{ "title": "ImitSAT · From Research to Shared Understanding", "purpose": "Explore how one research paper develops through drafts, notes, and revisions—and the writing decisions along the way.", "skill_taxonomy": { "artifact_kind": "writing_skill_framework_not_validated_labels", "principles": [ "Multi-label when supported.", "Reader-goal tags express intended value, not measured gains.", "Do not force coverage or inflate routine edits.", "Proposed tasks are not historical prompts or ready-made training labels." ], "skills": [ { "id": "WS01", "label": "Scientific framing", "reader_purpose": "Make the contribution recognizable and its significance assessable.", "subskills": [ "problem and gap formulation", "contribution identity", "novelty and scope", "relation between the formulation and prior approaches" ], "evidence_to_look_for": [ "A change in what the manuscript presents as its contribution or justification.", "A formulation promoted across the argument, even when technical ingredients existed earlier." ], "insufficient_evidence": [ "A new keyword alone.", "An inferred moment of invention." ], "review_questions": [ "Does the earlier text already contain the underlying concept?", "After equally fluent paraphrasing, does the contribution or justification differ?" ], "reader_goal_tags": [ "understand", "scrutinize" ], "potential_tasks": [ "Explain how positioning changed using evidence.", "Propose a context-consistent framing while preserving technical meaning." ] }, { "id": "WS02", "label": "Literature positioning", "reader_purpose": "Help a reader see why prior work matters to the present argument.", "subskills": [ "organizing related work", "gap construction", "similarities and differences", "citation function and attribution" ], "evidence_to_look_for": [ "Reorganizing or connecting prior-work discussion to a contribution or limitation." ], "insufficient_evidence": [ "A changed citation count with no demonstrated argumentative change.", "Claiming cited research is correct from the manuscript alone." ], "review_questions": [ "What argumentative job does the cited work perform before and after?", "Is the relationship supported, or only asserted by the manuscript?" ], "reader_goal_tags": [ "understand", "scrutinize", "build_on" ], "potential_tasks": [ "Construct a literature argument using eligible context.", "Identify unsupported comparative statements." ] }, { "id": "WS03", "label": "Argument and structure", "reader_purpose": "Make the account coherent and navigable.", "subskills": [ "planning and drafting", "introducing new content", "section ordering", "argument dependencies", "cross-section narrative" ], "evidence_to_look_for": [ "Meaningful additions, reorganizations, or transitions that change how the explanation develops." ], "insufficient_evidence": [ "A file rename or restore mistaken for conceptual reorganization." ], "review_questions": [ "Which reader dependency or argumentative gap is addressed?", "Did meaningful drafting occur, rather than only polishing an existing paragraph?" ], "reader_goal_tags": [ "understand", "build_on" ], "potential_tasks": [ "Plan or revise a section with its role in the paper specified.", "Identify missing premises or misplaced explanation." ] }, { "id": "WS04", "label": "Technical explanation", "reader_purpose": "Make methods and their conditions understandable and usable.", "subskills": [ "definitions and notation", "assumptions", "equations and algorithms", "levels of explanation", "implementation detail" ], "evidence_to_look_for": [ "A source-supported change in how a technical procedure or assumption is explained." ], "insufficient_evidence": [ "A wording change interpreted as proof that the implementation changed." ], "review_questions": [ "What is clearer or more explicit, and what remains unknown?", "Is the account faithful to available contemporaneous technical context?" ], "reader_goal_tags": [ "understand", "scrutinize", "build_on" ], "potential_tasks": [ "Clarify a method without changing its scientific meaning.", "Identify missing definitions or assumptions." ] }, { "id": "WS05", "label": "Evidence and claim calibration", "reader_purpose": "Let readers judge what the evidence supports and where it stops.", "subskills": [ "experimental question", "comparison and cost explanation", "claim-to-evidence alignment", "scope and uncertainty", "limitations" ], "evidence_to_look_for": [ "Changes in the explanation of a comparison, experimental claim, cost, or limitation." ], "insufficient_evidence": [ "Treating manuscript claims as independently verified results.", "Attributing publication acceptance to particular edits." ], "review_questions": [ "What evidence is actually available, as opposed to merely described?", "Does a claim become narrower, broader, better explained, or remain unresolved?" ], "reader_goal_tags": [ "understand", "scrutinize", "build_on" ], "potential_tasks": [ "Explain a comparison under its stated conditions.", "Revise a claim without extending its evidential scope." ] }, { "id": "WS06", "label": "Revision judgment and consistency", "reader_purpose": "Improve a document without losing valid content or introducing contradictions.", "subskills": [ "feedback interpretation when recorded", "prioritizing changes", "preserving valid text", "cross-section consistency", "reversions and regressions" ], "evidence_to_look_for": [ "A source-linked response to recorded feedback or a change reconciling multiple passages." ], "insufficient_evidence": [ "Invented feedback or intention.", "Every later revision treated as improvement." ], "review_questions": [ "What should remain unchanged?", "Does the revision introduce inconsistencies elsewhere?", "Is the feedback connection recorded or only conjectured?" ], "reader_goal_tags": [ "understand", "scrutinize", "build_on" ], "potential_tasks": [ "Address a real or explicitly derived revision request while preserving valid content.", "Check terminology and contribution claims across sections." ] }, { "id": "WS07", "label": "Language and presentation", "reader_purpose": "Improve readability and precision without distorting the research.", "subskills": [ "wording and concision", "terminology and notation", "captions and tables", "visual explanation when sources exist", "references and formatting" ], "evidence_to_look_for": [ "Meaning-preserving expression changes, or source-supported presentation changes." ], "insufficient_evidence": [ "Unavailable historical images treated as reconstructed evidence.", "Cosmetic changes described as conceptual innovation." ], "review_questions": [ "Does compression lose a caveat or technical condition?", "Are the relevant figures/tables/captions actually available?" ], "reader_goal_tags": [ "understand", "scrutinize", "build_on" ], "potential_tasks": [ "Compress an explanation while preserving qualifications.", "Improve a caption using only available evidence." ] } ], "provenance_note": "Writing skills identified by the curators to help readers explore the sample. These associations are interpretations, not independently validated labels.", "description": "Seven overlapping ways to interpret the writing skills illustrated by these selected passages. Associations are commentary, not validated training labels." }, "chapters": [ { "id": "T1", "title": "Make the contribution explicit", "reader_question": "What is the scientific contribution, and how is it positioned?", "scope": "Selected manuscript states across the introduction and methods; not a causal reconstruction of the research idea.", "steps": [ { "id": "T1-01", "title": "The method is described before its later framing", "primary_skill_id": "WS01", "skill_focus": { "skill_id": "WS01", "label": "Scientific framing", "subskill": "Separate method ingredients from their framing", "evidence_linked_rationale": "Expert-prefix targets already appear in the earlier introduction; later positioning must be read against that baseline.", "evidence_ids": [ "A01", "A09" ], "status": "curator_interpretation_not_validated" }, "reader_goal": { "id": "understand", "purpose": "Recognize the existing technical idea before interpreting a later change in presentation.", "evidence_ids": [ "A01", "A09" ], "kind": "proposed_reader_value", "measured": false }, "evidence_ids": [ "A01", "A09" ], "original_instruction": { "status": "not_established_for_this_reading_step", "text": null }, "quality_preference": null, "training_example_status": "not_prepared_or_evaluated", "learning_opportunity": { "kind": "newly_authored_task_sketch_not_historical_prompt", "question": "Explain what is already specified about supervision, without claiming when the idea was invented.", "input_evidence_ids": [ "A01", "A09" ], "view": "retrospective_source_inspection", "answers": "Open review question; no gold answer, preference or reward supplied.", "cutoff_safe_training_export": false, "training_or_evaluation_run": false, "evidence_overlap_group": "ImitSAT-single-project" }, "reading_order_in_chapter": 1, "reading_view": { "title": "The method is described before its later framing", "reader_problem": "When a paper adopts a new label, did the method change—or did the explanation become more explicit?", "plain_context": "The paper describes a solver that learns its next decision from examples of earlier solver decisions. You do not need the solver details to follow the writing problem.", "observed_development": "At Overleaf revision 256, the introduction already describes sequences of decisions and a next-decision target for every prefix.", "why_this_matters": { "text": "Ground a contribution statement in what the method already does. Otherwise, a clearer later label can be mistaken for a newly invented method.", "kind": "analyst_interpretation_of_reader_value", "evidence_ids": [ "A01", "A09", "A02", "A07" ], "measured_reader_gain": false }, "history_value": { "text": "The earlier account