[ { "id": "IF01", "dimension": "Task Fulfillment / Compliance", "criterion": "Requirement: Distinguish the identity shortcut from the residual branch.\nVerification method: Check the actual delivered content against the task statement and original materials; the author's self-report cannot replace evidence from the work.\nPass criterion: Fully satisfy this requirement; omissions, errors, or partial satisfaction result in No. Evaluation-tool failures cannot be grounds for No.\nVerification boundary: Check against the locked task statement and the provided original materials; do not substitute the author's self-report for inspection.\nTask basis: prompt.md, original line 9: 1. Distinguish the identity shortcut from the residual branch; explain learning F(x) and producing F(x)+x.", "answers": { "Yes": "Complete all necessary checks for this criterion, with actual evidence supporting satisfaction of the requirement; equivalent implementations within the task's scope are allowed.", "No": "At least one necessary requirement is not satisfied; identify evidence of actual omissions, errors, or partial satisfaction." }, "compliance_groups": [ "G01" ], "benchmark_source": "vas" }, { "id": "IF02", "dimension": "Task Fulfillment / Compliance", "criterion": "Requirement: Explain learning F(x).\nVerification method: Check the actual delivered content against the task statement and original materials; the author's self-report cannot replace evidence from the work.\nPass criterion: Fully satisfy this requirement; omissions, errors, or partial satisfaction result in No. Evaluation-tool failures cannot be grounds for No.\nVerification boundary: Check against the locked task statement and the provided original materials; do not substitute the author's self-report for inspection.\nTask basis: prompt.md, original line 9: 1. Distinguish the identity shortcut from the residual branch; explain learning F(x) and producing F(x)+x.", "answers": { "Yes": "Complete all necessary checks for this criterion, with actual evidence supporting satisfaction of the requirement; equivalent implementations within the task's scope are allowed.", "No": "At least one necessary requirement is not satisfied; identify evidence of actual omissions, errors, or partial satisfaction." }, "compliance_groups": [ "G01" ], "benchmark_source": "vas" }, { "id": "IF03", "dimension": "Task Fulfillment / Compliance", "criterion": "Requirement: Explain producing F(x)+x.\nVerification method: Check the actual delivered content against the task statement and original materials; the author's self-report cannot replace evidence from the work.\nPass criterion: Fully satisfy this requirement; omissions, errors, or partial satisfaction result in No. Evaluation-tool failures cannot be grounds for No.\nVerification boundary: Check against the locked task statement and the provided original materials; do not substitute the author's self-report for inspection.\nTask basis: prompt.md, original line 9: 1. Distinguish the identity shortcut from the residual branch; explain learning F(x) and producing F(x)+x.", "answers": { "Yes": "Complete all necessary checks for this criterion, with actual evidence supporting satisfaction of the requirement; equivalent implementations within the task's scope are allowed.", "No": "At least one necessary requirement is not satisfied; identify evidence of actual omissions, errors, or partial satisfaction." }, "compliance_groups": [ "G01" ], "benchmark_source": "vas" }, { "id": "IF04", "dimension": "Task Fulfillment / Compliance", "criterion": "Requirement: Cover ImageNet experiments.\nVerification method: Check the actual delivered content against the task statement and original materials; the author's self-report cannot replace evidence from the work.\nPass criterion: Fully satisfy this requirement; omissions, errors, or partial satisfaction result in No. Evaluation-tool failures cannot be grounds for No.\nVerification boundary: Check against the locked task statement and the provided original materials; do not substitute the author's self-report for inspection.\nTask basis: prompt.md, original line 10: 2. Cover at least ImageNet and CIFAR-10 experiments, correctly matching metrics, model depths, and datasets.", "answers": { "Yes": "Complete all necessary checks for this criterion, with actual evidence supporting satisfaction of the requirement; equivalent implementations within the task's scope are allowed.", "No": "At least one necessary requirement is not satisfied; identify evidence of actual omissions, errors, or partial satisfaction." }, "compliance_groups": [ "G02" ], "benchmark_source": "vas" }, { "id": "IF05", "dimension": "Task Fulfillment / Compliance", "criterion": "Requirement: Cover CIFAR-10 experiments.\nVerification method: Check the actual delivered content against the task statement and original materials; the author's self-report cannot replace evidence from the work.\nPass criterion: Fully satisfy this requirement; omissions, errors, or partial satisfaction result in No. Evaluation-tool failures cannot be grounds for No.\nVerification boundary: Check against the locked task statement and the provided original materials; do not substitute the author's self-report for inspection.\nTask basis: prompt.md, original line 10: 2. Cover at least ImageNet and CIFAR-10 experiments, correctly matching metrics, model depths, and datasets.", "answers": { "Yes": "Complete all necessary checks for this criterion, with actual evidence supporting satisfaction of the requirement; equivalent implementations within the task's scope are allowed.", "No": "At least one necessary requirement is not satisfied; identify evidence of actual omissions, errors, or partial satisfaction." }, "compliance_groups": [ "G02" ], "benchmark_source": "vas" }, { "id": "IF06", "dimension": "Task Fulfillment / Compliance", "criterion": "Requirement: Match experimental values to datasets.\nVerification method: Check the actual delivered content against the task statement and original materials; the author's self-report cannot replace evidence from the work.\nPass criterion: Fully satisfy this requirement; omissions, errors, or partial satisfaction result in No. Evaluation-tool failures cannot be grounds for No.\nVerification boundary: Check against the locked task statement and the provided original materials; do not substitute the author's self-report for inspection.\nTask basis: prompt.md, original line 10: 2. Cover at least ImageNet and CIFAR-10 experiments, correctly matching metrics, model depths, and datasets.", "answers": { "Yes": "Complete all necessary checks for this criterion, with actual evidence supporting satisfaction of the requirement; equivalent implementations within the task's scope are allowed.", "No": "At least one necessary requirement is not satisfied; identify evidence of actual omissions, errors, or partial satisfaction." }, "compliance_groups": [ "G02" ], "benchmark_source": "vas" }, { "id": "IF07", "dimension": "Task Fulfillment / Compliance", "criterion": "Requirement: Match experimental metrics to model depths.\nVerification method: Check the actual delivered content against the task statement and original materials; the author's self-report cannot replace evidence from the work.\nPass criterion: Fully satisfy this requirement; omissions, errors, or partial satisfaction result in No. Evaluation-tool failures cannot be grounds for No.\nVerification boundary: Check against the locked task statement and the provided original materials; do not substitute the author's self-report for inspection.\nTask basis: prompt.md, original line 10: 2. Cover at least ImageNet and CIFAR-10 experiments, correctly matching metrics, model depths, and datasets.", "answers": { "Yes": "Complete all necessary checks for this criterion, with actual evidence supporting satisfaction of the requirement; equivalent implementations within the task's scope are allowed.", "No": "At least one necessary requirement is not satisfied; identify evidence of actual omissions, errors, or partial satisfaction." }, "compliance_groups": [ "G02" ], "benchmark_source": "vas" }, { "id": "IF08", "dimension": "Task Fulfillment / Compliance", "criterion": "Requirement: Do not simply equate degradation in training deeper networks with overfitting.\nVerification method: Check the actual delivered content against the task statement and original materials; the author's self-report cannot replace evidence from the work.\nPass criterion: Fully satisfy this requirement; omissions, errors, or partial satisfaction result in No. Evaluation-tool failures cannot be grounds for No.\nVerification boundary: Check against the locked task statement and the provided original materials; do not substitute the author's self-report for inspection.\nTask basis: prompt.md, original line 11: 3. Do not simply describe degradation in training deeper networks as overfitting. Explicitly identify limitations and extrapolations.", "answers": { "Yes": "Complete all necessary checks for this criterion, with actual evidence supporting satisfaction of the requirement; equivalent implementations within the task's scope are allowed.", "No": "At least one necessary requirement is not satisfied; identify evidence of actual omissions, errors, or partial satisfaction." }, "compliance_groups": [ "G03" ], "benchmark_source": "vas" }, { "id": "IF09", "dimension": "Task Fulfillment / Compliance", "criterion": "Requirement: Explicitly identify limitations of the method or experiments.\nVerification method: Check the actual delivered content against the task statement and original materials; the author's self-report cannot replace evidence from the work.\nPass criterion: Fully satisfy this requirement; omissions, errors, or partial satisfaction result in No. Evaluation-tool failures cannot be grounds for No.\nVerification boundary: Check against the locked task statement and the provided original materials; do not substitute the author's self-report for inspection.\nTask basis: prompt.md, original line 11: 3. Do not simply describe degradation in training deeper networks as overfitting. Explicitly identify limitations and extrapolations.", "answers": { "Yes": "Complete all necessary checks for this criterion, with actual evidence supporting satisfaction of the requirement; equivalent implementations within the task's scope are allowed.", "No": "At least one necessary requirement is not satisfied; identify evidence of actual omissions, errors, or partial satisfaction." }, "compliance_groups": [ "G03" ], "benchmark_source": "vas" }, { "id": "IF10", "dimension": "Task Fulfillment / Compliance", "criterion": "Requirement: Explicitly distinguish extrapolations from supported conclusions.\nVerification method: Check the actual delivered content against the task statement and original materials; the author's self-report cannot replace evidence from the work.\nPass criterion: Fully satisfy this requirement; omissions, errors, or partial satisfaction result in No. Evaluation-tool failures cannot be grounds for No.\nVerification boundary: Check against the locked task statement and the provided original materials; do not substitute the author's self-report for inspection.\nTask basis: prompt.md, original line 11: 3. Do not simply describe degradation in training deeper networks as overfitting. Explicitly identify limitations and extrapolations.", "answers": { "Yes": "Complete all necessary checks for this criterion, with actual evidence supporting satisfaction of the requirement; equivalent implementations within the task's scope are allowed.", "No": "At least one necessary requirement is not satisfied; identify evidence of actual omissions, errors, or partial satisfaction." }, "compliance_groups": [ "G03" ], "benchmark_source": "vas" }, { "id": "IF11", "dimension": "Task Fulfillment / Compliance", "criterion": "Requirement: Deliver a PPTX with exactly 8 slides.\nVerification method: Inspect actual rendering, playback, or execution results as appropriate to the artifact format, and check the relevant native structure; do not judge solely from screenshots or filenames.\nPass criterion: Fully satisfy this requirement; omissions, errors, or partial satisfaction result in No. Evaluation-tool failures cannot be grounds for No.\nVerification boundary: Check against the locked task statement and the provided original materials; do not substitute the author's self-report for inspection.\nTask basis: prompt.md, original line 3: Using the materials in `input/`, create an **8-slide PPTX** in English for a research group meeting in the relevant field. Preserve product names, paper titles, and necessary technical terminology.", "answers": { "Yes": "Complete all necessary checks for this criterion, with actual evidence supporting satisfaction of the requirement; equivalent implementations within the task's scope are allowed.", "No": "At least one necessary requirement is not satisfied; identify evidence of actual omissions, errors, or partial satisfaction." }, "compliance_groups": [ "G04" ], "benchmark_source": "vas" }, { "id": "IF12", "dimension": "Task Fulfillment / Compliance", "criterion": "Requirement: Use English in the artifact.\nVerification method: Inspect actual rendering, playback, or execution results as appropriate to the artifact format, and check the relevant native structure; do not judge solely from screenshots or filenames.