLARIFICATION_20261108.md"
      ]
    }
  ],
  "accepted_errata": [
    {
      "reviewer": "E",
      "path": "/workspace/shared/candidate_review_batch001/reviewer_e/ATT-0086/CLARIFICATION_20261108.json",
      "sha256": "7876e397bb3c4e99846104b0b6bfd88b1696ea29139d9a75c9f8ccb3bfb23a83",
      "companion_sha256": {
        "CLARIFICATION_20261108.md": "30f5050dc93ea032edcd61e69a1b3b46d193ddb3ae81a935021c0e687789cc9f"
      },
      "same_vote_not_additional_reviewer": true,
      "scores_changed": false
    }
  ],
  "input_pdf_sha256": "0235fd4cd7a51ae05b313532abdc7d0fe4dc86ac23bc29c1105f5e3cccb814f8",
  "no_experiments_run": true
}