precedes the comparison note at revision 439. A later contribution list alone would not establish that the supervision account already existed before that note.", "kind": "earlier_state_in_a_trajectory", "evidence_ids": [ "A01", "A09", "A02", "A07" ], "endpoint_is_published_final": false, "model_advantage_demonstrated": false }, "evidence_ids": [ "A01", "A09", "A02", "A07" ], "review_status": "curator_commentary_not_validated", "boundary": "This establishes an earlier written description, not when the method was invented or whether implementation stayed unchanged.", "focus_passages": [ { "evidence_id": "A01", "label": "Earlier introduction", "text": "This suggests aligning learning targets with solver behavior by modeling the decision sequence itself.", "start_character_in_excerpt": 171, "end_character_in_excerpt": 273, "verbatim": true, "verbatim_relative_to": "source_excerpt" }, { "evidence_id": "A09", "label": "Supervision in the same version", "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.", "start_character_in_excerpt": 0, "end_character_in_excerpt": 302, "verbatim": true, "verbatim_relative_to": "source_excerpt" } ], "reason_evidence": { "status": "analyst_interpretation", "explanation": "A proposed explanation of communicative value, not recovered writer intention.", "evidence_ids": [ "A01", "A09", "A02", "A07" ] }, "short_skill": "Separate method ingredients from their framing", "skill_ids": [ "WS01" ] } }, { "id": "T1-02", "title": "A note asks for a scientific comparison", "primary_skill_id": "WS01", "skill_focus": { "skill_id": "WS01", "label": "Scientific framing", "subskill": "Turn a comparison request into a framing question", "evidence_linked_rationale": "The actual note requests an imitation-learning comparison, rather than a grammar change.", "evidence_ids": [ "A01", "A09", "A02" ], "status": "curator_interpretation_not_validated" }, "reader_goal": { "id": "scrutinize", "purpose": "Know which scientific distinction the comparison needs to make explicit.", "evidence_ids": [ "A01", "A09", "A02" ], "kind": "proposed_reader_value", "measured": false }, "evidence_ids": [ "A01", "A09", "A02" ], "original_instruction": { "status": "recorded_source_note", "evidence_id": "A02", "text": "Use imitation learning to compare with graphq" }, "quality_preference": null, "training_example_status": "not_prepared_or_evaluated", "learning_opportunity": { "kind": "newly_authored_task_sketch_not_historical_prompt", "question": "Given the earlier account and this recorded note, identify the comparison to develop and what further evidence is needed.", "input_evidence_ids": [ "A01", "A09", "A02" ], "view": "retrospective_source_inspection", "answers": "Open review question; no gold answer, preference or reward supplied.", "cutoff_safe_training_export": false, "training_or_evaluation_run": false, "evidence_overlap_group": "ImitSAT-single-project" }, "reading_order_in_chapter": 2, "reading_view": { "title": "A note asks for a scientific comparison", "reader_problem": "What comparison should help the reader recognize the contribution?", "plain_context": "The existing draft describes learning from solver examples. A later working note asks to use imitation learning as the comparison frame.", "observed_development": "The note explicitly requests an imitation-learning comparison with “graphq”. This is recorded text, not a reconstructed prompt.", "why_this_matters": { "text": "The recorded request concerns how the approach is positioned against another approach, not merely making the language more fluent.", "kind": "analyst_interpretation_of_reader_value", "evidence_ids": [ "A09", "A02" ], "measured_reader_gain": false }, "history_value": { "text": "A final or later paragraph may present the comparison, but would not establish this recorded intermediate request or its order relative to the earlier supervision description.", "kind": "recorded_request_in_history", "evidence_ids": [ "A09", "A02" ], "endpoint_is_published_final": false, "model_advantage_demonstrated": false }, "evidence_ids": [ "A09", "A02" ], "review_status": "curator_commentary_not_validated", "boundary": "The note names “graphq”; its author, original conversation and causal effect on later revisions are unknown.", "focus_passages": [ { "evidence_id": "A09", "label": "Already in the earlier introduction", "text": "Each prefix paired with its subsequent decision provides a clear stepwise target.", "start_character_in_excerpt": 221, "end_character_in_excerpt": 302, "verbatim": true, "verbatim_relative_to": "source_excerpt" }, { "evidence_id": "A02", "label": "Recorded working note", "text": "Use imitation learning to compare with graphq", "start_character_in_excerpt": 0, "end_character_in_excerpt": 45, "verbatim": true, "verbatim_relative_to": "source_excerpt" } ], "reason_evidence": { "status": "recorded_request", "explanation": "A request is present in a working note; author, original context and causal effect are unknown.", "evidence_ids": [ "A09", "A02" ] }, "short_skill": "Turn a comparison request into a framing question", "skill_ids": [ "WS01", "WS02" ] } }, { "id": "T1-03", "title": "Explain why the model fits the job", "primary_skill_id": "WS04", "skill_focus": { "skill_id": "WS04", "label": "Technical explanation", "subskill": "Explain why the modeling choice fits the operation", "evidence_linked_rationale": "The methods passage connects one requested branch, the available prefix, and autoregressive next-element prediction.", "evidence_ids": [ "A03" ], "status": "curator_interpretation_not_validated" }, "reader_goal": { "id": "understand", "purpose": "Follow the relation between the solver operation and the learning formulation.", "evidence_ids": [ "A03" ], "kind": "proposed_reader_value", "measured": false }, "evidence_ids": [ "A03" ], "original_instruction": { "status": "not_established_for_this_reading_step", "text": null }, "quality_preference": null, "training_example_status": "not_prepared_or_evaluated", "learning_opportunity": { "kind": "newly_authored_task_sketch_not_historical_prompt", "question": "Identify the inputs, output and stated reason for using an autoregressive model.", "input_evidence_ids": [ "A03" ], "view": "retrospective_source_inspection", "answers": "Open review question; no gold answer, preference or reward supplied.", "cutoff_safe_training_export": false, "training_or_evaluation_run": false, "evidence_overlap_group": "ImitSAT-single-project" }, "reading_order_in_chapter": 3, "reading_view": { "title": "Explain why the model fits the job", "reader_problem": "Why should this modeling choice make sense to a reader?", "plain_context": "The solver asks for one decision at a time. The method passage describes a model that predicts a next item from the decisions already given.", "observed_development": "The methods passage states the operational requirement, then says next-element sequence prediction fits that requirement.", "why_this_matters": { "text": "A technical label is more assessable when the explanation connects it to the operation it must perform.", "kind": "analyst_interpretation_of_reader_value", "evidence_ids": [ "A03", "A02" ], "measured_reader_gain": false }, "history_value": { "text": "This is a supporting methods snapshot. Its place after the framing note supplies context, but we do not have a before/after methods pair showing this explanation being introduced.", "kind": "supporting_snapshot", "evidence_ids": [ "A03", "A02" ], "endpoint_is_published_final": false, "model_advantage_demonstrated": false }, "evidence_ids": [ "A03", "A02" ], "review_status": "curator_commentary_not_validated", "boundary": "The link to the introduction is thematic and cross-file. It does not prove that the note caused a methods revision.", "focus_passages": [ { "evidence_id": "A03", "label": "Method-section passage", "text": "A learner is now introduced to imitate the expert keytrace.\nThe CDCL solver requests one branch at a time, so the learner must map the formula and a prefix of the expert trace to the next signed variable under a small computational budget.\nAn Autoregressive (AR) Model approach fits this need, since it conditions on a prefix and predicts the next element in a sequence.", "start_character_in_excerpt": 39, "end_character_in_excerpt": 409, "verbatim": true, "verbatim_relative_to": "source_excerpt" } ], "reason_evidence": { "status": "source_stated_rationale", "explanation": "The manuscript itself offers a reason; this is not independent validation.", "evidence_ids": [ "A03", "A02" ] }, "short_skill": "Explain why the modeling choice fits the operation", "skill_ids": [ "WS04", "WS01" ] } }, { "id": "T1-07", "title": "Give the reader a route through the method", "primary_skill_id": "WS03", "skill_focus": { "skill_id": "WS03", "label": "Argument and structure", "subskill": "Order an explanation by its dependencies", "evidence_linked_rationale": "The roadmap places the representation before the prediction model and its online use.", "evidence_ids": [ "A03" ], "status": "curator_interpretation_not_validated" }, "reader_goal": { "id": "understand", "purpose": "Follow a technical explanation in an explicit sequence.", "evidence_ids": [ "A03" ], "kind": "proposed_reader_value", "measured": false }, "evidence_ids": [ "A03" ], "original_instruction": { "status": "not_established_for_this_reading_step", "text": null }, "quality_preference": null, "training_example_status": "not_prepared_or_evaluated", "learning_opportunity": { "kind": "newly_authored_task_sketch_not_historical_prompt", "question": "Identify the announced explanation order and the dependencies it suggests; distinguish the roadmap from verified completeness.", "input_evidence_ids": [ "A03" ], "view": "retrospective_source_inspection", "answers": "Open review question; no gold answer, preference or reward supplied.", "cutoff_safe_training_export": false, "training_or_evaluation_run": false, "evidence_overlap_group": "ImitSAT-single-project" }, "reading_order_in_chapter": 4, "reading_view": { "title": "Give the reader a route through the method", "reader_problem": "What should be explained first so later details have a place?", "plain_context": "This passage announces three parts: prepare the model input, explain the prediction, and describe use in the solver.", "observed_development": "The source gives an explicit reading order after stating the modeling problem.", "why_this_matters": { "text": "A route through the explanation can reduce the need to infer how separate technical pieces connect.", "kind": "analyst_interpretation_of_reader_value", "evidence_ids": [ "A03" ], "measured_reader_gain": false }, "history_value": { "text": "This is a structure example available from one state, not a recovered planning conversation or a newly observed revision. It serves as a current-text control.", "kind": "single_state_control", "evidence_ids": [ "A03" ], "endpoint_is_published_final": false, "model_advantage_demonstrated": false }, "evidence_ids": [ "A03" ], "review_status": "curator_commentary_not_validated", "boundary": "Downstream method sections and the process used to plan them are absent.", "focus_passages": [ { "evidence_id": "A03", "label": "Method roadmap", "text": "The presentation proceeds in three steps.