\nPass criterion: Fully satisfy this requirement; omissions, errors, or partial satisfaction result in No. Evaluation-tool failures cannot be grounds for No.\nVerification boundary: Check against the locked task statement and the provided original materials; do not substitute the author's self-report for inspection.\nTask basis: prompt.md, original line 3: Using the materials in `input/`, create an **8-slide PPTX** in English for a research group meeting in the relevant field. Preserve product names, paper titles, and necessary technical terminology.", "answers": { "Yes": "Complete all necessary checks for this criterion, with actual evidence supporting satisfaction of the requirement; equivalent implementations within the task's scope are allowed.", "No": "At least one necessary requirement is not satisfied; identify evidence of actual omissions, errors, or partial satisfaction." }, "compliance_groups": [ "G05" ], "benchmark_source": "vas" }, { "id": "IF13", "dimension": "Task Fulfillment / Compliance", "criterion": "Requirement: Do not add product, statistical, or experimental claims unsupported by the input sources.\nVerification method: Check the actual delivered content against the task statement and original materials; the author's self-report cannot replace evidence from the work.\nPass criterion: Fully satisfy this requirement; omissions, errors, or partial satisfaction result in No. Evaluation-tool failures cannot be grounds for No.\nVerification boundary: Check against the locked task statement and the provided original materials; do not substitute the author's self-report for inspection.\nTask basis: prompt.md, original line 5: See `input/reading-scope.md` for the reading scope. Use the provided facts and assets; do not add unsupported product, statistical, or experimental claims. If auxiliary TXT/CSV files conflict with the original PDF/webpage, the original takes precedence.", "answers": { "Yes": "Complete all necessary checks for this criterion, with actual evidence supporting satisfaction of the requirement; equivalent implementations within the task's scope are allowed.", "No": "At least one necessary requirement is not satisfied; identify evidence of actual omissions, errors, or partial satisfaction." }, "compliance_groups": [ "G06" ], "benchmark_source": "vas" }, { "id": "IF14", "dimension": "Task Fulfillment / Compliance", "criterion": "Requirement: Keep text editable.\nVerification method: Inspect actual rendering, playback, or execution results as appropriate to the artifact format, and check the relevant native structure; do not judge solely from screenshots or filenames.\nPass criterion: Fully satisfy this requirement; omissions, errors, or partial satisfaction result in No. Evaluation-tool failures cannot be grounds for No.\nVerification boundary: Check against the locked task statement and the provided original materials; do not substitute the author's self-report for inspection.\nTask basis: prompt.md, original line 17: - Preserve editable text, shapes, and data charts; a full-slide raster image must not be the sole content.", "answers": { "Yes": "Complete all necessary checks for this criterion, with actual evidence supporting satisfaction of the requirement; equivalent implementations within the task's scope are allowed.", "No": "At least one necessary requirement is not satisfied; identify evidence of actual omissions, errors, or partial satisfaction." }, "compliance_groups": [ "G07" ], "benchmark_source": "vas" }, { "id": "IF15", "dimension": "Task Fulfillment / Compliance", "criterion": "Requirement: Keep shapes and data charts editable.\nVerification method: Inspect actual rendering, playback, or execution results as appropriate to the artifact format, and check the relevant native structure; do not judge solely from screenshots or filenames.\nPass criterion: Fully satisfy this requirement; omissions, errors, or partial satisfaction result in No. Evaluation-tool failures cannot be grounds for No.\nVerification boundary: Check against the locked task statement and the provided original materials; do not substitute the author's self-report for inspection.\nTask basis: prompt.md, original line 17: - Preserve editable text, shapes, and data charts; a full-slide raster image must not be the sole content.", "answers": { "Yes": "Complete all necessary checks for this criterion, with actual evidence supporting satisfaction of the requirement; equivalent implementations within the task's scope are allowed.", "No": "At least one necessary requirement is not satisfied; identify evidence of actual omissions, errors, or partial satisfaction." }, "compliance_groups": [ "G07" ], "benchmark_source": "vas" }, { "id": "IF16", "dimension": "Task Fulfillment / Compliance", "criterion": "Requirement: Do not use a full-slide raster image as the sole content.\nVerification method: Inspect actual rendering, playback, or execution results as appropriate to the artifact format, and check the relevant native structure; do not judge solely from screenshots or filenames.\nPass criterion: Fully satisfy this requirement; omissions, errors, or partial satisfaction result in No. Evaluation-tool failures cannot be grounds for No.\nVerification boundary: Check against the locked task statement and the provided original materials; do not substitute the author's self-report for inspection.\nTask basis: prompt.md, original line 17: - Preserve editable text, shapes, and data charts; a full-slide raster image must not be the sole content.", "answers": { "Yes": "Complete all necessary checks for this criterion, with actual evidence supporting satisfaction of the requirement; equivalent implementations within the task's scope are allowed.", "No": "At least one necessary requirement is not satisfied; identify evidence of actual omissions, errors, or partial satisfaction." }, "compliance_groups": [ "G07" ], "benchmark_source": "vas" }, { "id": "IF17", "dimension": "Task Fulfillment / Compliance", "criterion": "Requirement: Cite factual sources in the artifact or sources.md.\nVerification method: Check the actual delivered content against the task statement and original materials; the author's self-report cannot replace evidence from the work.\nPass criterion: Fully satisfy this requirement; omissions, errors, or partial satisfaction result in No. Evaluation-tool failures cannot be grounds for No.\nVerification boundary: Check against the locked task statement and the provided original materials; do not substitute the author's self-report for inspection.\nTask basis: prompt.md, original line 19: - Cite facts and images in the artifact or `sources.md`; explain final file locations and any necessary opening instructions in `delivery.md`.", "answers": { "Yes": "Complete all necessary checks for this criterion, with actual evidence supporting satisfaction of the requirement; equivalent implementations within the task's scope are allowed.", "No": "At least one necessary requirement is not satisfied; identify evidence of actual omissions, errors, or partial satisfaction." }, "compliance_groups": [ "G08" ], "benchmark_source": "vas" }, { "id": "IF18", "dimension": "Task Fulfillment / Compliance", "criterion": "Requirement: Cite image sources in the artifact or sources.md.\nVerification method: Check the actual delivered content against the task statement and original materials; the author's self-report cannot replace evidence from the work.\nPass criterion: Fully satisfy this requirement; omissions, errors, or partial satisfaction result in No. Evaluation-tool failures cannot be grounds for No.\nVerification boundary: Check against the locked task statement and the provided original materials; do not substitute the author's self-report for inspection.\nTask basis: prompt.md, original line 19: - Cite facts and images in the artifact or `sources.md`; explain final file locations and any necessary opening instructions in `delivery.md`.", "answers": { "Yes": "Complete all necessary checks for this criterion, with actual evidence supporting satisfaction of the requirement; equivalent implementations within the task's scope are allowed.", "No": "At least one necessary requirement is not satisfied; identify evidence of actual omissions, errors, or partial satisfaction." }, "compliance_groups": [ "G08" ], "benchmark_source": "vas" }, { "id": "IF19", "dimension": "Task Fulfillment / Compliance", "criterion": "Requirement: Explain final file locations in delivery.\nVerification method: Inspect actual rendering, playback, or execution results as appropriate to the artifact format, and check the relevant native structure; do not judge solely from screenshots or filenames.\nPass criterion: Fully satisfy this requirement; omissions, errors, or partial satisfaction result in No. Evaluation-tool failures cannot be grounds for No.\nVerification boundary: Check against the locked task statement and the provided original materials; do not substitute the author's self-report for inspection.\nTask basis: prompt.md, original line 19: - Cite facts and images in the artifact or `sources.md`; explain final file locations and any necessary opening instructions in `delivery.md`.", "answers": { "Yes": "Complete all necessary checks for this criterion, with actual evidence supporting satisfaction of the requirement; equivalent implementations within the task's scope are allowed.", "No": "At least one necessary requirement is not satisfied; identify evidence of actual omissions, errors, or partial satisfaction." }, "compliance_groups": [ "G09" ], "benchmark_source": "vas" }, { "id": "IF20", "dimension": "Task Fulfillment / Compliance", "criterion": "Requirement: Explain any necessary opening procedures in delivery.\nVerification method: Inspect actual rendering, playback, or execution results as appropriate to the artifact format, and check the relevant native structure; do not judge solely from screenshots or filenames.\nPass criterion: Fully satisfy this requirement; omissions, errors, or partial satisfaction result in No. Evaluation-tool failures cannot be grounds for No.\nVerification boundary: Check against the locked task statement and the provided original materials; do not substitute the author's self-report for inspection.\nTask basis: prompt.md, original line 19: - Cite facts and images in the artifact or `sources.md`; explain final file locations and any necessary opening instructions in `delivery.md`.", "answers": { "Yes": "Complete all necessary checks for this criterion, with actual evidence supporting satisfaction of the requirement; equivalent implementations within the task's scope are allowed.", "No": "At least one necessary requirement is not satisfied; identify evidence of actual omissions, errors, or partial satisfaction." }, "compliance_groups": [ "G09" ], "benchmark_source": "vas" }, { "id": "IF21", "dimension": "Task Fulfillment / Compliance", "criterion": "Requirement: Deliver a corresponding PNG preview for each slide with the PPTX.\nVerification method: Inspect actual rendering, playback, or execution results as appropriate to the artifact format, and check the relevant native structure; do not judge solely from screenshots or filenames.\nPass criterion: Fully satisfy this requirement; omissions, errors, or partial satisfaction result in No. Evaluation-tool failures cannot be grounds for No.\nVerification boundary: Check against the locked task statement and the provided original materials; do not substitute the author's self-report for inspection.\nTask basis: prompt.md, original line 18: - Supply per-slide PNG previews with PPTX; supply compilable source, required assets, and PDF with TEX. PNG-only deliverables do not additionally require source files.", "answers": { "Yes": "Complete all necessary checks for this criterion, with actual evidence supporting satisfaction of the requirement; equivalent implementations within the task's scope are allowed.", "No": "At least one necessary requirement is not satisfied; identify evidence of actual omissions, errors, or partial satisfaction." }, "compliance_groups": [ "G10" ], "benchmark_source": "vas" }, { "id": "C1", "dimension": "Content", "criterion": "Conceptual precision required by researchers\nAssessment focus: Explain the residual branch, identity shortcut, and F(x)+x to researchers in the relevant field, neither conflating concepts nor elaborating excessively on basics.\nObservation scope: The 8-slide research group meeting PPTX and corresponding PNGs; researchers reading projected slides.\nObservation basis: Check the actual deliverables against the task statement and provided originals; focus on the residual branch and identity shortcut, the bases used for ImageNet/CIFAR experiments, and boundaries of extrapolation.\nScoring boundary: Whether terminology, concepts, and wording enable this task's audience to understand accurately; IF checks factual implementation, and C1 does not award points based on the number of terms.\nFull-refinement check (+2): Where the chosen method explanation is most detailed, shortcuts, residual branches and any stated dimensional qualification remain concise and unambiguous.