{ "schema": "project-experience.writing-task-inputs.v1", "version": "1.0", "status": "constructed_example_not_evaluated", "historical_prompts": false, "training_or_evaluation_run": false, "context_rule": "Only supplied excerpts with Overleaf revision <= cutoff are included. Prompts and selected cutoffs were designed retrospectively; this is not original-context replay or a sealed evaluation.", "tasks": [ { "id": "WA1", "title": "Make the formulation explicit without manufacturing novelty", "action": "Revise scientific positioning", "cutoff_revision": 439, "prompt": "Using only the supplied introduction excerpts and working note, write a short replacement passage that makes learning from expert examples explicit. Preserve the described prefix-to-next-decision supervision. Respond to the note's comparison request, but do not invent facts about the comparator named “graphq.” If the supplied material cannot support a substantive comparison, identify the missing evidence separately rather than assert an advantage. Do not claim a new invention date, priority, measured speedup, training stability or algorithm switch.", "output_request": "A proposed introduction passage of roughly 80–140 words, followed by a brief missing-evidence note. The length is a new task constraint, not a historical requirement.", "skill_ids": [ "WS01", "WS05", "WS06" ], "input_note": "The revision-439 note is supplied with older revision-256 passages. This is deliberately a constructed archival exercise; the full introduction at revision 439 is unavailable.", "current_source_ids": [ "A02" ], "earlier_context_ids": [ "A01", "A09" ], "sources": [ { "evidence_id": "A02", "source_file": "introduction.tex", "content_kind": "working_note", "text": "Use imitation learning to compare with graphq", "text_view": "full_source_excerpt", "presented_text_sha256": "c576b1f6d1559dd8c497d453c750aebcedcbbc6361f315bd47021ef55b79f88f", "context_status": "A selected manuscript passage, not a complete record of the context available when it was written. The records do not establish who wrote it; scientific claims are unverified.", "excerpt_sha256": "c576b1f6d1559dd8c497d453c750aebcedcbbc6361f315bd47021ef55b79f88f", "manuscript_revision": 439 }, { "evidence_id": "A01", "source_file": "introduction.tex", "content_kind": "manuscript_prose", "text": "Most learning-based SAT methods either provide instance-level guidance or learn state-centric policies; neither directly models the decision sequence executed at runtime.\nThis suggests aligning learning targets with solver behavior by modeling the decision sequence itself. Long‑context sequence models—particularly Transformers~\\citep{vaswani2017attention}-are well-suited to this formulation and can be used to propose the next branch online within CDCL.", "text_view": "full_source_excerpt", "presented_text_sha256": "3c563569a464806b863e25684389da27e5b06c30bf79f221801f55e670e13456", "context_status": "A selected manuscript passage, not a complete record of the context available when it was written. The records do not establish who wrote it; scientific claims are unverified.", "excerpt_sha256": "3c563569a464806b863e25684389da27e5b06c30bf79f221801f55e670e13456", "manuscript_revision": 256 }, { "evidence_id": "A09", "source_file": "introduction.tex", "content_kind": "manuscript_prose", "text": "Training data are [keytraces] distilled from solved instances. From a full CDCL run, contiguous backtracks are collapsed to obtain a near conflict‑free sequence of branching decisions along a solution or refutation path. Each prefix paired with its subsequent decision provides a clear stepwise target.", "text_view": "full_source_excerpt", "presented_text_sha256": "4b72f3dad06de7b0d9f0ee8a7acaa7d6dfdb62d60f6a9dc86fd866b7406455d7", "context_status": "A selected manuscript passage, not a complete record of the context available when it was written. The records do not establish who wrote it; scientific claims are unverified.", "excerpt_sha256": "4b72f3dad06de7b0d9f0ee8a7acaa7d6dfdb62d60f6a9dc86fd866b7406455d7", "manuscript_revision": 256 } ], "prompt_origin": "newly_authored_derived_instruction_not_historical_prompt", "historical_note_available": true, "reference_completion_included": false, "independent_project_group": "ImitSAT-single-project" }, { "id": "WA2", "title": "Connect literature strands without rewriting the science", "action": "Draft an argument bridge", "cutoff_revision": 1597, "prompt": "Revise the supplied draft bridge so a reader can follow the connection from prior solver guidance and imitation learning to the proposal. Use only the supplied manuscript passages; their descriptions of prior work and performance are not independently verified. Repair the opening fragment, separate motivation from mechanism, and keep the next-decision supervision account intact. Do not add citations, strengthen benefit claims or invent a new experimental result. Put any unsupported-benefit concern in a separate short review note.", "output_request": "One revised paragraph and a short explanation of what was preserved, changed and still needs evidence. These instructions are newly authored.", "skill_ids": [ "WS02", "WS03", "WS07", "WS05" ], "input_note": "The current bridge is paired with selected older passages from the same related-work file, not a complete file snapshot at the cutoff.", "current_source_ids": [ "B04" ], "earlier_context_ids": [ "B02", "A08" ], "sources": [ { "evidence_id": "B04", "source_file": "related works.tex", "content_kind": "manuscript_prose", "text": "Based on the above limitation with neural guidance for SAT, and inspired by imitation learning. We propose \\modelname{}, a branching policy for CDCL that clones a near conflict‑free KeyTrace distilled from solver runs. We cast branching as prefix-conditioned sequence prediction and train an autoregressive next-decision model on compact sequences of surviving decisions, providing dense, decision-level supervision with low per-query cost. These designs help reduce propagation and improve wall-clock time under small query budgets.", "text_view": "full_source_excerpt", "presented_text_sha256": "d3bd52fcc95805ef9a0bd17792caa3651a840a5d4964738529325687d26e5915", "context_status": "A selected manuscript passage, not a complete record of the context available when it was written. The records do not establish who wrote it; scientific claims are unverified.", "excerpt_sha256": "d3bd52fcc95805ef9a0bd17792caa3651a840a5d4964738529325687d26e5915", "manuscript_revision": 1597 }, { "evidence_id": "B02", "source_file": "related works.tex", "content_kind": "manuscript_prose", "text": "\\section{Related Works}\n\n\\paragraph{Neural guidance for SAT and CDCL.}\nEarly learning approaches focused on instance-level prediction, using Graph Neural Network (GNNs)~\\citep{scarselli2008graph} to classify SAT or UNSAT, as seen in NeuroSAT~\\citep{selsam2018learning,selsam2019guiding} and ~\\citep{cameron2020predicting}. Recent work has explored whether Transformers can learn solver behavior directly~\\citep{pan2025can}.\nIn parallel with these model-based approaches, complementary efforts target data and benchmarking, including G2SAT~\\citep{you2019g2sat} and G4SATBench~\\citep{lig4satbench}.\nBuilding on these foundations, a second line integrates learning inside solvers to shape specific components: for example, NeuroSelect~\\citep{liu2024neuroselect} learns clause deletion policies, NeuroBack~\\citep{wang2024neuroback} improves phase initialization with GNNs, and RDC‑SAT~\\citep{zhai2025learning} adopts a divide-and-conquer strategy via reinforcement learning. This approach leads to targeted enhancements within solver mechanisms.\nMore concretely, within the CDCL branching loop, several methods exemplify this integration: NeuroSAT~\\citep{selsam2019guiding} has been used to guide variable selection; Graph-Q-SAT~\\citep {kurin2020can} trains an RL agent queried online during search based on instance information; and SATformer~\\citep{shi2023satformer} trains a GNN Transformer model to initialize the CDCL that indirectly influences branching thereafter.\n\n\n\\paragraph{Imitation learning.}", "text_view": "full_source_excerpt", "presented_text_sha256": "214a40d95caf6e9de312dce02260293e33fb53b3aaec31906f4534399b08ffeb", "context_status": "A selected manuscript passage, not a complete record of the context available when it was written. The records do not establish who wrote it; scientific claims are unverified.", "excerpt_sha256": "214a40d95caf6e9de312dce02260293e33fb53b3aaec31906f4534399b08ffeb", "manuscript_revision": 1576 }, { "evidence_id": "A08", "source_file": "related works.tex", "content_kind": "manuscript_prose", "text": "\\paragraph{Imitation learning for control.}\nImitation learning (IL) learns policies directly from expert demonstrations, that is, sequences of states with associated actions~\\citep{osa2018algorithmic,zare2024survey}. A simple example is behavior cloning (BC), which utilizes supervised learning to map observed situations to expert choices~\\citep{pomerleau1991efficient}.\nBuilding on the principles of imitation learning, the Decision Transformer~\\citep{chen2021decision} is similar to behavior cloning, framing reinforcement learning as sequence modeling, where an autoregressive Transformer is trained to predict the next action given a sequence rollout of returns, states, and actions. This view connects control to next-token prediction and attains competitive performance without explicit value function learning.\nThe application of imitation learning extends beyond traditional domains. For instance, beyond robotics and games, IL has guided decision-making in exact optimization solvers. In mixed-integer linear programming, policies learned to imitate strong branching can be used within branch-and-bound and achieve strong results~\\citep{gasse2019exact}. Related work also learns branching policies that integrate into branch-and-bound~\\citep{zarpellon2021parameterizing}.", "text_view": "full_source_excerpt", "presented_text_sha256": "241387f79147cf211eb2162a15051cb4946a63ed85f769293b9f0b4c1ed37670", "context_status": "A selected manuscript passage, not a complete record of the context available when it was written. The records do not establish who wrote it; scientific claims are unverified.", "excerpt_sha256": "241387f79147cf211eb2162a15051cb4946a63ed85f769293b9f0b4c1ed37670", "manuscript_revision": 