\nWe first build a compact serialization that the learner can read, and then introduce the next decision AR model.\nFinally, we describe online use inside CDCL.", "start_character_in_excerpt": 410, "end_character_in_excerpt": 609, "verbatim": true, "verbatim_relative_to": "source_excerpt" } ], "reason_evidence": { "status": "analyst_interpretation", "explanation": "A proposed explanation of communicative value, not recovered writer intention.", "evidence_ids": [ "A03" ] }, "short_skill": "Order an explanation by its dependencies", "skill_ids": [ "WS03", "WS04" ] } }, { "id": "T1-04", "title": "Give the existing method an explicit identity", "primary_skill_id": "WS01", "skill_focus": { "skill_id": "WS01", "label": "Scientific framing", "subskill": "Name the proposal and explain its supervision", "evidence_linked_rationale": "The introduction pairs an imitation-learner identity with a description of expert-target construction.", "evidence_ids": [ "A01", "A09", "A03", "A04" ], "status": "curator_interpretation_not_validated" }, "reader_goal": { "id": "understand", "purpose": "Recognize the proposal without losing how it learns.", "evidence_ids": [ "A01", "A09", "A03", "A04" ], "kind": "proposed_reader_value", "measured": false }, "evidence_ids": [ "A01", "A09", "A03", "A04" ], "original_instruction": { "status": "not_established_for_this_reading_step", "text": null }, "quality_preference": null, "training_example_status": "not_prepared_or_evaluated", "learning_opportunity": { "kind": "newly_authored_task_sketch_not_historical_prompt", "question": "Explain how the later introduction presents an existing supervision account as a recognizable contribution.", "input_evidence_ids": [ "A01", "A09", "A04" ], "view": "retrospective_source_inspection", "answers": "Open review question; no gold answer, preference or reward supplied.", "cutoff_safe_training_export": false, "training_or_evaluation_run": false, "evidence_overlap_group": "ImitSAT-single-project" }, "reading_order_in_chapter": 5, "reading_view": { "title": "Give the existing method an explicit identity", "reader_problem": "How can a reader tell what kind of contribution is being proposed?", "plain_context": "Earlier prose described learning from decision sequences. The later introduction now calls the proposal an imitation learner and immediately explains the examples it learns from.", "observed_development": "Across the selected introductions, the supervision description is made part of an explicit imitation-learning identity.", "why_this_matters": { "text": "Naming the formulation together with its mechanism makes the contribution easier to inspect than a new label on its own.", "kind": "analyst_interpretation_of_reader_value", "evidence_ids": [ "A01", "A09", "A04" ], "measured_reader_gain": false }, "history_value": { "text": "The earlier and later passages show a change in presentation around already described ingredients. The later introduction alone cannot establish which ingredients preceded that framing.", "kind": "historical_comparison", "evidence_ids": [ "A01", "A09", "A04" ], "endpoint_is_published_final": false, "model_advantage_demonstrated": false }, "evidence_ids": [ "A01", "A09", "A04" ], "review_status": "curator_commentary_not_validated", "boundary": "These are selected passages, not a complete aligned paragraph replacement. Replay, efficiency and solver guarantees remain manuscript assertions.", "focus_passages": [ { "evidence_id": "A01", "label": "Earlier account", "text": "This suggests aligning learning targets with solver behavior by modeling the decision sequence itself.", "start_character_in_excerpt": 171, "end_character_in_excerpt": 273, "verbatim": true, "verbatim_relative_to": "source_excerpt" }, { "evidence_id": "A04", "label": "Later introduction identity", "text": "We introduce \\emph{[modelname]}, an imitation learner for CDCL branching.", "start_character_in_excerpt": 611, "end_character_in_excerpt": 684, "verbatim": true, "verbatim_relative_to": "source_excerpt" }, { "evidence_id": "A04", "label": "Identity connected to supervision", "text": "Then, the learner reads the instance together with a prefix of the keytrace and predicts the next signed variable as the next branch decision.", "start_character_in_excerpt": 933, "end_character_in_excerpt": 1075, "verbatim": true, "verbatim_relative_to": "source_excerpt" } ], "reason_evidence": { "status": "analyst_interpretation", "explanation": "A proposed explanation of communicative value, not recovered writer intention.", "evidence_ids": [ "A01", "A09", "A04" ] }, "short_skill": "Name the proposal and explain its supervision", "skill_ids": [ "WS01", "WS04" ] } }, { "id": "T1-08", "title": "Explain the intervention—and its boundary", "primary_skill_id": "WS04", "skill_focus": { "skill_id": "WS04", "label": "Technical explanation", "subskill": "Bound the proposed intervention", "evidence_linked_rationale": "The source describes when the learned policy acts and when the native solver takes over.", "evidence_ids": [ "A04" ], "status": "curator_interpretation_not_validated" }, "reader_goal": { "id": "build_on", "purpose": "Understand the claimed integration boundary before attempting to reuse the approach.", "evidence_ids": [ "A04" ], "kind": "proposed_reader_value", "measured": false }, "evidence_ids": [ "A04" ], "original_instruction": { "status": "not_established_for_this_reading_step", "text": null }, "quality_preference": null, "training_example_status": "not_prepared_or_evaluated", "learning_opportunity": { "kind": "newly_authored_task_sketch_not_historical_prompt", "question": "Extract the deployment boundary and fallback, separating them from unverified guarantees.", "input_evidence_ids": [ "A04" ], "view": "retrospective_source_inspection", "answers": "Open review question; no gold answer, preference or reward supplied.", "cutoff_safe_training_export": false, "training_or_evaluation_run": false, "evidence_overlap_group": "ImitSAT-single-project" }, "reading_order_in_chapter": 6, "reading_view": { "title": "Explain the intervention—and its boundary", "reader_problem": "Does the learned component replace the solver, or intervene in one part of it?", "plain_context": "The introduction says the model supplies some branching decisions, with fallback to the native rule; other solver parts are described as unchanged.", "observed_development": "The passage states both what the learner does and where it stops.", "why_this_matters": { "text": "A bounded description helps a reader separate the proposed component from the larger system and inspect the claims made for each.", "kind": "analyst_interpretation_of_reader_value", "evidence_ids": [ "A04" ], "measured_reader_gain": false }, "history_value": { "text": "This is a useful single-state explanation. No earlier intervention description is supplied, so it is not evidence of a newly discovered boundary.", "kind": "single_state_control", "evidence_ids": [ "A04" ], "endpoint_is_published_final": false, "model_advantage_demonstrated": false }, "evidence_ids": [ "A04" ], "review_status": "curator_commentary_not_validated", "boundary": "The claimed preservation of completeness and robustness is not verified here; the code and proof are absent.", "focus_passages": [ { "evidence_id": "A04", "label": "Deployment and fallback", "text": "At solve time, the learner acts as a drop-in branching rule under a small query budget, and the solver falls back to the native heuristic when the model is uncertain. \nAll other parts of CDCL remain unchanged, so completeness and robustness are preserved.", "start_character_in_excerpt": 1076, "end_character_in_excerpt": 1331, "verbatim": true, "verbatim_relative_to": "source_excerpt" } ], "reason_evidence": { "status": "analyst_interpretation", "explanation": "A proposed explanation of communicative value, not recovered writer intention.", "evidence_ids": [ "A04" ] }, "short_skill": "Bound the proposed intervention", "skill_ids": [ "WS04", "WS05" ] } }, { "id": "T1-05", "title": "Turn a label into a reasoned comparison", "primary_skill_id": "WS02", "skill_focus": { "skill_id": "WS02", "label": "Literature positioning", "subskill": "Build a comparison with an explicit rationale", "evidence_linked_rationale": "The same-version comparator and contrast passages connect learning paradigms to claimed supervision and optimization properties.", "evidence_ids": [ "A10", "A06" ], "status": "curator_interpretation_not_validated" }, "reader_goal": { "id": "scrutinize", "purpose": "Locate the proposed