\nExceptional check (+3): An economical conceptual explanation makes the distinction between learning a residual and producing the final mapping immediately graspable, resolving a likely expert-reader misconception more effectively than an otherwise flawless definition-and-diagram presentation.", "answers": { "-3": "The expression of the relationship between residual learning and the output is fundamentally wrong, making the method incomprehensible.", "-2": "For “Conceptual precision required by researchers”: Ambiguity in key concepts or terms repeatedly causes misunderstanding; some information remains discernible, but the target audience cannot reliably understand the main meaning.", "-1": "For “Conceptual precision required by researchers”: One substantive conceptual ambiguity or terminology explanation unsuitable for the audience requires readers to fill in important meaning themselves.", "0": "Meets basic requirements: The meanings of F(x), shortcut, and the output are accurate, and the necessary terminology is sufficient to understand the method. Satisfies the necessary quality for this criterion, with no substantive defects; performance is ordinary.", "1": "Accomplished but not yet fully refined: meets tier 0 and demonstrates a substantive, observable quality gain. The notation beside the residual block makes the distinction between F(x), x and their sum quicker to understand. A specific remaining opportunity or failure to meet the complete +2 quality conditions keeps the item at this tier; explain the actual gap in reason. Ordinary correctness alone is not excellent execution.", "2": "Fully refined within scope: meets tier 0 and achieves the following throughout every applicable part of this item, not just one successful local example. The formula, diagram labels and explanation reinforce the same distinction without requiring readers to reconstruct it across slides. Difficult-case check: Where the chosen method explanation is most detailed, shortcuts, residual branches and any stated dimensional qualification remain concise and unambiguous. Across the full applicable scope, there is no identifiable substantive weakness or feasible task-relevant improvement with material benefit under this criterion. Establish this through complete observation and an active search for plausible shortcomings; absence of a noticed error does not prove full refinement.", "3": "Exceptional beyond conventional full refinement: fully meets +2, then demonstrates a non-obvious design with a substantial additional benefit. An economical conceptual explanation makes the distinction between learning a residual and producing the final mapping immediately graspable, resolving a likely expert-reader misconception more effectively than an otherwise flawless definition-and-diagram presentation. In reason identify the choice, real difficulty, benefit beyond an equally correct/complete/polished conventional solution, and observed evidence. These benefit descriptions do not prescribe a unique design. Repeating +2 strengths, adding complexity or expressing amazement is insufficient; without the exceptional evidence, the ceiling is +2." }, "weight": 9, "benchmark_source": "vas" }, { "id": "C2", "dimension": "Content", "criterion": "Correspondence between experimental conditions and research claims\nAssessment focus: Evaluate how experimental comparisons support claims, making datasets, depths, metrics, and comparison bases explicit; a larger number of correct values alone does not earn a high score.\nObservation scope: The 8-slide research group meeting PPTX and corresponding PNGs; researchers reading projected slides.\nObservation basis: Check the actual deliverables against the task statement and provided originals; focus on the residual branch and identity shortcut, the bases used for ImageNet/CIFAR experiments, and boundaries of extrapolation.\nScoring boundary: The support relationship between task claims and evidence/data/actual processes; do not re-evaluate graphical encoding or how easy source documents are to locate.\nScenario reference · Content quality (no additional thresholds): Method definitions, comparison conditions, and limitations are nearby and concise, allowing researchers to judge accurately what evidence supports each conclusion; no additional papers or experiments are required.\nScoring boundary: C2 evaluates whether evidence in the artifact supports its claims; D3 evaluates whether required sources/records are traceable. Do not penalize both functions for the same “missing citation.”\nScoring boundary: Each root cause is counted only once, and independent functional impacts each require evidence; an IF failure does not automatically incur an additional quality penalty.\nApplicable boundaries for this task (apply to all relevant criteria; no additional scoring criteria): No new ablations, reproduction training, or citation of every value in the fact tables is required.\nFull-refinement check (+2): The most easily confused chosen results, such as different depths or evaluation settings, remain directly comparable without implying unsupported equivalence; more reported values earn no bonus.\nExceptional check (+3): The presentation exposes a meaningful relationship between the selected experimental comparisons that a conventional accurate table would leave readers to reconstruct, while preserving the non-equivalence of datasets and evaluation settings.", "answers": { "-3": "Main conclusions are supported by incompatible conditions or incorrect evidence, invalidating the research argument.", "-2": "For “Correspondence between experimental conditions and research claims”: Main claims are clearly disconnected from their bases, and key comparisons/explanations consistently cannot be supported by the existing materials or processes.", "-1": "For “Correspondence between experimental conditions and research claims”: One important claim lacks appropriately corresponding support, or the scope of evidence does not match the strength of the wording, requiring independent inference.", "0": "Meets basic requirements: ImageNet/CIFAR evidence supports the stated conclusions, and comparison conditions are sufficiently explicit. Satisfies the necessary quality for this criterion, with no substantive defects; performance is ordinary.", "1": "Accomplished but not yet fully refined: meets tier 0 and demonstrates a substantive, observable quality gain. One experimental comparison places its dataset, metric and comparison basis beside the claim, reducing source reconstruction. A specific remaining opportunity or failure to meet the complete +2 quality conditions keeps the item at this tier; explain the actual gap in reason. Ordinary correctness alone is not excellent execution.", "2": "Fully refined within scope: meets tier 0 and achieves the following throughout every applicable part of this item, not just one successful local example. The chosen ImageNet and CIFAR-10 comparisons connect conditions, observations and appropriately limited claims throughout the argument. Difficult-case check: The most easily confused chosen results, such as different depths or evaluation settings, remain directly comparable without implying unsupported equivalence; more reported values earn no bonus. Across the full applicable scope, there is no identifiable substantive weakness or feasible task-relevant improvement with material benefit under this criterion. Establish this through complete observation and an active search for plausible shortcomings; absence of a noticed error does not prove full refinement.", "3": "Exceptional beyond conventional full refinement: fully meets +2, then demonstrates a non-obvious design with a substantial additional benefit. The presentation exposes a meaningful relationship between the selected experimental comparisons that a conventional accurate table would leave readers to reconstruct, while preserving the non-equivalence of datasets and evaluation settings. In reason identify the choice, real difficulty, benefit beyond an equally correct/complete/polished conventional solution, and observed evidence. These benefit descriptions do not prescribe a unique design. Repeating +2 strengths, adding complexity or expressing amazement is insufficient; without the exceptional evidence, the ceiling is +2." }, "weight": 9, "benchmark_source": "vas" }, { "id": "C3", "dimension": "Content", "criterion": "Content selection and conclusion boundaries in an academic presentation\nAssessment focus: Emphasize the method, experiments, and limitations; clarify the distinction between degradation and overfitting and the boundaries of extrapolation, rather than using an accumulation of paper information as a substitute for a group meeting presentation.\nObservation scope: The 8-slide research group meeting PPTX and corresponding PNGs; researchers reading projected slides.\nObservation basis: Check the actual deliverables against the task statement and provided originals; focus on the residual branch and identity shortcut, the bases used for ImageNet/CIFAR experiments, and boundaries of extrapolation.\nScoring boundary: Content selection, density, tone, and necessary qualifications in relation to the scenario's purpose; not font size or visual priorities on the page.\nTask scenario: Academic projection / 8-slide PPTX.\nScenario reference · Audience (no additional thresholds): Researchers in the relevant field; projected presentation for group meeting discussion.\nScenario reference · Content (no additional thresholds): Precision in expressing the residual method, experimental conditions, and limitations; distinguish F(x) from F(x)+x, comparable experiments, and conclusion boundaries, reducing the burden of filling in meaning independently.\nScoring boundary: C3 considers what is selected, how deeply it is explained, and tone; O1 considers overall semantic relationships; V1 considers what the eye notices first.\nFull-refinement check (+2): The densest experiment discussion still preserves the distinction between training degradation, overfitting and the scope of the reported evidence without crowding out the main contribution.\nExceptional check (+3): The selection makes the boundary between evidence and extrapolation part of the central insight, achieving unusual explanatory economy without suppressing a needed limitation.", "answers": { "-3": "Degradation or extrapolation is presented as the opposite conclusion, or content selection distorts core research judgments.", "-2": "For “Content selection and conclusion boundaries in an academic presentation”: Content emphasis or tone clearly deviates from the task, necessary qualifications are persistently buried, and understanding/judgment is significantly impaired.", "-1": "For “Content selection and conclusion boundaries in an academic presentation”: Clear redundancy, omission of important qualifications, or a tone mismatched to the scenario increases the target audience's comprehension burden.", "0": "Meets basic requirements: Key points and necessary limitations are complete, with no unsupported generalization. Satisfies the necessary quality for this criterion, with no substantive defects; performance is ordinary.", "1": "Accomplished but not yet fully refined: meets tier 0 and demonstrates a substantive, observable quality gain. One explicit boundary helps distinguish the paper's supported conclusion from extrapolation. A specific remaining opportunity or failure to meet the complete +2 quality conditions keeps the item at this tier; explain the actual gap in reason. Ordinary correctness alone is not excellent execution.", "2": "Fully refined within scope: meets tier 0 and achieves the following throughout every applicable part of this item, not just one successful local example. Method, evidence and limitations receive purposeful emphasis, so the degradation discussion is not displaced by general background. Difficult-case check: The densest experiment discussion still preserves the distinction between training degradation, overfitting and the scope of the reported evidence without crowding out the main contribution. Across the full applicable scope, there is no identifiable substantive weakness or feasible task-relevant improvement with material benefit under this criterion. Establish this through complete observation and an active search for plausible shortcomings; absence of a noticed error does not prove full refinement.", "3": "Exceptional beyond conventional full refinement: fully meets +2, then demonstrates a non-obvious design with a substantial additional benefit. The selection makes the boundary between evidence and extrapolation part of the central insight, achieving unusual explanatory economy without suppressing a needed limitation. In reason identify the choice, real difficulty, benefit beyond an equally correct/complete/polished conventional solution, and observed evidence. These benefit descriptions do not prescribe a unique design. Repeating +2 strengths, adding complexity or expressing amazement is insufficient; without the exceptional evidence, the ceiling is +2." }, "weight": 7, "benchmark_source": "vas" }, { "id": "O1", "dimension": "Organization", "criterion": "Overall argument linking method, experiments, and limitations\nAssessment focus: The overarching logic explains why the method was proposed, what the evidence supports, and which issues remain unresolved; no fixed slide order is imposed.