{"id":"34ff6a8a-ffab-40b7-9932-b651a064087a","arxiv_id":"2606.00837","paper_version":1,"verdict":"UNVERDICTED","confidence":"LOW","novelty_score":7.0,"correctness_risk":"unknown","formal_verification":"none","parameter_count":0,"one_line_summary":"CoFi is a two-stage coarse-to-fine sampler that enforces global coherence via scaffold alignment before restoring local structure with a pretrained prior, yielding better quality and 2-8x fewer evaluations across planning and generation tasks.","lead":"The paper introduces CoFi, an inference-time sampler for compositional diffusion that first builds a global scaffold from local plans then refines details at intermediate noise. This approach targets long-horizon robotic planning and generation tasks where standard composition struggles with global structure.","discovery_kind":"new_method","skeptic_critique":{"model":"grok-4.3","headline":"Existence and robustness of an intermediate noise level that preserves the imposed scaffold while restoring local details is unverified","rationale":"The reader's weakest assumption matches the load-bearing step exactly. Because the provided text is abstract-only, no further internal inconsistency can be diagnosed, but the claim that the same prior works at an intermediate level without destroying either global or local properties remains the point that must be empirically secured before the sampler can be considered reliable.","tokens_in":1808,"tokens_out":298,"duration_ms":15001,"concrete_test":"In the robotic planning experiments, fix the coarse scaffold and re-run the second stage at noise levels t=0.2, 0.4, 0.6, 0.8 (normalized); if global coherence (task success or route validity) drops >15% or local quality metrics degrade outside a narrow band, the intermediate-level assumption fails to generalize.","verdict_should_be":"CONDITIONAL","load_bearing_attack":"CoFi's second stage diffuses the coarse scaffold to an intermediate noise level before re-applying the pretrained short-horizon prior. This step implicitly requires that (a) the prior's denoising trajectory at that level can add fine structure without erasing the global arrangement imposed in stage 1, and (b) a single such level works across tasks without task-specific retuning. The abstract provides no equation, schedule, or ablation characterizing this level or demonstrating stability under perturbation.","agreement_with_reader":"agree"},"referee_report":{"model":"grok-4.3","summary":"The manuscript proposes Coarse-to-Fine Compositional Diffusion (CoFi), an inference-time sampler that first aligns local denoised estimates from a pretrained short-horizon diffusion prior around a shared coarse structure to form a global scaffold, then diffuses this scaffold to an intermediate noise level and reapplies the same prior to restore fine details while preserving global coherence. It claims this yields better global coherence and local sample quality than prior compositional baselines across long-horizon robotic planning, panoramic image generation, and long video generation, while requiring 2-8x fewer denoiser evaluations.","tokens_in":1910,"tokens_out":344,"duration_ms":12530,"significance":"If the intermediate noise level can be shown to reliably exist and generalize, the approach would provide an efficient way to extend short-horizon diffusion priors to long-horizon tasks without retraining or expensive inference-time optimization, addressing a practical bottleneck in compositional generation for robotics and sequential data.","major_comments":[{"comment":"Abstract (method description): The second stage relies on the existence of an intermediate noise level at which the pretrained local prior can add fine structure without erasing the imposed global scaffold; no equation, noise schedule, selection procedure, or ablation is supplied to characterize this level, demonstrate its stability under perturbation, or show that a single choice works across tasks without retuning.","section":"Abstract"}],"minor_comments":[{"comment":"The abstract states performance and efficiency gains but supplies no quantitative tables, error analysis, or baseline comparisons that would allow verification of the 2-8x claim or assessment of variance.","section":"Abstract"}],"recommendation":"major_revision","confidential_remarks":null},"author_rebuttal":{"model":"grok-4.3","summary":"We thank the referee for the careful review and constructive comment. We address the concern about the intermediate noise level below.","responses":[{"response":"We agree that the manuscript does not supply an explicit equation, noise schedule, selection procedure, or ablation study characterizing the intermediate noise level used in the second stage. In the revised version we will add these elements to the method section and abstract: an equation defining the target noise level, the fixed schedule employed, the selection procedure (based on matching scaffold scale to the prior's effective receptive field on a small validation set), and an ablation demonstrating stability under small perturbations as well as consistency of the same level across the three evaluated tasks.","revision_made":"yes","referee_comment":"[Abstract] Abstract (method description): The second stage relies on the existence of an intermediate noise level at which the pretrained local prior can add fine structure without erasing the imposed global scaffold; no equation, noise schedule, selection procedure, or ablation is supplied to characterize this level, demonstrate its stability under perturbation, or show that a single choice works across tasks without retuning."}],"tokens_in":1335,"tokens_out":247,"duration_ms":19997,"standing_objections":[]},"desk_editor":{"model":"grok-4.3","letter":"The paper's actual new element is the explicit two-stage sampler: first force local denoised estimates to align on a shared coarse scaffold for global structure, then diffuse that scaffold to an intermediate noise level and run the pretrained short-horizon prior again to restore fine details. This is presented as different from both simple overlapping composition and the heavier propagation or optimization baselines.\n\nIt does a straightforward job spelling out why local consistency alone can still produce implausible long-horizon outputs and why existing fixes scale poorly with more local plans. The efficiency numbers—2-8x fewer denoiser evaluations while claiming gains in both coherence and sample quality across robotic planning, panoramas, and video—are the practical hook that could interest people who actually run these models.