distinction and the claims that would need supporting evidence.", "evidence_ids": [ "A10", "A06" ], "kind": "proposed_reader_value", "measured": false }, "evidence_ids": [ "A10", "A06" ], "original_instruction": { "status": "not_established_for_this_reading_step", "text": null }, "quality_preference": null, "training_example_status": "not_prepared_or_evaluated", "learning_opportunity": { "kind": "newly_authored_task_sketch_not_historical_prompt", "question": "Separate the comparison itself from claims about convergence, stability and the comparator that need verification.", "input_evidence_ids": [ "A10", "A06" ], "view": "retrospective_source_inspection", "answers": "Open review question; no gold answer, preference or reward supplied.", "cutoff_safe_training_export": false, "training_or_evaluation_run": false, "evidence_overlap_group": "ImitSAT-single-project" }, "reading_order_in_chapter": 7, "reading_view": { "title": "Turn a label into a reasoned comparison", "reader_problem": "Why does the manuscript ask readers to favor this learning formulation?", "plain_context": "The selected later review text contrasts learning from expert examples with an exploration-based alternative. The task here is to inspect the argument, not accept the comparison as true.", "observed_development": "Paired review-macro passages now state a comparator criticism and a proposed advantage of dense supervision.", "why_this_matters": { "text": "The manuscript supplies a rationale that readers can challenge. That is a different writing move from only naming imitation learning.", "kind": "analyst_interpretation_of_reader_value", "evidence_ids": [ "A02", "A10", "A06", "A09" ], "measured_reader_gain": false }, "history_value": { "text": "Read with the earlier note and supervision description, the record shows a comparison being articulated around an existing account. A later paragraph cannot reconstruct those earlier states or establish a switch of algorithms.", "kind": "historical_comparison", "evidence_ids": [ "A02", "A10", "A06", "A09" ], "endpoint_is_published_final": false, "model_advantage_demonstrated": false }, "evidence_ids": [ "A02", "A10", "A06", "A09" ], "review_status": "curator_commentary_not_validated", "boundary": "Comparator descriptions and convergence/stability claims are unverified. Review-macro text may not equal compiled paper text.", "focus_passages": [ { "evidence_id": "A10", "label": "Comparator assertion in the manuscript", "text": "Graph-Q-SAT~\\citep{kurin2020can} introduces an online agent within CDCL, yet it relies on reinforcement learning (RL), which requires extensive exploration and can be unstable due to sparse rewards and delayed feedback.", "start_character_in_excerpt": 389, "end_character_in_excerpt": 608, "verbatim": true, "verbatim_relative_to": "source_excerpt" }, { "evidence_id": "A06", "label": "Proposed contrast in the manuscript", "text": "In contrast, we adopt imitation learning, which learns directly from expert traces. By training on KeyTraces, i.e., collapsed sequences of surviving decisions from full solver runs, our approach provides clear, dense supervision at every branching step.", "start_character_in_excerpt": 13, "end_character_in_excerpt": 266, "verbatim": true, "verbatim_relative_to": "source_excerpt" } ], "reason_evidence": { "status": "source_stated_rationale", "explanation": "The manuscript itself offers a reason; this is not independent validation.", "evidence_ids": [ "A02", "A10", "A06", "A09" ] }, "short_skill": "Build a comparison with an explicit rationale", "skill_ids": [ "WS01", "WS02", "WS03" ] } }, { "id": "T1-06", "title": "Move the framing into the contribution claim", "primary_skill_id": "WS01", "skill_focus": { "skill_id": "WS01", "label": "Scientific framing", "subskill": "Promote a formulation into contribution claims", "evidence_linked_rationale": "The contribution list presents the learning paradigm and sequential targets as explicit contributions.", "evidence_ids": [ "A07" ], "status": "curator_interpretation_not_validated" }, "reader_goal": { "id": "understand", "purpose": "Find what the paper is asking readers to recognize as its contribution.", "evidence_ids": [ "A07" ], "kind": "proposed_reader_value", "measured": false }, "evidence_ids": [ "A07" ], "original_instruction": { "status": "not_established_for_this_reading_step", "text": null }, "quality_preference": null, "training_example_status": "not_prepared_or_evaluated", "learning_opportunity": { "kind": "newly_authored_task_sketch_not_historical_prompt", "question": "Separate the two stated contributions, and retain the qualifications needed to assess them.", "input_evidence_ids": [ "A07" ], "view": "retrospective_source_inspection", "answers": "Open review question; no gold answer, preference or reward supplied.", "cutoff_safe_training_export": false, "training_or_evaluation_run": false, "evidence_overlap_group": "ImitSAT-single-project" }, "reading_order_in_chapter": 8, "reading_view": { "title": "Move the framing into the contribution claim", "reader_problem": "What does the paper now ask readers to recognize as its contribution?", "plain_context": "The later contribution list foregrounds imitation learning and expert-derived decisions. It also makes a priority claim using “first.”", "observed_development": "The formulation appears as an explicit contribution-list claim, rather than only in the method description or explanatory introduction.", "why_this_matters": { "text": "This makes the claimed contribution identifiable—but also creates an obligation to support the priority and technical claims.", "kind": "analyst_interpretation_of_reader_value", "evidence_ids": [ "A09", "A02", "A04", "A07" ], "measured_reader_gain": false }, "history_value": { "text": "The selected history links earlier supervision, an explicit comparison request, and later contribution wording. The list alone cannot establish that developmental path.", "kind": "cross_state_development", "evidence_ids": [ "A09", "A02", "A04", "A07" ], "endpoint_is_published_final": false, "model_advantage_demonstrated": false }, "evidence_ids": [ "A09", "A02", "A04", "A07" ], "review_status": "curator_commentary_not_validated", "boundary": "The priority claim and conflict-free-target wording are not validated. This is a selected later state, not an inspected final publication.", "focus_passages": [ { "evidence_id": "A04", "label": "Earlier introduction", "text": "We introduce \\emph{[modelname]}, an imitation learner for CDCL branching.", "start_character_in_excerpt": 611, "end_character_in_excerpt": 684, "verbatim": true, "verbatim_relative_to": "source_excerpt" }, { "evidence_id": "A07", "label": "Later contribution list", "text": "We propose ImitSAT, the first branching policy for CDCL solvers based on imitation learning.", "start_character_in_excerpt": 80, "end_character_in_excerpt": 172, "verbatim": true, "verbatim_relative_to": "source_excerpt" } ], "reason_evidence": { "status": "analyst_interpretation", "explanation": "A proposed explanation of communicative value, not recovered writer intention.", "evidence_ids": [ "A09", "A02", "A04", "A07" ] }, "short_skill": "Promote a formulation into contribution claims", "skill_ids": [ "WS01", "WS03" ] } }, { "id": "T1-09", "title": "A stronger claim is not automatically a better claim", "primary_skill_id": "WS05", "skill_focus": { "skill_id": "WS05", "label": "Evidence and claim calibration", "subskill": "Audit priority and guarantee language", "evidence_linked_rationale": "The priority and target-quality phrases create specific verification obligations.", "evidence_ids": [ "A07" ], "status": "curator_interpretation_not_validated" }, "reader_goal": { "id": "scrutinize", "purpose": "Tell a clear contribution statement apart from evidence establishing it.", "evidence_ids": [ "A07" ], "kind": "proposed_reader_value", "measured": false }, "evidence_ids": [ "A07" ], "original_instruction": { "status": "not_established_for_this_reading_step", "text": null }, "quality_preference": null, "training_example_status": "not_prepared_or_evaluated", "learning_opportunity": { "kind": "newly_authored_task_sketch_not_historical_prompt", "question": "Identify claims that need priority checks, technical conditions or empirical support without silently deleting source text.", "input_evidence_ids": [ "A07" ], "view": "retrospective_source_inspection", "answers": "Open review question; no gold answer, preference or reward supplied.", "cutoff_safe_training_export": false, "training_or_evaluation_run": false, "evidence_overlap_group": "ImitSAT-single-project" }, "reading_order_in_chapter": 9, "reading_view": { "title": "A stronger claim is not automatically a better claim", "reader_problem": "What evidence does confident contribution wording require?", "plain_context": "The contribution list includes “first” and “conflict-free.” Both are stronger commitments than simply naming the proposed approach.", "observed_development": "The supplied text contains a priority assertion and a technical assertion that a reviewer would need to examine.", "why_this_matters": { "text": "Making a claim clear also makes its evidential burden visible. An attractive writing example should not reward unsupported confidence.", "kind": "analyst_interpretation_of_reader_value", "evidence_ids": [ "A07" ], "measured_reader_gain": false }, "history_value": { "text": "This is a retrospective claim-audit view on one state. The concern is ours; no historical reviewer feedback or resulting correction is supplied.", "kind": "single_state_control", "evidence_ids": [ "A07" ], "endpoint_is_published_final": false, "model_advantage_demonstrated": false }, "evidence_ids": [ "A07" ], "review_status": "curator_commentary_not_validated", "boundary": "Neither exhaustive prior-work coverage nor conflict-free behavior is established. No historical intention to overstate is