\nObservation scope: The 8-slide research group meeting PPTX and corresponding PNGs; researchers reading projected slides.\nObservation basis: Check the actual deliverables against the task statement and provided originals; focus on the residual branch and identity shortcut, the bases used for ImageNet/CIFAR experiments, and boundaries of extrapolation.\nScoring boundary: The logic of the overall argumentative/decision/operational/conceptual path, not compliance with a fixed structural template.\nScenario reference · Organizational quality (no additional thresholds): Method motivation, structure, experiments, and limitations progress closely, with no repetitive exposition within 8 slides, and conclusions naturally lead to evidence-based group meeting discussion.\nScoring boundary: C3 considers what is selected, how deeply it is explained, and tone; O1 considers overall semantic relationships; V1 considers what the eye notices first.\nFull-refinement check (+2): The transition from experimental gains to their limits remains reasoned and compact within eight slides; no extra discussion slide is needed.\nExceptional check (+3): A non-obvious argument structure makes the reason for the method, its experimental test and its limits illuminate one another, materially reducing the reasoning steps of a conventional method-results-limits account.", "answers": { "-3": "The sections cannot form a valid argument, and the core research question is disconnected from the conclusions.", "-2": "For “Overall argument linking method, experiments, and limitations”: The overall path has major breaks or an incorrect order, and the main explanations/judgments require substantial external supplementation to hold.", "-1": "For “Overall argument linking method, experiments, and limitations”: The main thread has one clear jump or instance of repetitive exposition, requiring the target audience to actively piece together key relationships.", "0": "Meets basic requirements: Method motivation, experiments, and limitations connect into an understandable argument. Satisfies the necessary quality for this criterion, with no substantive defects; performance is ordinary.", "1": "Accomplished but not yet fully refined: meets tier 0 and demonstrates a substantive, observable quality gain. A transition explains why a method or experiment answers the preceding research question. A specific remaining opportunity or failure to meet the complete +2 quality conditions keeps the item at this tier; explain the actual gap in reason. Ordinary correctness alone is not excellent execution.", "2": "Fully refined within scope: meets tier 0 and achieves the following throughout every applicable part of this item, not just one successful local example. The method, both experimental settings and limitations form an evidence-led argument rather than a sequence of independent summaries. Difficult-case check: The transition from experimental gains to their limits remains reasoned and compact within eight slides; no extra discussion slide is needed. Across the full applicable scope, there is no identifiable substantive weakness or feasible task-relevant improvement with material benefit under this criterion. Establish this through complete observation and an active search for plausible shortcomings; absence of a noticed error does not prove full refinement.", "3": "Exceptional beyond conventional full refinement: fully meets +2, then demonstrates a non-obvious design with a substantial additional benefit. A non-obvious argument structure makes the reason for the method, its experimental test and its limits illuminate one another, materially reducing the reasoning steps of a conventional method-results-limits account. In reason identify the choice, real difficulty, benefit beyond an equally correct/complete/polished conventional solution, and observed evidence. These benefit descriptions do not prescribe a unique design. Repeating +2 strengths, adding complexity or expressing amazement is insufficient; without the exceptional evidence, the ceiling is +2." }, "weight": 6, "benchmark_source": "vas" }, { "id": "O2", "dimension": "Organization", "criterion": "Local organization of claims and evidence within a slide\nAssessment focus: Evaluate the semantic relationships among conclusions, formulas/charts, and necessary explanations within a slide; visual prominence is assessed separately under V1.\nObservation scope: The 8-slide research group meeting PPTX and corresponding PNGs; researchers reading projected slides.\nObservation basis: Check the actual deliverables against the task statement and provided originals; focus on the residual branch and identity shortcut, the bases used for ImageNet/CIFAR experiments, and boundaries of extrapolation.\nScoring boundary: The semantic sequence and relationships of explanations within a page, stage, node, or comparison unit.\nFull-refinement check (+2): On the densest chosen evidence slide, exceptions and comparison bases remain attached to the correct claims without a reader having to infer their scope.\nExceptional check (+3): The local arrangement lets a reader test a claim against its condition and countervailing evidence in one coherent unit, beyond a polished but separate claim-chart-note arrangement.", "answers": { "-3": "Main slides contain only fragmented materials, preventing readers from establishing relationships between claims and evidence.", "-2": "For “Local organization of claims and evidence within a slide”: The order of objects, premises, explanations, or results within multiple key units is confused, significantly obstructing local understanding.", "-1": "For “Local organization of claims and evidence within a slide”: Relationships or the order of explanation within one key unit are unclear, requiring readers to consult other units to understand.", "0": "Meets basic requirements: The relationship between claims and supporting materials is apparent on each main slide. Satisfies the necessary quality for this criterion, with no substantive defects; performance is ordinary.", "1": "Accomplished but not yet fully refined: meets tier 0 and demonstrates a substantive, observable quality gain. A claim is locally paired with its supporting formula, chart or qualifying text, reducing semantic lookup. A specific remaining opportunity or failure to meet the complete +2 quality conditions keeps the item at this tier; explain the actual gap in reason. Ordinary correctness alone is not excellent execution.", "2": "Fully refined within scope: meets tier 0 and achieves the following throughout every applicable part of this item, not just one successful local example. Claims, comparisons and qualifications form coherent groups on the main evidence slides. Difficult-case check: On the densest chosen evidence slide, exceptions and comparison bases remain attached to the correct claims without a reader having to infer their scope. Across the full applicable scope, there is no identifiable substantive weakness or feasible task-relevant improvement with material benefit under this criterion. Establish this through complete observation and an active search for plausible shortcomings; absence of a noticed error does not prove full refinement.", "3": "Exceptional beyond conventional full refinement: fully meets +2, then demonstrates a non-obvious design with a substantial additional benefit. The local arrangement lets a reader test a claim against its condition and countervailing evidence in one coherent unit, beyond a polished but separate claim-chart-note arrangement. In reason identify the choice, real difficulty, benefit beyond an equally correct/complete/polished conventional solution, and observed evidence. These benefit descriptions do not prescribe a unique design. Repeating +2 strengths, adding complexity or expressing amazement is insufficient; without the exceptional evidence, the ceiling is +2." }, "weight": 5, "benchmark_source": "vas" }, { "id": "O3", "dimension": "Organization", "criterion": "Cross-slide conceptual continuity and closure of discussion\nAssessment focus: Examine how method terminology, experimental questions, and limitations carry across slides, and whether final conclusions return to existing evidence; no additional discussion slide is required.\nObservation scope: The 8-slide research group meeting PPTX and corresponding PNGs; researchers reading projected slides.\nObservation basis: Check the actual deliverables against the task statement and provided originals; focus on the residual branch and identity shortcut, the bases used for ImageNet/CIFAR experiments, and boundaries of extrapolation.\nScoring boundary: Semantic continuity of method terminology, experimental questions, and limitations across the 8 slides, and how the existing summary brings earlier content to a close; do not require an additional ending or discussion slide.\nScoring boundary: O3 evaluates cross-slide semantic continuity, V7 evaluates actual visible consistency, and V8 evaluates scanning and looking back during static projection; impacts sharing a root cause require proof of distinct impacts on the respective criterion.\nFull-refinement check (+2): Where ImageNet and CIFAR-10 evidence meet in the conclusion, their different scopes remain explicit and no later slide is needed to repair an earlier referent.\nExceptional check (+3): A purposeful conceptual carry-over turns the shift between ImageNet and CIFAR-10 into cumulative understanding rather than simple continuity, without implying the experiments are interchangeable.", "answers": { "-3": "Cross-slide semantic breaks or an ending that reverses earlier evidence prevent research understanding from forming.", "-2": "For “Cross-slide conceptual continuity and closure of discussion”: Multiple semantic breaks or contradictory referents across slides/stages or within a single-slide path prevent understanding from accumulating reliably.", "-1": "For “Cross-slide conceptual continuity and closure of discussion”: One necessary transition or path closure is ambiguous, requiring readers to reconstruct the relationship between preceding and subsequent objects/states.", "0": "Meets basic requirements: Cross-slide references are traceable, and the final summary remains connected to earlier content. Satisfies the necessary quality for this criterion, with no substantive defects; performance is ordinary.", "1": "Accomplished but not yet fully refined: meets tier 0 and demonstrates a substantive, observable quality gain. A slide transition explicitly carries forward a method term or question. A specific remaining opportunity or failure to meet the complete +2 quality conditions keeps the item at this tier; explain the actual gap in reason. Ordinary correctness alone is not excellent execution.", "2": "Fully refined within scope: meets tier 0 and achieves the following throughout every applicable part of this item, not just one successful local example. Terms, comparison objects and conclusions retain their meaning from method through experiments to limitations. Difficult-case check: Where ImageNet and CIFAR-10 evidence meet in the conclusion, their different scopes remain explicit and no later slide is needed to repair an earlier referent. Across the full applicable scope, there is no identifiable substantive weakness or feasible task-relevant improvement with material benefit under this criterion. Establish this through complete observation and an active search for plausible shortcomings; absence of a noticed error does not prove full refinement.", "3": "Exceptional beyond conventional full refinement: fully meets +2, then demonstrates a non-obvious design with a substantial additional benefit. A purposeful conceptual carry-over turns the shift between ImageNet and CIFAR-10 into cumulative understanding rather than simple continuity, without implying the experiments are interchangeable. In reason identify the choice, real difficulty, benefit beyond an equally correct/complete/polished conventional solution, and observed evidence. These benefit descriptions do not prescribe a unique design. Repeating +2 strengths, adding complexity or expressing amazement is insufficient; without the exceptional evidence, the ceiling is +2." }, "weight": 4, "benchmark_source": "vas" }, { "id": "V1", "dimension": "Visuals", "criterion": "Visual hierarchy of research claims and evidence\nAssessment focus: Evaluate the relative prominence of conclusions, method diagrams, experimental evidence, and supplementary materials across the 8 projected slides, establishing a restrained academic focus.\nObservation scope: The 8-slide research group meeting PPTX and corresponding PNGs; researchers reading projected slides.\nObservation basis: Check the actual deliverables against the task statement and provided originals; focus on the residual branch and identity shortcut, the bases used for ImageNet/CIFAR experiments, and boundaries of extrapolation.\nScoring boundary: Relative prominence of titles, main content, evidence, and supporting information; evaluate attention entry points, core focus, and strength of emphasis.\nScoring boundary: C3 considers what is selected, how deeply it is explained, and tone; O1 considers overall semantic relationships; V1 considers what the eye notices first.