\n\nThe soft spot is the intermediate noise step itself. The abstract gives no equation, schedule, or ablation for choosing or validating that level, and nothing shows it works without task-specific tuning or that the prior reliably adds local structure without erasing the imposed scaffold. If that assumption fails on some tasks or requires per-domain fiddling, the claimed efficiency advantage disappears. The provided text also skips any equations or error analysis, so the central performance claims stay hard to verify.\n\nThis is for researchers working on diffusion-based planning or long-sequence generation who need lighter inference. A reader who cares about robotics applications or compositional generative models would get concrete value from the experiments once the method details and ablations are checked.\n\nI would send it to peer review. The procedure is distinct enough and the efficiency angle concrete enough to deserve referee time, even if the noise-level robustness needs more evidence.","headline":"CoFi's scaffold-then-intermediate-refine sampler is a distinct inference trick that targets global coherence cheaply, but the key assumption about the noise level lacks supporting detail or tests.","tokens_in":2377,"tokens_out":407,"would_cite":false,"duration_ms":14759,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"grok-4.3","headline":"CoFi produces globally coherent long-horizon outputs by first aligning local diffusion estimates into a shared coarse scaffold then restoring fine details at an intermediate noise level with the same pretrained prior.","keywords":["diffusion models","compositional generation","long-horizon planning","robotic planning","video generation","image generation","inference-time sampling"],"falsifier":"Running the alignment stage on a short-horizon prior and measuring whether local consistency metrics drop sharply, or testing whether denoising the scaffold at the chosen intermediate noise level erases the global arrangement in held-out long-horizon examples.","tokens_in":2732,"feed_emoji":"","tokens_out":671,"duration_ms":11152,"temperature":0.7,"pith_summary":"The paper establishes that standard compositional diffusion enforces only local consistency between neighboring plans, which can still yield implausible global structures in long-horizon tasks. CoFi addresses this by separating the process into two stages: first aligning denoised local estimates around one shared coarse structure to form a global scaffold, then diffusing that scaffold to an intermediate noise level and denoising it again with the original short-horizon prior to recover local details. This approach is shown to improve both global coherence and local sample quality across robotic planning, panoramic image generation, and long video generation while using 2-8 times fewer denoiser calls than prior methods that rely on repeated propagation or inference-time optimization.","feed_headline":"CoFi cuts diffusion evaluations 2-8x while raising long-horizon coherence","feed_subtitle":"Alignment around one coarse scaffold followed by intermediate-noise refinement preserves global structure without retraining the local prior","key_machinery":"Coarse-to-Fine Compositional Diffusion (CoFi) sampler, which separates global structure formation from local detail refinement at inference time.","core_discovery":"CoFi first aligns local denoised estimates around a shared coarse structure, producing a global scaffold that captures the long-range task-level arrangement. It then diffuses this scaffold to an intermediate noise level and denoises it with the same pretrained local prior, restoring local fine structure while preserving the scaffold-induced global coherence.","pith_inferences":["The two-stage separation could be applied to any pretrained diffusion prior whose training horizon is shorter than the target output length.","Choosing the intermediate noise level might be automated by monitoring when global scaffold statistics stabilize during the first stage.","The approach suggests that global structure can be imposed without retraining if the prior already encodes the right local statistics."],"forward_implications":["Locally compatible plans no longer form implausible global routes or task sequences.","Global coherence and local sample quality both improve over standard compositional baselines.","The method requires 2-8 times fewer denoiser evaluations than repeated propagation or optimization approaches.","The same procedure applies without modification to robotic planning, panoramic images, and long video generation."],"fun_headline_variants":["CoFi aligns local plans around shared coarse scaffold","CoFi first forms global scaffold then refines local details","Coarse structure guides CoFi refinement at intermediate noise","CoFi separates global scaffold from local prior refinement"],"cache_read_input_tokens":2112,"weakest_assumption_plain":"Forcing alignment around a shared coarse structure will not destroy the local consistency properties of the pretrained short-horizon prior, and an intermediate noise level exists that lets the same prior restore fine details without erasing the imposed global scaffold.","fun_headline_variants_meta":{"raw":{"variants":["CoFi aligns local plans around shared coarse scaffold","CoFi first forms global scaffold then refines local details","Coarse structure guides CoFi refinement at intermediate noise","CoFi separates global scaffold from local prior refinement"]},"model":"grok-4.3","cost_usd":0.004649,"raw_usage":{"total_tokens":2306,"prompt_tokens":678,"num_sources_used":0,"completion_tokens":60,"cost_in_usd_ticks":46487000,"prompt_tokens_details":{"text_tokens":678,"audio_tokens":0,"image_tokens":0,"cached_tokens":256},"completion_tokens_details":{"audio_tokens":0,"reasoning_tokens":1568,"accepted_prediction_tokens":0,"rejected_prediction_tokens":0}},"tokens_in":678,"tokens_out":60,"duration_ms":11424,"temperature":1.0,"reasoning_tokens":1568,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-06-28T18:22:44.800799+00:00","model_set":{"reader":"grok-4.3"},"falsifier":"Running the alignment stage on a short-horizon prior and measuring whether local consistency metrics drop sharply, or testing whether denoising the scaffold at the chosen intermediate noise level erases the global arrangement in held-out long-horizon examples.","supporting_citations":[],"review_version":1}