inferred.", "focus_passages": [ { "evidence_id": "A07", "label": "Priority assertion", "text": "We propose ImitSAT, the first branching policy for CDCL solvers based on imitation learning.", "start_character_in_excerpt": 80, "end_character_in_excerpt": 172, "verbatim": true, "verbatim_relative_to": "source_excerpt" }, { "evidence_id": "A07", "label": "Technical assertion", "text": "These sequences serve as clean, conflict-free training targets and align naturally with prefix-conditioned autoregressive modeling.", "start_character_in_excerpt": 435, "end_character_in_excerpt": 566, "verbatim": true, "verbatim_relative_to": "source_excerpt" } ], "reason_evidence": { "status": "analyst_interpretation", "explanation": "A proposed explanation of communicative value, not recovered writer intention.", "evidence_ids": [ "A07" ] }, "short_skill": "Audit priority and guarantee language", "skill_ids": [ "WS05", "WS01" ] } } ] }, { "id": "T2", "title": "Build the literature argument", "reader_question": "How do organized references become a connected argument for a paper?", "scope": "Four selected related-work stages: organized prose with an empty conceptual heading, a supplied conceptual strand, and a bridge before and after repair. This preview does not include the initial citation scratchpad.", "steps": [ { "id": "T2-02", "title": "Organize references by the questions they answer", "primary_skill_id": "WS02", "skill_focus": { "skill_id": "WS02", "label": "Literature positioning", "subskill": "Organize citations by argumentative role", "evidence_linked_rationale": "The supplied B02 text organizes prior work by role while leaving a separate conceptual heading empty.", "evidence_ids": [ "B02" ], "status": "curator_interpretation_not_validated" }, "reader_goal": { "id": "understand", "purpose": "Understand relationships among cited approaches rather than read an undifferentiated list.", "evidence_ids": [ "B02" ], "kind": "proposed_reader_value", "measured": false }, "evidence_ids": [ "B02", "A08" ], "original_instruction": { "status": "not_established_for_this_reading_step", "text": null }, "quality_preference": null, "training_example_status": "not_prepared_or_evaluated", "learning_opportunity": { "kind": "newly_authored_task_sketch_not_historical_prompt", "question": "Identify the organizing roles in the prose; do not treat cited-work descriptions as verified.", "input_evidence_ids": [ "B02" ], "view": "retrospective_source_inspection", "answers": "Open review question; no gold answer, preference or reward supplied.", "cutoff_safe_training_export": false, "training_or_evaluation_run": false, "evidence_overlap_group": "ImitSAT-single-project" }, "reading_order_in_chapter": 1, "reading_view": { "title": "Organize references by the questions they answer", "reader_problem": "Which prior works describe a model, a dataset, or a component inside the solver?", "plain_context": "This selected draft groups prior work by role: predictions about a problem, data and benchmarks, and learned components inside a solver. A separate learning-method heading has not yet been filled.", "observed_development": "At revision 1576, the related-work section contains thematic prose and an empty imitation-learning heading. This preview begins at that partly completed state.", "why_this_matters": { "text": "The reader can begin to see the landscape and the remaining gap. The empty heading also shows that this intermediate draft is not yet a complete argument.", "kind": "analyst_interpretation_of_reader_value", "evidence_ids": [ "B02" ], "measured_reader_gain": false }, "history_value": { "text": "This intermediate draft exposes an unfinished part of the argument. A later selected state fills the heading; the connected paragraph alone would not establish that order.", "kind": "selected_intermediate_state", "evidence_ids": [ "B02", "A08" ], "endpoint_is_published_final": false, "model_advantage_demonstrated": false }, "evidence_ids": [ "B02", "A08" ], "review_status": "curator_commentary_not_validated", "boundary": "Descriptions of the cited papers remain manuscript assertions; the references themselves were not inspected.", "focus_passages": [ { "evidence_id": "B02", "label": "One organizing sentence", "text": "In parallel with these model-based approaches, complementary efforts target data and benchmarking, including G2SAT~\\citep{you2019g2sat} and G4SATBench~\\citep{lig4satbench}.", "start_character_in_excerpt": 424, "end_character_in_excerpt": 596, "verbatim": true, "verbatim_relative_to": "source_excerpt" }, { "evidence_id": "B02", "label": "Another organizing transition; sentence continues", "text": "Building on these foundations, a second line integrates learning inside solvers to shape specific components:", "start_character_in_excerpt": 597, "end_character_in_excerpt": 706, "verbatim": true, "verbatim_relative_to": "source_excerpt" }, { "evidence_id": "B02", "label": "An unfinished subsection heading", "text": "\\paragraph{Imitation learning.}", "start_character_in_excerpt": 1470, "end_character_in_excerpt": 1501, "verbatim": true, "verbatim_relative_to": "source_excerpt" } ], "reason_evidence": { "status": "analyst_interpretation", "explanation": "A proposed explanation of communicative value, not recovered writer intention.", "evidence_ids": [ "B02" ] }, "short_skill": "Organize citations by argumentative role", "skill_ids": [ "WS02", "WS03" ] } }, { "id": "T2-03", "title": "Add the conceptual strand the argument needs", "primary_skill_id": "WS02", "skill_focus": { "skill_id": "WS02", "label": "Literature positioning", "subskill": "Add a conceptual literature strand", "evidence_linked_rationale": "The earlier empty imitation-learning heading is followed by a selected subsection connecting expert demonstrations, sequence modeling and solver examples.", "evidence_ids": [ "B02", "A08" ], "status": "curator_interpretation_not_validated" }, "reader_goal": { "id": "understand", "purpose": "See the conceptual connection between the method and relevant areas of prior work.", "evidence_ids": [ "B02", "A08" ], "kind": "proposed_reader_value", "measured": false }, "evidence_ids": [ "B02", "A08" ], "original_instruction": { "status": "not_established_for_this_reading_step", "text": null }, "quality_preference": null, "training_example_status": "not_prepared_or_evaluated", "learning_opportunity": { "kind": "newly_authored_task_sketch_not_historical_prompt", "question": "Describe the new strand and its role; identify external claims that require the cited sources.", "input_evidence_ids": [ "B02", "A08" ], "view": "retrospective_source_inspection", "answers": "Open review question; no gold answer, preference or reward supplied.", "cutoff_safe_training_export": false, "training_or_evaluation_run": false, "evidence_overlap_group": "ImitSAT-single-project" }, "reading_order_in_chapter": 2, "reading_view": { "title": "Add the conceptual strand the argument needs", "reader_problem": "What connects the solver literature to the proposed learning formulation?", "plain_context": "The later related-work passage introduces learning from expert demonstrations and its use in decision-making, including optimization solvers.", "observed_development": "The previously empty imitation-learning heading is followed in a later selected state by a substantive conceptual strand.", "why_this_matters": { "text": "The section can now connect a domain-specific problem to a broader learning approach, rather than adding citations without explaining their role.", "kind": "analyst_interpretation_of_reader_value", "evidence_ids": [ "B02", "A08" ], "measured_reader_gain": false }, "history_value": { "text": "The blank heading and later passage show this strand being supplied separately from the solver survey. A later section alone would not establish that sequence.", "kind": "historical_comparison", "evidence_ids": [ "B02", "A08" ], "endpoint_is_published_final": false, "model_advantage_demonstrated": false }, "evidence_ids": [ "B02", "A08" ], "review_status": "curator_commentary_not_validated", "boundary": "This is selected same-file development, not a claim about when the author learned the literature. Cited-work accuracy is unverified.", "focus_passages": [ { "evidence_id": "B02", "label": "Earlier: a heading without content", "text": "\\paragraph{Imitation learning.}", "start_character_in_excerpt": 1470, "end_character_in_excerpt": 1501, "verbatim": true, "verbatim_relative_to": "source_excerpt" }, { "evidence_id": "A08", "label": "Later: the strand has an explanation", "text": "Imitation learning (IL) learns policies directly from expert demonstrations, that is, sequences of states with associated actions~\\citep{osa2018algorithmic,zare2024survey}.", "start_character_in_excerpt": 44, "end_character_in_excerpt": 216, "verbatim": true, "verbatim_relative_to": "source_excerpt" }, { "evidence_id": "A08", "label": "Later: a connection to solvers", "text": "For instance, beyond robotics and games, IL has guided decision-making in exact optimization solvers.", "start_character_in_excerpt": 893, "end_character_in_excerpt": 994, "verbatim": true, "verbatim_relative_to": "source_excerpt" } ], "reason_evidence": { "status": "analyst_interpretation", "explanation": "A proposed explanation of communicative value, not recovered writer intention.", "evidence_ids": [ "B02", "A08" ] }, "short_skill": "Add a conceptual literature strand", "skill_ids": [ "WS02", "WS01" ] } }, { "id": "T2-04", "title": "Make the literature lead somewhere", "primary_skill_id": "WS03", "skill_focus": { "skill_id": "WS03", "label": "Argument and structure", "subskill": "Connect literature limitations to the proposal", "evidence_linked_rationale": "A paper-specific closing bridge is present even though its first sentence is grammatically fragmented.", "evidence_ids": [ "B04" ], "status": "curator_interpretation_not_validated" }, "reader_goal": { "id": "understand", "purpose": "Follow