\nNegative-score scope: Judge the extent and severity of defects in this visual criterion; functional unusability is not a prerequisite. Retain the concrete defect examples and assess only permitted edits and newly introduced regressions where applicable.\nFull-refinement check (+2): On the densest comparison slide, the conclusion is quickly found while necessary conditions remain visibly subordinate but accessible.\nExceptional check (+3): Visual emphasis directs attention to the decisive scientific relationship rather than just the largest result, enabling a more accurate first-glance interpretation than a flawless conventional heading-and-chart hierarchy.", "answers": { "-3": "Neither the main research conclusions nor the evidence can be located.", "-2": "Emphasis hierarchies are reversed on multiple slides, with supplementary materials overpowering main claims.", "-1": "A group of slides has clear competition for emphasis, making the focus on main evidence unstable.", "0": "Meets basic requirements: The relative priority of claims and supporting materials is distinguishable on each slide. Satisfies the necessary quality for this criterion, with no substantive defects; performance is ordinary.", "1": "Accomplished but not yet fully refined: meets tier 0 and demonstrates a substantive, observable quality gain. One research claim is visually distinguished from its supporting detail without suppressing conditions. A specific remaining opportunity or failure to meet the complete +2 quality conditions keeps the item at this tier; explain the actual gap in reason. Ordinary correctness alone is not excellent execution.", "2": "Fully refined within scope: meets tier 0 and achieves the following throughout every applicable part of this item, not just one successful local example. Headings, method diagrams and experimental evidence establish complementary levels of emphasis across the deck. Difficult-case check: On the densest comparison slide, the conclusion is quickly found while necessary conditions remain visibly subordinate but accessible. Across the full applicable scope, there is no identifiable substantive weakness or feasible task-relevant improvement with material benefit under this criterion. Establish this through complete observation and an active search for plausible shortcomings; absence of a noticed error does not prove full refinement.", "3": "Exceptional beyond conventional full refinement: fully meets +2, then demonstrates a non-obvious design with a substantial additional benefit. Visual emphasis directs attention to the decisive scientific relationship rather than just the largest result, enabling a more accurate first-glance interpretation than a flawless conventional heading-and-chart hierarchy. In reason identify the choice, real difficulty, benefit beyond an equally correct/complete/polished conventional solution, and observed evidence. These benefit descriptions do not prescribe a unique design. Repeating +2 strengths, adding complexity or expressing amazement is insufficient; without the exceptional evidence, the ceiling is +2." }, "weight": 5, "benchmark_source": "vas" }, { "id": "V2", "dimension": "Visuals", "criterion": "Overall composition of research layouts\nAssessment focus: Evaluate area proportions, grids, edge alignment, visual balance, and whitespace among charts, explanatory columns, and method diagrams, organizing projected layouts suitable for group meeting discussion.\nObservation scope: The 8-slide research group meeting PPTX and corresponding PNGs; researchers reading projected slides.\nObservation basis: Check the actual deliverables against the task statement and provided originals; focus on the residual branch and identity shortcut, the bases used for ImageNet/CIFAR experiments, and boundaries of extrapolation.\nScoring boundary: Area proportions, grids, edges, visual balance, density, and whitespace among information blocks; evaluate overall composition and spatial organization.\nNegative-score scope: Judge the extent and severity of defects in this visual criterion; functional unusability is not a prerequisite. Retain the concrete defect examples and assess only permitted edits and newly introduced regressions where applicable.\nFull-refinement check (+2): The largest chosen table or method diagram fits comfortably with its annotations at projection scale; no forced common layout or extra whitespace quota is required.\nExceptional check (+3): A content-specific spatial solution accommodates both detailed experimental evidence and a clear overview with substantially less visual search than an equally legible conventional grid.", "answers": { "-3": "The main graphical and textual blocks cannot form a usable layout.", "-2": "Module proportions and grids are disordered on multiple slides, with severely unbalanced space for evidence.", "-1": "Repeatedly mismatched edges of graphical and textual blocks or imbalanced empty and dense areas make entire slides feel loose or crowded.", "0": "Meets basic requirements: The proportions and divisions of graphical and textual blocks are basically coordinated. Satisfies the necessary quality for this criterion, with no substantive defects; performance is ordinary.", "1": "Accomplished but not yet fully refined: meets tier 0 and demonstrates a substantive, observable quality gain. A diagram or table is allocated space according to its actual information load. A specific remaining opportunity or failure to meet the complete +2 quality conditions keeps the item at this tier; explain the actual gap in reason. Ordinary correctness alone is not excellent execution.", "2": "Fully refined within scope: meets tier 0 and achieves the following throughout every applicable part of this item, not just one successful local example. Chart area, explanatory text and whitespace form balanced layouts across sparse and dense slides. Difficult-case check: The largest chosen table or method diagram fits comfortably with its annotations at projection scale; no forced common layout or extra whitespace quota is required. Across the full applicable scope, there is no identifiable substantive weakness or feasible task-relevant improvement with material benefit under this criterion. Establish this through complete observation and an active search for plausible shortcomings; absence of a noticed error does not prove full refinement.", "3": "Exceptional beyond conventional full refinement: fully meets +2, then demonstrates a non-obvious design with a substantial additional benefit. A content-specific spatial solution accommodates both detailed experimental evidence and a clear overview with substantially less visual search than an equally legible conventional grid. In reason identify the choice, real difficulty, benefit beyond an equally correct/complete/polished conventional solution, and observed evidence. These benefit descriptions do not prescribe a unique design. Repeating +2 strengths, adding complexity or expressing amazement is insufficient; without the exceptional evidence, the ceiling is +2." }, "weight": 8, "benchmark_source": "vas" }, { "id": "V3", "dimension": "Visuals", "criterion": "Text presentation for academic projection\nAssessment focus: Evaluate the complete visibility of formulas, body text, conditions, legends, and source text in the actual visuals, along with font-size proportions, contrast, line spacing, and line breaks at projection scale.\nObservation scope: The 8-slide research group meeting PPTX and corresponding PNGs; researchers reading projected slides.\nObservation basis: Check the actual deliverables against the task statement and provided originals; focus on the residual branch and identity shortcut, the bases used for ImageNet/CIFAR experiments, and boundaries of extrapolation.\nScoring boundary: The completeness of text visibility, font sizes, contrast, line spacing, and line breaks in the actual delivered visuals; evaluate text presentation and reading at the target scale.\nNegative-score scope: Judge the extent and severity of defects in this visual criterion; functional unusability is not a prerequisite. Retain the concrete defect examples and assess only permitted edits and newly introduced regressions where applicable.\nFull-refinement check (+2): The longest chosen condition and smallest necessary chart labels remain comfortably readable alongside formulas, without shrinking all detail to fit.\nExceptional check (+3): Typographic treatment makes the structure of a difficult formula or qualification apparent as it is read, yielding a discernible comprehension benefit beyond impeccable readability and styling.", "answers": { "-3": "Core formulas and necessary body text are essentially illegible.", "-2": "Key conditions or legends on multiple slides are missing, obscured, or severely difficult to read.", "-1": "Necessary text is locally obscured or truncated, or font-size proportions and density persistently hinder reading.", "0": "Meets basic requirements: Body text, formulas, legends, necessary conditions, and source text are complete and discernible in the actual PPTX and corresponding PNGs; font sizes, contrast, and line spacing suit projected reading at a group meeting. There is no substantive obscuring, truncation, or difficulty reading; performance is ordinary.", "1": "Accomplished but not yet fully refined: meets tier 0 and demonstrates a substantive, observable quality gain. Semantic line breaks or typographic sizing make a formula or condition easier to read than ordinary legibility. A specific remaining opportunity or failure to meet the complete +2 quality conditions keeps the item at this tier; explain the actual gap in reason. Ordinary correctness alone is not excellent execution.", "2": "Fully refined within scope: meets tier 0 and achieves the following throughout every applicable part of this item, not just one successful local example. Formula notation, body copy, chart labels and source text use coordinated size and spacing at projection scale. Difficult-case check: The longest chosen condition and smallest necessary chart labels remain comfortably readable alongside formulas, without shrinking all detail to fit. Across the full applicable scope, there is no identifiable substantive weakness or feasible task-relevant improvement with material benefit under this criterion. Establish this through complete observation and an active search for plausible shortcomings; absence of a noticed error does not prove full refinement.", "3": "Exceptional beyond conventional full refinement: fully meets +2, then demonstrates a non-obvious design with a substantial additional benefit. Typographic treatment makes the structure of a difficult formula or qualification apparent as it is read, yielding a discernible comprehension benefit beyond impeccable readability and styling. In reason identify the choice, real difficulty, benefit beyond an equally correct/complete/polished conventional solution, and observed evidence. These benefit descriptions do not prescribe a unique design. Repeating +2 strengths, adding complexity or expressing amazement is insufficient; without the exceptional evidence, the ceiling is +2." }, "weight": 7, "benchmark_source": "vas" }, { "id": "V4", "dimension": "Visuals", "criterion": "Encoding and correspondence in research graphics\nAssessment focus: Evaluate the discernibility and semantic correspondence of graphical marks and arrows in the residual structure and of experimental charts, enabling reliable interpretation of methodological relationships and experimental comparisons.\nObservation scope: The 8-slide research group meeting PPTX and corresponding PNGs; researchers reading projected slides.\nObservation basis: Check the actual deliverables against the task statement and provided originals; focus on the residual branch and identity shortcut, the bases used for ImageNet/CIFAR experiments, and boundaries of extrapolation.\nScoring boundary: Recognizability and semantic correspondence of graphical marks, lines, arrows, symbols, and dynamic states; evaluate accurate decoding and reliable comparison.\nNegative-score scope: Judge the extent and severity of defects in this visual criterion; functional unusability is not a prerequisite. Retain the concrete defect examples and assess only permitted edits and newly introduced regressions where applicable.\nFull-refinement check (+2): The most complex chosen residual diagram or multi-condition comparison remains directly decodable without misleading scales or ambiguous connections.