1585 } ], "prompt_origin": "newly_authored_derived_instruction_not_historical_prompt", "historical_note_available": false, "reference_completion_included": false, "independent_project_group": "ImitSAT-single-project" }, { "id": "WA3", "title": "Consolidate a mixed draft and preserve the current scope", "action": "Revise without regression", "cutoff_revision": 2895, "prompt": "Edit the current appendix excerpt into a coherent explanation using only the supplied material. Retain the definition of query budget, distinguish saved solver work from the computational cost of model calls, and remove repeated active prose. Keep the scope of the current draft's runtime assertion; do not restore an older broader claim from historical context. Do not infer an optimal budget or create results from a figure that is not supplied. Return a proposed revision, then a short list of claims still needing experimental support.", "output_request": "A revised section body and a brief evidence-check note. Preserve useful qualifications; no fixed historical wording is required.", "skill_ids": [ "WS04", "WS05", "WS06", "WS07" ], "input_note": "Comment-only lines are left out of this task input. The complete manuscript excerpts, including comments, remain available in the source library. The current appendix is at revision 2895; optional earlier context is at revision 2614.", "current_source_ids": [ "C08" ], "earlier_context_ids": [ "C06" ], "sources": [ { "evidence_id": "C08", "source_file": "appendix.tex", "content_kind": "mixed_prose_and_comments", "text": "\\section{Query Budget Analysis}\n\\label{sec:query-budget}\nIn this section, we study how the query budget, the number of times the solver consults the model during search, affects both effectiveness and computational cost. We vary the budget from 1 to 10 calls, and also include an all-calls configuration, to understand how much benefit each additional model query provides.\n\nAs shown in Figure~\\ref{fig:query}, \nthe performance gain exhibits clear diminishing returns: the majority of improvement is obtained from the first three queries, while additional calls beyond six provide only marginal benefits. Because each query introduces non-trivial latency, we interpret the reduction in propagations achieved by the model as a benefit that must be weighed against the computational overhead of issuing more queries.\n\nperformance exhibits diminishing returns as the query budget increases. Most improvement comes from the first three queries. Additional queries beyond six calls offer only marginal gains.\nWe view the reduction in propagations achieved by the model as a benefit that must be traded off against the computational cost of additional queries.\nTo balance this trade-off, we set a default query budget of 3.\nMoreover, as shown in Figure~\\ref{fig:wallclock-methods}, GQSAT also exhibit worse wall-clock performance as the query budget increases from 3 to 5.", "text_view": "comment_lines_omitted", "presented_text_sha256": "179a678e40edba25a6097dd06ccb965761655e20dc72776775052dacd59f6984", "context_status": "A selected manuscript passage, not a complete record of the context available when it was written. The records do not establish who wrote it; scientific claims are unverified.", "excerpt_sha256": "2977a8090fad4c8f08af643bce95278a35db9390add20a7a02cef4799343d91c", "manuscript_revision": 2895 }, { "evidence_id": "C06", "source_file": "appendix.tex", "content_kind": "mixed_prose_and_comments", "text": "\\section{Query Budget Analysis}\nWe evaluate our method across different query budgets from 1 to 10 model calls and an all-calls setting.\nAs shown in Figure~\\ref{fig:query}, performance exhibits diminishing returns as the query budget increases. Most improvement comes from the first three queries. Additional queries beyond six calls offer only marginal gains.\nWe view the reduction in propagations achieved by the model as a benefit that must be traded off against the computational cost of additional queries.\nTo balance this trade-off, we set a default query budget of 3.\nMoreover, as shown in Figure~\\ref{fig:wallclock-methods}, both our method and GQSAT exhibit worse wall-clock performance as the query budget increases from 3 to 5.", "text_view": "comment_lines_omitted", "presented_text_sha256": "0ccf4605824e9f2139a736dbea8c9f47159cc9998b0a9b293bec83f5f1e7d95e", "context_status": "A selected manuscript passage, not a complete record of the context available when it was written. The records do not establish who wrote it; scientific claims are unverified.", "excerpt_sha256": "798b4f47187458fa64e6c2036d5f4d4f124499b2738336538895cca64bff12fa", "manuscript_revision": 2614 } ], "prompt_origin": "newly_authored_derived_instruction_not_historical_prompt", "historical_note_available": false, "reference_completion_included": false, "independent_project_group": "ImitSAT-single-project" } ], "contributor_privacy": "Private contributor labels use generic markers; scientific wording is preserved.", "use_notice": "For discussing a possible data partnership. Training, redistribution and other uses require a separate agreement." }