why the literature discussion leads to this proposal.", "evidence_ids": [ "B04" ], "kind": "proposed_reader_value", "measured": false }, "evidence_ids": [ "B04" ], "original_instruction": { "status": "not_established_for_this_reading_step", "text": null }, "quality_preference": null, "training_example_status": "not_prepared_or_evaluated", "learning_opportunity": { "kind": "newly_authored_task_sketch_not_historical_prompt", "question": "Identify the intended argumentative connection separately from its sentence-level defect.", "input_evidence_ids": [ "B04" ], "view": "retrospective_source_inspection", "answers": "Open review question; no gold answer, preference or reward supplied.", "cutoff_safe_training_export": false, "training_or_evaluation_run": false, "evidence_overlap_group": "ImitSAT-single-project" }, "reading_order_in_chapter": 3, "reading_view": { "title": "Make the literature lead somewhere", "reader_problem": "How should prior work connect to this paper's proposal?", "plain_context": "The passage tries to connect limitations of learned solver guidance with inspiration from imitation learning, then describes the proposed policy.", "observed_development": "A gap-to-proposal bridge is visible, but its opening motivation is stranded in a sentence fragment.", "why_this_matters": { "text": "The scientific relationship and the sentence expressing it are separate issues: a useful idea can already be present in an unfinished explanation.", "kind": "analyst_interpretation_of_reader_value", "evidence_ids": [ "B02", "A08", "B04", "B05" ], "measured_reader_gain": false }, "history_value": { "text": "This state preserves an imperfect bridge between the earlier literature strands and the later fluent paragraph. It shows that the relationship precedes the language repair.", "kind": "intermediate_state_in_history", "evidence_ids": [ "B02", "A08", "B04", "B05" ], "endpoint_is_published_final": false, "model_advantage_demonstrated": false }, "evidence_ids": [ "B02", "A08", "B04", "B05" ], "review_status": "curator_commentary_not_validated", "boundary": "The original revision request is absent. Low query cost and performance benefits remain assertions, not verified findings.", "focus_passages": [ { "evidence_id": "B04", "label": "Historical opening, including its sentence fragment", "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.", "start_character_in_excerpt": 0, "end_character_in_excerpt": 218, "verbatim": true, "verbatim_relative_to": "source_excerpt" } ], "reason_evidence": { "status": "analyst_interpretation", "explanation": "A proposed explanation of communicative value, not recovered writer intention.", "evidence_ids": [ "B02", "A08", "B04", "B05" ] }, "short_skill": "Connect literature limitations to the proposal", "skill_ids": [ "WS02", "WS03" ] } }, { "id": "T2-05", "title": "Repair the connection without changing the contribution", "primary_skill_id": "WS07", "skill_focus": { "skill_id": "WS07", "label": "Language and presentation", "subskill": "Repair sentence logic while preserving the account", "evidence_linked_rationale": "The recorded revision joins a fragment to the proposal and separates mechanism and supervision into sentences.", "evidence_ids": [ "B04", "B05" ], "status": "curator_interpretation_not_validated" }, "reader_goal": { "id": "understand", "purpose": "Follow motivation, proposal and mechanism without repairing the sentence mentally.", "evidence_ids": [ "B04", "B05" ], "kind": "proposed_reader_value", "measured": false }, "evidence_ids": [ "B04", "B05" ], "original_instruction": { "status": "not_established_for_this_reading_step", "text": null }, "quality_preference": null, "training_example_status": "not_prepared_or_evaluated", "learning_opportunity": { "kind": "newly_authored_task_sketch_not_historical_prompt", "question": "Compare the two passages: what is repaired, what is reorganized, and which technical claims remain?", "input_evidence_ids": [ "B04", "B05" ], "view": "retrospective_source_inspection", "answers": "Open review question; no gold answer, preference or reward supplied.", "cutoff_safe_training_export": false, "training_or_evaluation_run": false, "evidence_overlap_group": "ImitSAT-single-project" }, "reading_order_in_chapter": 4, "reading_view": { "title": "Repair the connection without changing the contribution", "reader_problem": "Can the revision make the logic readable without inventing a new scientific idea?", "plain_context": "The earlier opening separates motivation from the proposal awkwardly. The later sentence connects them and gives the mechanism its own sentence.", "observed_development": "The sentence fragment becomes a connected motivation-to-proposal statement; the technical account is expressed in clearer units.", "why_this_matters": { "text": "A reader can follow why the proposal is introduced before processing how it works. This is a language-and-argument repair inside an existing frame.", "kind": "analyst_interpretation_of_reader_value", "evidence_ids": [ "B04", "B05" ], "measured_reader_gain": false }, "history_value": { "text": "The before/after pair shows what the revision preserves as well as repairs. Later prose alone would hide the original fragment and could not establish that this was repair rather than a new idea.", "kind": "historical_comparison", "evidence_ids": [ "B04", "B05" ], "endpoint_is_published_final": false, "model_advantage_demonstrated": false }, "evidence_ids": [ "B04", "B05" ], "review_status": "curator_commentary_not_validated", "boundary": "The later wording is an observed alternative, not a uniquely correct answer or proof of improved reader performance.", "focus_passages": [ { "evidence_id": "B04", "label": "Before · B04", "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.", "start_character_in_excerpt": 0, "end_character_in_excerpt": 218, "verbatim": true, "verbatim_relative_to": "source_excerpt" }, { "evidence_id": "B05", "label": "After · B05", "text": "Motivated by the limitations of neural guidance for SAT detailed above, and drawing inspiration from imitation learning, we propose \\modelname{}. This branching policy for CDCL clones a near conflict-free KeyTrace distilled from solver runs.", "start_character_in_excerpt": 0, "end_character_in_excerpt": 241, "verbatim": true, "verbatim_relative_to": "source_excerpt" } ], "reason_evidence": { "status": "analyst_interpretation", "explanation": "A proposed explanation of communicative value, not recovered writer intention.", "evidence_ids": [ "B04", "B05" ] }, "short_skill": "Repair sentence logic while preserving the account", "skill_ids": [ "WS03", "WS04", "WS07" ] } }, { "id": "T2-06", "title": "Keep a tiny edit in proportion", "primary_skill_id": "WS07", "skill_focus": { "skill_id": "WS07", "label": "Language and presentation", "subskill": "Distinguish local polish from substantive change", "evidence_linked_rationale": "The selected endpoints change a demonstrative while preserving the substantive account.", "evidence_ids": [ "B05", "B06" ], "status": "curator_interpretation_not_validated" }, "reader_goal": { "id": "scrutinize", "purpose": "Keep an ordinary expression edit in proportion to major writing decisions.", "evidence_ids": [ "B05", "B06" ], "kind": "proposed_reader_value", "measured": false }, "evidence_ids": [ "B05", "B06" ], "original_instruction": { "status": "not_established_for_this_reading_step", "text": null }, "quality_preference": null, "training_example_status": "not_prepared_or_evaluated", "learning_opportunity": { "kind": "newly_authored_task_sketch_not_historical_prompt", "question": "Classify the observed change without inventing a scientific contribution or a preference label.", "input_evidence_ids": [ "B05", "B06" ], "view": "retrospective_source_inspection", "answers": "Open review question; no gold answer, preference or reward supplied.", "cutoff_safe_training_export": false, "training_or_evaluation_run": false, "evidence_overlap_group": "ImitSAT-single-project" }, "reading_order_in_chapter": 5, "reading_view": { "title": "Keep a tiny edit in proportion", "reader_problem": "Does every version difference deserve a big learning claim?", "plain_context": "One selected phrase changes from “This branching policy” to “The branching policy.” The surrounding account is substantially the same in these excerpts.", "observed_development": "The selected endpoint difference is a local determiner change.", "why_this_matters": { "text": "A credible dataset should retain small edits without presenting each as a new conceptual decision.", "kind": "analyst_interpretation_of_reader_value", "evidence_ids": [ "B05", "B06" ], "measured_reader_gain": false }, "history_value": { "text": "The history establishes the edit, but its standalone trajectory value is limited. This is deliberately a local-expression control, not a headline example.", "kind": "local_edit_control", "evidence_ids": [ "B05", "B06" ], "endpoint_is_published_final": false, "model_advantage_demonstrated": false }, "evidence_ids": [ "B05", "B06" ], "review_status": "curator_commentary_not_validated", "boundary": "The endpoints are not adjacent versions. No preference or reason for the change is established.", "focus_passages": [ { "evidence_id": "B05", "label": "Before · B05", "text": "This branching policy for CDCL clones a near conflict-free KeyTrace distilled from solver runs.", "start_character_in_excerpt": 146, "end_character_in_excerpt": 241, "verbatim": true, "verbatim_relative_to": "source_excerpt" }, { "evidence_id": "B06", "label": "After · B06", "text": "The branching policy for CDCL clones a near conflict-free KeyTrace distilled from solver runs.", "start_character_in_excerpt": 146, "end_character_in_excerpt": 240, "verbatim": true, "verbatim_relative_to": "source_excerpt" } ], "reason_evidence": { "status": "analyst_interpretation", "explanation": "A