\nExceptional check (+3): A graphic encoding makes a difficult residual or experimental relationship directly visible instead of requiring mental reconstruction from correctly labelled parts; fidelity and comparison conditions remain intact.", "answers": { "-3": "The method's main path or core experimental encoding is reversed, invalidating the explanation.", "-2": "Main arrow or legend relationships are repeatedly incorrect, making comparisons unreliable.", "-1": "One key connection or comparison encoding has clear ambiguity.", "0": "Meets basic requirements: Structural relationships and experimental encoding are basically correct and usable. Satisfies the necessary quality for this criterion, with no substantive defects; performance is ordinary.", "1": "Accomplished but not yet fully refined: meets tier 0 and demonstrates a substantive, observable quality gain. A residual connection or experimental comparison uses a clear visual correspondence that reduces decoding effort. A specific remaining opportunity or failure to meet the complete +2 quality conditions keeps the item at this tier; explain the actual gap in reason. Ordinary correctness alone is not excellent execution.", "2": "Fully refined within scope: meets tier 0 and achieves the following throughout every applicable part of this item, not just one successful local example. Labels, connections and chart encodings reinforce the represented method and experimental relationships. Difficult-case check: The most complex chosen residual diagram or multi-condition comparison remains directly decodable without misleading scales or ambiguous connections. Across the full applicable scope, there is no identifiable substantive weakness or feasible task-relevant improvement with material benefit under this criterion. Establish this through complete observation and an active search for plausible shortcomings; absence of a noticed error does not prove full refinement.", "3": "Exceptional beyond conventional full refinement: fully meets +2, then demonstrates a non-obvious design with a substantial additional benefit. A graphic encoding makes a difficult residual or experimental relationship directly visible instead of requiring mental reconstruction from correctly labelled parts; fidelity and comparison conditions remain intact. In reason identify the choice, real difficulty, benefit beyond an equally correct/complete/polished conventional solution, and observed evidence. These benefit descriptions do not prescribe a unique design. Repeating +2 strengths, adding complexity or expressing amazement is insufficient; without the exceptional evidence, the ceiling is +2." }, "weight": 5, "benchmark_source": "vas" }, { "id": "V5", "dimension": "Visuals", "criterion": "Scenario-specific aesthetic suitability\nAssessment focus: For a projected group meeting presentation to researchers in the relevant field, evaluate an evidence-led, rigorous, and restrained research-discussion character; whether graphical abstraction, typography, panels, and imagery work together to serve academic communication. Coordination of colors alone, lightness and darkness, and saturation/area relationships are assigned to V9; this criterion requires evidence of scenario expression beyond color.\nObservation scope: The 8-slide research group meeting PPTX and corresponding PNGs; researchers reading projected slides.\nObservation basis: Check the actual deliverables against the task statement and provided originals; focus on the residual branch and identity shortcut, the bases used for ImageNet/CIFAR experiments, and boundaries of extrapolation.\nScoring boundary: V5 evaluates scenario character and the integration of non-color design language; V9 evaluates the aesthetics of color relationships themselves. Both require evidence for their respective quality objectives.\nScenario reference · Visual language (no additional thresholds): Rigorous and credible, with evidence first; structural diagrams, formulas, and experimental charts form a unified research narrative, restrained but not mechanically templated.\nScenario reference · Design constraints (no additional thresholds): Promotional-style grand slogans, decoration overpowering experimental conditions, or stacks of paper screenshots making evidence impossible to compare during the group meeting.\nScenario reference · Design objectives (no additional thresholds): Within 8 slides and at projection scale, typography, experimental charts, and structural diagrams jointly reflect the precision of research discussion; simple or more expressive styles are allowed as long as evidence remains dominant.\nNegative-score scope: Judge the extent and severity of defects in this visual criterion; functional unusability is not a prerequisite. Retain the concrete defect examples and assess only permitted edits and newly introduced regressions where applicable.\nFull-refinement check (+2): The densest scientific material integrates with simpler method slides without decorative competition or a change to promotional rhetoric.\nExceptional check (+3): A distinctive, restrained research visual language expresses the logic of residual reasoning or evidence comparison through its graphic form, beyond polished academic styling and without decorative distraction.", "answers": { "-3": "The visual mode of expression conflicts with this task's audience or core purpose, defeating the main scenario purpose.", "-2": "Graphical abstraction, typographic tone, imagery, or panel language is systematically unsuitable for the audience, severely harming credibility, approachability, or task expression.", "-1": "Multiple instances of stylistic patchwork, expressive tone, or graphical details do not fit this task's scenario, clearly reducing the overall sense of finish; do not deduct points directly for being “not modern enough.”", "0": "Meets basic requirements: The overall expression suits the audience, purpose, and medium and is basically coordinated, but the integration of scenario character is ordinary. Satisfies the necessary quality for this criterion, with no substantive defects; performance is ordinary.", "1": "Accomplished but not yet fully refined: meets tier 0 and demonstrates a substantive, observable quality gain. A non-color choice in typography or diagram abstraction supports evidence-led research discussion. A specific remaining opportunity or failure to meet the complete +2 quality conditions keeps the item at this tier; explain the actual gap in reason. Ordinary correctness alone is not excellent execution.", "2": "Fully refined within scope: meets tier 0 and achieves the following throughout every applicable part of this item, not just one successful local example. Diagrams, tables and typography share a restrained visual language that keeps research evidence central. Difficult-case check: The densest scientific material integrates with simpler method slides without decorative competition or a change to promotional rhetoric. Across the full applicable scope, there is no identifiable substantive weakness or feasible task-relevant improvement with material benefit under this criterion. Establish this through complete observation and an active search for plausible shortcomings; absence of a noticed error does not prove full refinement.", "3": "Exceptional beyond conventional full refinement: fully meets +2, then demonstrates a non-obvious design with a substantial additional benefit. A distinctive, restrained research visual language expresses the logic of residual reasoning or evidence comparison through its graphic form, beyond polished academic styling and without decorative distraction. In reason identify the choice, real difficulty, benefit beyond an equally correct/complete/polished conventional solution, and observed evidence. These benefit descriptions do not prescribe a unique design. Repeating +2 strengths, adding complexity or expressing amazement is insufficient; without the exceptional evidence, the ceiling is +2." }, "weight": 7, "benchmark_source": "vas" }, { "id": "V6", "dimension": "Visuals", "criterion": "Craftsmanship of research graphics and components\nAssessment focus: Evaluate the geometric precision and optical alignment of structural diagram endpoints, line widths, chart boundaries, image proportions, local label components, and numerical markers.\nObservation scope: The 8-slide research group meeting PPTX and corresponding PNGs; researchers reading projected slides.\nObservation basis: Check the actual deliverables against the task statement and provided originals; focus on the residual branch and identity shortcut, the bases used for ImageNet/CIFAR experiments, and boundaries of extrapolation.\nScoring boundary: Graphic contours, line widths, endpoints, image proportions and cropping, and geometric and optical alignment within components; evaluate precision of local construction.\nNegative-score scope: Judge the extent and severity of defects in this visual criterion; functional unusability is not a prerequisite. Retain the concrete defect examples and assess only permitted edits and newly introduced regressions where applicable.\nFull-refinement check (+2): Crowded junctions, small labels and chart boundaries retain the same finish as simple components, without requiring a particular drawing style.\nExceptional check (+3): Precision in the most demanding scientific drawing resolves a genuine ambiguity of correspondence at crowded connections or annotations, beyond uniformly clean edges and alignment.", "answers": { "-3": "The construction of key research graphics is broken.", "-2": "Main connection endpoints, chart contours, or image treatments exhibit severe distortion in multiple places.", "-1": "Repeated misalignment, rough lines, or imprecise local components reduce the level of finish.", "0": "Meets basic requirements: Endpoints, contours, proportions, cropping, and local alignment of necessary graphics and components are correct and usable. Satisfies the necessary quality for this criterion, with no substantive defects; performance is ordinary.", "1": "Accomplished but not yet fully refined: meets tier 0 and demonstrates a substantive, observable quality gain. A set of diagram endpoints or numerical labels is precisely aligned and visibly easier to inspect. A specific remaining opportunity or failure to meet the complete +2 quality conditions keeps the item at this tier; explain the actual gap in reason. Ordinary correctness alone is not excellent execution.", "2": "Fully refined within scope: meets tier 0 and achieves the following throughout every applicable part of this item, not just one successful local example. Connections, line weights and local labels remain carefully finished across the method and experimental figures. Difficult-case check: Crowded junctions, small labels and chart boundaries retain the same finish as simple components, without requiring a particular drawing style. Across the full applicable scope, there is no identifiable substantive weakness or feasible task-relevant improvement with material benefit under this criterion. Establish this through complete observation and an active search for plausible shortcomings; absence of a noticed error does not prove full refinement.", "3": "Exceptional beyond conventional full refinement: fully meets +2, then demonstrates a non-obvious design with a substantial additional benefit. Precision in the most demanding scientific drawing resolves a genuine ambiguity of correspondence at crowded connections or annotations, beyond uniformly clean edges and alignment. In reason identify the choice, real difficulty, benefit beyond an equally correct/complete/polished conventional solution, and observed evidence. These benefit descriptions do not prescribe a unique design. Repeating +2 strengths, adding complexity or expressing amazement is insufficient; without the exceptional evidence, the ceiling is +2." }, "weight": 6, "benchmark_source": "vas" }, { "id": "V7", "dimension": "Visuals", "criterion": "Visible consistency across slides and delivery formats\nAssessment focus: Evaluate the stability of identity and presentation of the same research objects, symbols, and recurring styles across the 8 slides, the native PPTX, and delivered previews.\nObservation scope: The 8-slide research group meeting PPTX and corresponding PNGs; researchers reading projected slides.\nObservation basis: Check the actual deliverables against the task statement and provided originals; focus on the residual branch and identity shortcut, the bases used for ImageNet/CIFAR experiments, and boundaries of extrapolation.\nScoring boundary: Assess negatively only differences that affect identity, content correspondence, layout hierarchy, or use; do not deduct points for harmless differences in antialiasing, sampling, or font edges. If line wrapping causes obscuring, loss, or misaligned correspondences, assess negatively according to its actual extent; consistency in other areas cannot offset this defect. Scoring across criteria requires separate proof of independent impacts.\nScoring boundary: O3 evaluates cross-slide semantic continuity, V7 evaluates actual visible consistency, and V8 evaluates scanning and looking back during static projection; impacts sharing a root cause require proof of distinct impacts on the respective criterion.\nNegative-score scope: Judge the extent and severity of defects in this visual criterion; functional unusability is not a prerequisite. Retain the concrete defect examples and assess only permitted edits and newly introduced regressions where applicable.\nFull-refinement check (+2): The most detailed diagram and table preserve that identity across slides and exports, without font substitution, misalignment or style drift at the relevant objects.