proposed explanation of communicative value, not recovered writer intention.", "evidence_ids": [ "B05", "B06" ] }, "short_skill": "Distinguish local polish from substantive change", "skill_ids": [ "WS07" ] } } ] }, { "id": "T3", "title": "Explain the evidence and its cost", "reader_question": "What was measured, and why is the chosen query budget justified?", "scope": "Selected timing, conditional interpretation, budget-analysis and cross-section organization passages. Explicit references are distinguished from thematic links. No recovered execution logs or figures are supplied.", "steps": [ { "id": "T3-01", "title": "Define what the timing claim includes", "primary_skill_id": "WS05", "skill_focus": { "skill_id": "WS05", "label": "Evidence and claim calibration", "subskill": "Specify the measurement boundary", "evidence_linked_rationale": "The timing statement identifies start/stop conditions, excluded preprocessing and included inference costs.", "evidence_ids": [ "C03" ], "status": "curator_interpretation_not_validated" }, "reader_goal": { "id": "scrutinize", "purpose": "Know what a reported runtime includes and excludes.", "evidence_ids": [ "C03" ], "kind": "proposed_reader_value", "measured": false }, "evidence_ids": [ "C03" ], "original_instruction": { "status": "not_established_for_this_reading_step", "text": null }, "quality_preference": null, "training_example_status": "not_prepared_or_evaluated", "learning_opportunity": { "kind": "newly_authored_task_sketch_not_historical_prompt", "question": "Extract the stated measurement boundaries and list what would need implementation-level verification.", "input_evidence_ids": [ "C03" ], "view": "retrospective_source_inspection", "answers": "Open review question; no gold answer, preference or reward supplied.", "cutoff_safe_training_export": false, "training_or_evaluation_run": false, "evidence_overlap_group": "ImitSAT-single-project" }, "reading_order_in_chapter": 1, "reading_view": { "title": "Define what the timing claim includes", "reader_problem": "What exactly is included when the manuscript says a method is faster?", "plain_context": "The experiment passage counts model inference but excludes parsing and simplification from its solve-time measurement.", "observed_development": "The timing window, exclusions and model-call allowance are explicitly stated in this selected passage.", "why_this_matters": { "text": "A reader needs these boundaries to interpret a runtime comparison rather than assuming that every kind of cost is included.", "kind": "analyst_interpretation_of_reader_value", "evidence_ids": [ "C03" ], "measured_reader_gain": false }, "history_value": { "text": "This is an earlier main-text snapshot, not the first step of an adjacent appendix edit. It supplies context and a current-text control; the defining revision is not recovered.", "kind": "single_state_control", "evidence_ids": [ "C03" ], "endpoint_is_published_final": false, "model_advantage_demonstrated": false }, "evidence_ids": [ "C03" ], "review_status": "curator_commentary_not_validated", "boundary": "The passage specifies a protocol; it does not verify that the experiment implemented it.", "focus_passages": [ { "evidence_id": "C03", "label": "Timing protocol as stated in the manuscript", "text": "The timer starts when the CDCL solve loop begins and stops when the instance is solved; CNF parsing and simplification are not counted. All model inference costs are included.", "start_character_in_excerpt": 111, "end_character_in_excerpt": 286, "verbatim": true, "verbatim_relative_to": "source_excerpt" } ], "reason_evidence": { "status": "analyst_interpretation", "explanation": "A proposed explanation of communicative value, not recovered writer intention.", "evidence_ids": [ "C03" ] }, "short_skill": "Specify the measurement boundary", "skill_ids": [ "WS05", "WS04" ] } }, { "id": "T3-06", "title": "Separate a comparison rule from proof of fairness", "primary_skill_id": "WS05", "skill_focus": { "skill_id": "WS05", "label": "Evidence and claim calibration", "subskill": "Separate a control from the conclusion it supports", "evidence_linked_rationale": "The protocol equates query counts but gives no per-call cost evidence in this excerpt.", "evidence_ids": [ "C03" ], "status": "curator_interpretation_not_validated" }, "reader_goal": { "id": "scrutinize", "purpose": "See what is actually controlled and what remains unspecified.", "evidence_ids": [ "C03" ], "kind": "proposed_reader_value", "measured": false }, "evidence_ids": [ "C03" ], "original_instruction": { "status": "not_established_for_this_reading_step", "text": null }, "quality_preference": null, "training_example_status": "not_prepared_or_evaluated", "learning_opportunity": { "kind": "newly_authored_task_sketch_not_historical_prompt", "question": "List the stated controls and the additional information needed to assess the compute-budget claim.", "input_evidence_ids": [ "C03" ], "view": "retrospective_source_inspection", "answers": "Open review question; no gold answer, preference or reward supplied.", "cutoff_safe_training_export": false, "training_or_evaluation_run": false, "evidence_overlap_group": "ImitSAT-single-project" }, "reading_order_in_chapter": 2, "reading_view": { "title": "Separate a comparison rule from proof of fairness", "reader_problem": "Does giving two models the same number of calls establish equal computational cost?", "plain_context": "The manuscript assigns three calls to two methods and says this matches compute budgets. It does not supply a full cost measurement in this excerpt.", "observed_development": "A call-count rule and a fairness rationale appear together in the source.", "why_this_matters": { "text": "A reviewer can distinguish the stated protocol from the evidence needed to justify an equal-cost comparison.", "kind": "analyst_interpretation_of_reader_value", "evidence_ids": [ "C03" ], "measured_reader_gain": false }, "history_value": { "text": "This is our retrospective audit of one passage. No historical reviewer objection or ensuing protocol correction is supplied.", "kind": "single_state_control", "evidence_ids": [ "C03" ], "endpoint_is_published_final": false, "model_advantage_demonstrated": false }, "evidence_ids": [ "C03" ], "review_status": "curator_commentary_not_validated", "boundary": "Equal-cost validity is not decided here. This is not evidence that the actual comparison was unfair.", "focus_passages": [ { "evidence_id": "C03", "label": "Comparison as written", "text": "\\textsc{Arbiter} and Graph‑Q‑SAT receive 3 calls per instance to match compute budgets. SATformer adjusts VSIDS variable scores once at initialization.", "start_character_in_excerpt": 287, "end_character_in_excerpt": 438, "verbatim": true, "verbatim_relative_to": "source_excerpt" } ], "reason_evidence": { "status": "analyst_interpretation", "explanation": "A proposed explanation of communicative value, not recovered writer intention.", "evidence_ids": [ "C03" ] }, "short_skill": "Separate a control from the conclusion it supports", "skill_ids": [ "WS05" ] } }, { "id": "T3-07", "title": "The cost trade-off was already in the paper", "primary_skill_id": "WS05", "skill_focus": { "skill_id": "WS05", "label": "Evidence and claim calibration", "subskill": "State the regime of a claimed advantage", "evidence_linked_rationale": "The interpretation explicitly connects potential runtime gains to a savings-versus-overhead condition.", "evidence_ids": [ "C04" ], "status": "curator_interpretation_not_validated" }, "reader_goal": { "id": "scrutinize", "purpose": "Evaluate an advantage together with its stated conditions.", "evidence_ids": [ "C04" ], "kind": "proposed_reader_value", "measured": false }, "evidence_ids": [ "C04" ], "original_instruction": { "status": "not_established_for_this_reading_step", "text": null }, "quality_preference": null, "training_example_status": "not_prepared_or_evaluated", "learning_opportunity": { "kind": "newly_authored_task_sketch_not_historical_prompt", "question": "Extract the condition and the limited-gain regime; distinguish both from independently verified results.", "input_evidence_ids": [ "C04" ], "view": "retrospective_source_inspection", "answers": "Open review question; no gold answer, preference or reward supplied.", "cutoff_safe_training_export": false, "training_or_evaluation_run": false, "evidence_overlap_group": "ImitSAT-single-project" }, "reading_order_in_chapter": 3, "reading_view": { "title": "The cost trade-off was already in the paper", "reader_problem": "When does additional model computation lead to faster solving?", "plain_context": "This earlier main-text passage says saved solver work must exceed the cost of consulting the model.", "observed_development": "At revision 803, the manuscript already states the benefit-versus-overhead condition and describes ranges where the curves are close.", "why_this_matters": { "text": "The explanation makes the runtime claim conditional instead of treating fewer solver operations as automatic speedup.", "kind": "analyst_interpretation_of_reader_value", "evidence_ids": [ "C04", "C05", "C06" ], "measured_reader_gain": false }, "history_value": { "text": "This earlier context matters when reading the later appendix: its clearer benefit–cost sentence is explanation repair, not the first recorded appearance of the trade-off.", "kind": "earlier_context_for_later_revision", "evidence_ids": [ "C04", "C05", "C06" ], "endpoint_is_published_final": false, "model_advantage_demonstrated": false }, "evidence_ids": [ "C04", "C05", "C06" ], "review_status": "curator_commentary_not_validated", "boundary": "The plots and measurements are absent. Historical “Arbiter” wording is preserved; its exact naming relationship is not inferred.", "focus_passages": [ { "evidence_id": "C04", "label": "Condition on the runtime interpretation", "text": "Learning model-based branching