\nExceptional check (+3): A carefully chosen recurring visual identity allows a research object to be recognized across materially different diagram and experiment contexts with substantially less re-identification effort than ordinary consistent styling.", "answers": { "-3": "Main research objects cannot be matched across slides or formats.", "-2": "Object identity or necessary content undergoes major drift in multiple places.", "-1": "One type of recurring symbol or corresponding slide exhibits clear drift.", "0": "Meets basic requirements: Necessary objects, symbols, and content correspond correctly across slides and between the PPTX and the author's PNGs, with no drift affecting identity, content, layout hierarchy, or use; harmless antialiasing or font-edge differences do not incur deductions, and the overall manner of correspondence is ordinary.", "1": "Accomplished but not yet fully refined: meets tier 0 and demonstrates a substantive, observable quality gain. A recurring symbol or comparison object has a deliberately stable visible treatment across slides. A specific remaining opportunity or failure to meet the complete +2 quality conditions keeps the item at this tier; explain the actual gap in reason. Ordinary correctness alone is not excellent execution.", "2": "Fully refined within scope: meets tier 0 and achieves the following throughout every applicable part of this item, not just one successful local example. Recurring notation, chart styling and page components form a coherent visual identity in PPTX and previews. Difficult-case check: The most detailed diagram and table preserve that identity across slides and exports, without font substitution, misalignment or style drift at the relevant objects. Across the full applicable scope, there is no identifiable substantive weakness or feasible task-relevant improvement with material benefit under this criterion. Establish this through complete observation and an active search for plausible shortcomings; absence of a noticed error does not prove full refinement.", "3": "Exceptional beyond conventional full refinement: fully meets +2, then demonstrates a non-obvious design with a substantial additional benefit. A carefully chosen recurring visual identity allows a research object to be recognized across materially different diagram and experiment contexts with substantially less re-identification effort than ordinary consistent styling. In reason identify the choice, real difficulty, benefit beyond an equally correct/complete/polished conventional solution, and observed evidence. These benefit descriptions do not prescribe a unique design. Repeating +2 strengths, adding complexity or expressing amazement is insufficient; without the exceptional evidence, the ceiling is +2." }, "weight": 3, "benchmark_source": "vas" }, { "id": "V8", "dimension": "Visuals", "criterion": "Scanning rhythm for group meeting projection\nAssessment focus: Evaluate readers' eye movement among claims, charts, and condition areas, their entry into new slides, and the burden of looking back, supporting live research discussion.\nObservation scope: The 8-slide research group meeting PPTX and corresponding PNGs; researchers reading projected slides.\nObservation basis: Check the actual deliverables against the task statement and provided originals; focus on the residual branch and identity shortcut, the bases used for ImageNet/CIFAR experiments, and boundaries of extrapolation.\nScoring boundary: Evaluate eye movement from claims to charts, conditions, and explanations in static projection, entry into new slides, and the burden of looking back; animation, autoplay, or fixed dwell times are not required.\nScoring boundary: O3 evaluates cross-slide semantic continuity, V7 evaluates actual visible consistency, and V8 evaluates scanning and looking back during static projection; impacts sharing a root cause require proof of distinct impacts on the respective criterion.\nNegative-score scope: Judge the extent and severity of defects in this visual criterion; functional unusability is not a prerequisite. Retain the concrete defect examples and assess only permitted edits and newly introduced regressions where applicable.\nFull-refinement check (+2): The densest comparison can be scanned and then examined without repeatedly searching for its legend or condition; simple slides remain appropriately concise.\nExceptional check (+3): The viewing route allows both rapid research orientation and close evidential scrutiny with unusually little backtracking, through a content-specific scanning solution rather than mere cleanliness.", "answers": { "-3": "No scanning route can be established on the main pages.", "-2": "The visual path is repeatedly interrupted across multiple pages, and the main evidence cannot be traced smoothly.", "-1": "There is conspicuous, repeated, unproductive backtracking between evidence and explanation.", "0": "Meets basic requirements: On every necessary page, there is a followable scanning route among claims, evidence, and conditions, with no substantive obstacles to entering a new page or looking back; performance is ordinary.", "1": "Accomplished but not yet fully refined: meets tier 0 and demonstrates a substantive, observable quality gain. A claim-to-evidence viewing route reduces unnecessary backtracking on one slide. A specific remaining opportunity or failure to meet the complete +2 quality conditions keeps the item at this tier; explain the actual gap in reason. Ordinary correctness alone is not excellent execution.", "2": "Fully refined within scope: meets tier 0 and achieves the following throughout every applicable part of this item, not just one successful local example. Slide entry points, evidence blocks and qualifying notes support a steady research-presentation reading rhythm. Difficult-case check: The densest comparison can be scanned and then examined without repeatedly searching for its legend or condition; simple slides remain appropriately concise. Across the full applicable scope, there is no identifiable substantive weakness or feasible task-relevant improvement with material benefit under this criterion. Establish this through complete observation and an active search for plausible shortcomings; absence of a noticed error does not prove full refinement.", "3": "Exceptional beyond conventional full refinement: fully meets +2, then demonstrates a non-obvious design with a substantial additional benefit. The viewing route allows both rapid research orientation and close evidential scrutiny with unusually little backtracking, through a content-specific scanning solution rather than mere cleanliness. In reason identify the choice, real difficulty, benefit beyond an equally correct/complete/polished conventional solution, and observed evidence. These benefit descriptions do not prescribe a unique design. Repeating +2 strengths, adding complexity or expressing amazement is insufficient; without the exceptional evidence, the ceiling is +2." }, "weight": 3, "benchmark_source": "vas" }, { "id": "V9", "dimension": "Visuals", "criterion": "Color-scheme aesthetics and color relationships\nObject of assessment: Independently assess the aesthetic maturity of the color scheme for a research-group meeting presentation projected for researchers in the relevant field. Assess hue and light–dark relationships among backgrounds, headings/body text, experimental charts, method diagrams, and panels; focus on the area occupied by accent colors on each page and whether illustrations/native graphics are integrated.\nObservation scope: An 8-page research-group meeting PPTX and corresponding PNGs; projected reading by researchers.\nObservation basis: Check the actual deliverables against the task specification and supplied originals; the content focus is residual branches and identity shortcuts, ImageNet/CIFAR experimental definitions and scope, and extrapolation boundaries.\nScoring boundary: V9 independently assesses the aesthetic harmony and refinement of the color scheme; impaired reading caused by text contrast belongs to V3, mismatches between colors and object meanings belong to V4, and drift in color-code identity across pages belongs to V7. Attribute an issue across criteria only when distinct impacts on the respective criterion are demonstrated. V5 must not merely repeat V9's assessment of color-scheme quality, and V9 must not directly award a professional tier merely because the scheme is “appropriate to the setting.”\nApplicability limits: No fixed palette, modern theme, or greater number of colors is required; both modern and traditional approaches must demonstrate maturity through actual color relationships.; For editing tasks, assess only permitted areas and newly introduced regressions; do not automatically lower the tier for retaining the original palette, making no color changes, or performing local repairs.; When color values have not been read, use observable color descriptions; do not invent exact HEX values or treat a list of color names as analysis.\nObservation elements: Hue compatibility and warm–cool relationships; lightness hierarchy and color center of gravity; saturation × area proportions; the number, weight, and restraint of accent colors; integration of images, graphics, panels, and background colors\nFull-refinement check (+2): The densest experimental slide remains harmonious and restrained without competing accents or jarring inserted figures; evaluate color relationships rather than palette fashion.\nExceptional check (+3): A subtle relationship of hue, lightness and area gives dense figures and sparse explanation equally coherent color balance while retaining a distinctive restrained character; name the perceptual gain beyond a flawless conventional palette.", "answers": { "-3": "The relationships among primary colors, background, and accent colors are fundamentally out of control, and the color organization of core objects does not hold together; identify the actual relationships rather than judging failure by personal preference.", "-2": "Large areas of clashing hues or saturation occur in multiple places, with disordered light–dark relationships and color centers of gravity, causing systematically harsh, muddy, cluttered, or clearly unbalanced effects. This constitutes a severe color-scheme defect even if the content remains readable.", "-1": "There is clear, persistent imbalance in color compatibility or area: backgrounds and images are incompatible, accent colors are excessive, local areas are jarring, or adjacent colors clash; explain the color relationships and their locations.", "0": "Meets basic requirements: Colors in the required areas are basically harmonious, and the relationships among background, main subjects, and accent colors hold together, with no clear conflicts, but color integration is ordinary; correct semantic colors or common combinations alone demonstrate only a basic level. The necessary quality for this criterion is met, with no substantive defects; performance is ordinary.", "1": "Accomplished but not yet fully refined: meets tier 0 and demonstrates a substantive, observable quality gain. A specific hue, lightness or accent-area relationship improves the harmony of a research figure and its background. A specific remaining opportunity or failure to meet the complete +2 quality conditions keeps the item at this tier; explain the actual gap in reason. Ordinary correctness alone is not excellent execution.", "2": "Fully refined within scope: meets tier 0 and achieves the following throughout every applicable part of this item, not just one successful local example. Backgrounds, method diagrams and experimental charts share a controlled color balance suited to changing content density. Difficult-case check: The densest experimental slide remains harmonious and restrained without competing accents or jarring inserted figures; evaluate color relationships rather than palette fashion. Across the full applicable scope, there is no identifiable substantive weakness or feasible task-relevant improvement with material benefit under this criterion. Establish this through complete observation and an active search for plausible shortcomings; absence of a noticed error does not prove full refinement.", "3": "Exceptional beyond conventional full refinement: fully meets +2, then demonstrates a non-obvious design with a substantial additional benefit. A subtle relationship of hue, lightness and area gives dense figures and sparse explanation equally coherent color balance while retaining a distinctive restrained character; name the perceptual gain beyond a flawless conventional palette. In reason identify the choice, real difficulty, benefit beyond an equally correct/complete/polished conventional solution, and observed evidence. These benefit descriptions do not prescribe a unique design. Repeating +2 strengths, adding complexity or expressing amazement is insufficient; without the exceptional evidence, the ceiling is +2." }, "weight": 6, "benchmark_source": "vas" }, { "id": "D1", "dimension": "Delivery", "criterion": "Direct usability of research-group meeting files\nObject of assessment: Check only whether the required PPTX and per-page previews actually open, their page order, and their usable state; assess visual details separately under visual criteria.\nObservation scope: An 8-page research-group meeting PPTX and corresponding PNGs; projected reading by researchers.\nObservation basis: Check the actual deliverables against the task specification and supplied originals; the content focus is residual branches and identity shortcuts, ImageNet/CIFAR experimental definitions and scope, and extrapolation boundaries.\nScoring boundary: Opening, locating, and directly using the required files; V7 assesses consistency of actually visible content, not file existence as a substitute for quality.\nScenario reference · Delivery quality (no additional thresholds): Native objects and previews can be used directly for the research-group meeting; existing sources are clearly locatable, opening instructions are concise, and editing boundaries for charts and formulas are explicit, without requiring additional generation scripts.\nScoring boundary: Check direct usability, editability and maintainability as required, and verifiability of the required supporting basis separately; do not mechanically deduct under all three criteria for one missing document.\nFull-refinement check (+2): Even when several companion files exist, the final eight-slide deck and each corresponding preview are unambiguous without trial-and-error opening.\nExceptional check (+3): Within the required deck and previews, an economical handoff organization removes a real ambiguity or repeated selection step that persists in a conventionally fully refined file set; extra documentation or software is not a substitute.", "answers": { "-3": "There is no primary PPTX deliverable that can be opened and used.", "-2": "For “Direct usability of research-group meeting files”: The main file is partially usable, but necessary opening/use or identification of the final version is severely obstructed, requiring substantial manual remediation.", "-1": "For “Direct usability of research-group meeting files”: There are clear defects in the required files/versions/entry points, and the recipient must perform additional searching or remediation before normal use.", "0": "Meets basic requirements: The final PPTX and 8 corresponding PNG previews are complete and open normally, the page count and order are correct, and file roles can be confirmed; they can be used as specified in the task, with no substantive defects, and delivery organization is ordinary.", "1": "Accomplished but not yet fully refined: meets tier 0 and demonstrates a substantive, observable quality gain. The required deck and previews have a directly identifiable relationship that reduces file-selection effort. A specific remaining opportunity or failure to meet the complete +2 quality conditions keeps the item at this tier; explain the actual gap in reason. Ordinary correctness alone is not excellent execution.", "2": "Fully refined within scope: meets tier 0 and achieves the following throughout every applicable part of this item, not just one successful local example. Entry file, preview order and supplied opening guidance work together for immediate group-meeting use. Difficult-case check: Even when several companion files exist, the final eight-slide deck and each corresponding preview are unambiguous without trial-and-error opening. Across the full applicable scope, there is no identifiable substantive weakness or feasible task-relevant improvement with material benefit under this criterion. Establish this through complete observation and an active search for plausible shortcomings; absence of a noticed error does not prove full refinement.", "3": "Exceptional beyond conventional full refinement: fully meets +2, then demonstrates a non-obvious design with a substantial additional benefit. Within the required deck and previews, an economical handoff organization removes a real ambiguity or repeated selection step that persists in a conventionally fully refined file set; extra documentation or software is not a substitute. In reason identify the choice, real difficulty, benefit beyond an equally correct/complete/polished conventional solution, and observed evidence. These benefit descriptions do not prescribe a unique design. Repeating +2 strengths, adding complexity or expressing amazement is insufficient; without the exceptional evidence, the ceiling is +2." }, "weight": 4, "benchmark_source": "vas" }, { "id": "D2", "dimension": "Delivery", "criterion": "Editing structure of scientific-research objects\nObject of assessment: Assess the native editing organization of required text, shapes, and charts, and the necessary resources; generation scripts are not required.\nObservation scope: An 8-page research-group meeting PPTX and corresponding PNGs; projected reading by researchers.\nObservation basis: Check the actual deliverables against the task specification and supplied originals; the content focus is residual branches and identity shortcuts, ImageNet/CIFAR experimental definitions and scope, and extrapolation boundaries.\nScoring boundary: The editable structure, resources, and material organization of objects required by the task specification; engineering scripts or additional tests are not required.\nScoring boundary: Check direct usability, editability and maintainability as required, and verifiability of the required supporting basis separately; do not mechanically deduct under all three criteria for one missing document.\nTask-specific applicability boundary (applies to all relevant criteria; adds no scoring criteria): No fixed template, brand colors, or use of all paper figures is required; necessary readable figures from the original paper may be incorporated as raster images.\nFull-refinement check (+2): The densest native diagram or chart has clear object relationships and editable content without hidden raster substitutes; infer only what native structure supports, not an unperformed edit test.\nExceptional check (+3): Native object organization mirrors a meaningful scientific relationship so that locating the corresponding text, shape or chart component requires substantially less reconstruction than in a flawlessly editable but conventionally grouped deck.", "answers": { "-3": "Full-page rasterization or resource corruption makes core content impossible to edit as required.", "-2": "For “Editing structure of scientific-research objects”: The structure of major required objects or resources is severely damaged, so local modifications require extensive reconstruction or supplementation of necessary materials.", "-1": "For “Editing structure of scientific-research objects”: The editing organization of one important type of object/resource is clearly inadequate, requiring additional manual organization for maintenance.", "0": "Meets basic requirements: Necessary objects are editable, and resources are sufficient to display the content. The necessary quality for this criterion is met, with no substantive defects; performance is ordinary.", "1": "Accomplished but not yet fully refined: meets tier 0 and demonstrates a substantive, observable quality gain. A meaningful grouping or editable chart relationship makes one scientific object easier to locate and maintain. A specific remaining opportunity or failure to meet the complete +2 quality conditions keeps the item at this tier; explain the actual gap in reason. Ordinary correctness alone is not excellent execution.", "2": "Fully refined within scope: meets tier 0 and achieves the following throughout every applicable part of this item, not just one successful local example. Text, diagram shapes and chart data are organized coherently rather than merely being individually editable. Difficult-case check: The densest native diagram or chart has clear object relationships and editable content without hidden raster substitutes; infer only what native structure supports, not an unperformed edit test. Across the full applicable scope, there is no identifiable substantive weakness or feasible task-relevant improvement with material benefit under this criterion. Establish this through complete observation and an active search for plausible shortcomings; absence of a noticed error does not prove full refinement.", "3": "Exceptional beyond conventional full refinement: fully meets +2, then demonstrates a non-obvious design with a substantial additional benefit. Native object organization mirrors a meaningful scientific relationship so that locating the corresponding text, shape or chart component requires substantially less reconstruction than in a flawlessly editable but conventionally grouped deck. In reason identify the choice, real difficulty, benefit beyond an equally correct/complete/polished conventional solution, and observed evidence. These benefit descriptions do not prescribe a unique design. Repeating +2 strengths, adding complexity or expressing amazement is insufficient; without the exceptional evidence, the ceiling is +2." }, "weight": 3, "benchmark_source": "vas" }, { "id": "D3", "dimension": "Delivery", "criterion": "Sources of research evidence and handoff instructions\nObject of assessment: Assess whether fact/image sources and delivery within the existing requirements are easy to trace and open, without adding requirements for scripts or experimental reproduction.\nObservation scope: An 8-page research-group meeting PPTX and corresponding PNGs; projected reading by researchers.\nObservation basis: Check the actual deliverables against the task specification and supplied originals; the content focus is residual branches and identity shortcuts, ImageNet/CIFAR experimental definitions and scope, and extrapolation boundaries.\nScoring boundary: The verifiability of required sources and calculation/change/capture or usage instructions; select according to the actual requirements of this task, without insisting on all document types. This criterion assesses verification paths; it does not replace C2 in judging whether claims are supported, or D1 in judging whether files can be opened.\nScoring boundary: C2 assesses whether evidence in the deliverable supports claims; D3 assesses whether required sources/records can be traced. Do not deduct twice for the same “missing citation” under these two functions.\nScoring boundary: Check direct usability, editability and maintainability as required, and verifiability of the required supporting basis separately; do not mechanically deduct under all three criteria for one missing document.\nFull-refinement check (+2): The most easily confused dataset or metric can be traced to its actual source passage or table without guessing which result was used; no additional experiment or script is required.\nExceptional check (+3): A compact source-to-claim organization makes a difficult comparison basis directly auditable across the chosen results, reducing reconstruction beyond a complete, accurate citation list.", "answers": { "-3": "Severe failure: The main research evidence has no verifiable sources, or the instructions make the deliverable impossible to locate/open reliably.", "-2": "Clearly unacceptable: For “Sources of research evidence and handoff instructions”: Major sources or calculation/change/capture/usage records required by the task specification are extensively missing or contradictory, substantially impeding verification.", "-1": "Local defect: For “Sources of research evidence and handoff instructions”: One type of key provenance information or necessary instruction is not clearly locatable, requiring additional searching to confirm its basis or usage.", "0": "Meets basic requirements: Fact/image sources and necessary instructions required by the task specification are accurate, and the actual basis can be located and verified; they may be provided within figures or at document locations specified by the task.", "1": "Accomplished but not yet fully refined: meets tier 0 and demonstrates a substantive, observable quality gain. A claim or reused figure has a source locator that reduces the effort to check it. A specific remaining opportunity or failure to meet the complete +2 quality conditions keeps the item at this tier; explain the actual gap in reason. Ordinary correctness alone is not excellent execution.", "2": "Fully refined within scope: meets tier 0 and achieves the following throughout every applicable part of this item, not just one successful local example. Fact references, figure credits and existing delivery guidance map consistently to the chosen slides and files. Difficult-case check: The most easily confused dataset or metric can be traced to its actual source passage or table without guessing which result was used; no additional experiment or script is required. Across the full applicable scope, there is no identifiable substantive weakness or feasible task-relevant improvement with material benefit under this criterion. Establish this through complete observation and an active search for plausible shortcomings; absence of a noticed error does not prove full refinement.", "3": "Exceptional beyond conventional full refinement: fully meets +2, then demonstrates a non-obvious design with a substantial additional benefit. A compact source-to-claim organization makes a difficult comparison basis directly auditable across the chosen results, reducing reconstruction beyond a complete, accurate citation list. In reason identify the choice, real difficulty, benefit beyond an equally correct/complete/polished conventional solution, and observed evidence. These benefit descriptions do not prescribe a unique design. Repeating +2 strengths, adding complexity or expressing amazement is insufficient; without the exceptional evidence, the ceiling is +2." }, "weight": 3, "benchmark_source": "vas" } ]