introduces query overhead, so wall‑clock gains appear only once propagation savings exceed this cost.", "start_character_in_excerpt": 218, "end_character_in_excerpt": 350, "verbatim": true, "verbatim_relative_to": "source_excerpt" }, { "evidence_id": "C04", "label": "Limited-gain regime in the source", "text": "On smaller variable ranges, the curves are close since the model cost is comparable to the available propagation savings.", "start_character_in_excerpt": 546, "end_character_in_excerpt": 667, "verbatim": true, "verbatim_relative_to": "source_excerpt" } ], "reason_evidence": { "status": "source_stated_rationale", "explanation": "The manuscript itself offers a reason; this is not independent validation.", "evidence_ids": [ "C04", "C05", "C06" ] }, "short_skill": "State the regime of a claimed advantage", "skill_ids": [ "WS05", "WS04" ] } }, { "id": "T3-02", "title": "The appendix needs to say what is gained and paid", "primary_skill_id": "WS04", "skill_focus": { "skill_id": "WS04", "label": "Technical explanation", "subskill": "Diagnose an unclear trade-off", "evidence_linked_rationale": "The earlier sentence merges gain, reduced solver work and query counts instead of separating benefit from cost.", "evidence_ids": [ "C05" ], "status": "curator_interpretation_not_validated" }, "reader_goal": { "id": "understand", "purpose": "Understand why a default is chosen rather than only learn its numerical value.", "evidence_ids": [ "C05" ], "kind": "proposed_reader_value", "measured": false }, "evidence_ids": [ "C05" ], "original_instruction": { "status": "not_established_for_this_reading_step", "text": null }, "quality_preference": null, "training_example_status": "not_prepared_or_evaluated", "learning_opportunity": { "kind": "newly_authored_task_sketch_not_historical_prompt", "question": "Identify the explanatory ambiguity without inventing new experimental results.", "input_evidence_ids": [ "C05" ], "view": "retrospective_source_inspection", "answers": "Open review question; no gold answer, preference or reward supplied.", "cutoff_safe_training_export": false, "training_or_evaluation_run": false, "evidence_overlap_group": "ImitSAT-single-project" }, "reading_order_in_chapter": 4, "reading_view": { "title": "The appendix needs to say what is gained and paid", "reader_problem": "Why choose only a few model calls if more calls can reduce solver work?", "plain_context": "A model call may save work for the solver, but the call itself takes computation. The appendix needs to separate those two sides.", "observed_development": "The early appendix calls this a trade-off, but its sentence mixes reductions in solver work with query counts.", "why_this_matters": { "text": "The rationale for a default setting is hard to assess unless the text names both the benefit and the cost.", "kind": "analyst_interpretation_of_reader_value", "evidence_ids": [ "C05", "C04", "C06" ], "measured_reader_gain": false }, "history_value": { "text": "This draft preserves the explanatory ambiguity that the next selected state rewrites. The earlier main text already stated the cost condition, so the problem here is communication, not discovery.", "kind": "earlier_state_in_a_trajectory", "evidence_ids": [ "C05", "C04", "C06" ], "endpoint_is_published_final": false, "model_advantage_demonstrated": false }, "evidence_ids": [ "C05", "C04", "C06" ], "review_status": "curator_commentary_not_validated", "boundary": "Reported gains and the choice of three calls remain unverified manuscript claims; no causal reason for the revision is recorded.", "focus_passages": [ { "evidence_id": "C05", "label": "Earlier appendix wording", "text": "We treat the gain from the model, which reduces the propagation number and query times, as a trade-off. So to balance the gain and cost, we set a default query budget of 3.", "start_character_in_excerpt": 337, "end_character_in_excerpt": 509, "verbatim": true, "verbatim_relative_to": "source_excerpt" } ], "reason_evidence": { "status": "analyst_interpretation", "explanation": "A proposed explanation of communicative value, not recovered writer intention.", "evidence_ids": [ "C05", "C04", "C06" ] }, "short_skill": "Diagnose an unclear trade-off", "skill_ids": [ "WS04", "WS05" ] } }, { "id": "T3-03", "title": "Separate benefit from cost—without claiming a new result", "primary_skill_id": "WS04", "skill_focus": { "skill_id": "WS04", "label": "Technical explanation", "subskill": "Separate benefit from computational cost", "evidence_linked_rationale": "The recorded revision explicitly contrasts propagation savings with the computation required for additional queries.", "evidence_ids": [ "C05", "C06" ], "status": "curator_interpretation_not_validated" }, "reader_goal": { "id": "scrutinize", "purpose": "Assess the two sides of the stated trade-off.", "evidence_ids": [ "C05", "C06" ], "kind": "proposed_reader_value", "measured": false }, "evidence_ids": [ "C05", "C06" ], "original_instruction": { "status": "not_established_for_this_reading_step", "text": null }, "quality_preference": null, "training_example_status": "not_prepared_or_evaluated", "learning_opportunity": { "kind": "newly_authored_task_sketch_not_historical_prompt", "question": "Explain the change in the trade-off sentence and what it does not establish about the experiment.", "input_evidence_ids": [ "C05", "C06" ], "view": "retrospective_source_inspection", "answers": "Open review question; no gold answer, preference or reward supplied.", "cutoff_safe_training_export": false, "training_or_evaluation_run": false, "evidence_overlap_group": "ImitSAT-single-project" }, "reading_order_in_chapter": 5, "reading_view": { "title": "Separate benefit from cost—without claiming a new result", "reader_problem": "What has to be traded against what?", "plain_context": "Read “reduction in propagations” as less work inside the solver; extra model queries are additional computation. These are different quantities.", "observed_development": "The revised appendix explicitly identifies saved solver work as the benefit and additional queries as the computational cost.", "why_this_matters": { "text": "That distinction makes the stated choice of a default budget interpretable. It does not, by itself, prove that the choice is optimal.", "kind": "analyst_interpretation_of_reader_value", "evidence_ids": [ "C05", "C06", "C04" ], "measured_reader_gain": false }, "history_value": { "text": "The pair records how an existing cost idea becomes clearer in the appendix. The main text at revision 803 had already stated the runtime condition, so this is not evidence of a new experiment or first insight.", "kind": "historical_comparison", "evidence_ids": [ "C05", "C06", "C04" ], "endpoint_is_published_final": false, "model_advantage_demonstrated": false }, "evidence_ids": [ "C05", "C06", "C04" ], "review_status": "curator_commentary_not_validated", "boundary": "No original instruction or experimental figure is available. Reader benefit is proposed, not measured.", "focus_passages": [ { "evidence_id": "C05", "label": "Earlier appendix: mixed explanation", "text": "We treat the gain from the model, which reduces the propagation number and query times, as a trade-off.", "start_character_in_excerpt": 337, "end_character_in_excerpt": 440, "verbatim": true, "verbatim_relative_to": "source_excerpt" }, { "evidence_id": "C06", "label": "Later appendix: benefit and cost separated", "text": "We view the reduction in propagations achieved by the model as a benefit that must be traded off against the computational cost of additional queries.", "start_character_in_excerpt": 1010, "end_character_in_excerpt": 1160, "verbatim": true, "verbatim_relative_to": "source_excerpt" } ], "reason_evidence": { "status": "source_stated_rationale", "explanation": "The manuscript itself offers a reason; this is not independent validation.", "evidence_ids": [ "C05", "C06", "C04" ] }, "short_skill": "Separate benefit from computational cost", "skill_ids": [ "WS04", "WS05" ] } }, { "id": "T3-08", "title": "Put the short answer in the main text", "primary_skill_id": "WS03", "skill_focus": { "skill_id": "WS03", "label": "Argument and structure", "subskill": "Link concise explanation to detailed evidence", "evidence_linked_rationale": "The paragraph connects an analysis summary to an explicit appendix and figure reference.", "evidence_ids": [ "C07" ], "status": "curator_interpretation_not_validated" }, "reader_goal": { "id": "build_on", "purpose": "Locate the intended analysis when checking or extending the reported setting.", "evidence_ids": [ "C07" ], "kind": "proposed_reader_value", "measured": false }, "evidence_ids": [ "C07" ], "original_instruction": { "status": "not_established_for_this_reading_step", "text": null }, "quality_preference": null, "training_example_status": "not_prepared_or_evaluated", "learning_opportunity": { "kind": "newly_authored_task_sketch_not_historical_prompt", "question": "Identify what the main text communicates and what evidence the cited appendix must supply.", "input_evidence_ids": [ "C07" ], "view": "retrospective_source_inspection", "answers": "Open review question; no gold answer, preference or reward supplied.", "cutoff_safe_training_export": false, "training_or_evaluation_run": false, "evidence_overlap_group": "ImitSAT-single-project" }, "reading_order_in_chapter": 6, "reading_view": { "title": "Put the short answer in the main text", "reader_problem": "How can readers find the rationale without losing access to the detailed analysis?", "plain_context": "The main text summarizes the budget sweep and directs readers to an appendix and figure.", "observed_development": "The paragraph links a compact explanation of the default setting to named supporting locations.", "why_this_matters": { "text": "Readers can see the high-level rationale and know where to look for its support.", "kind": "analyst_interpretation_of_reader_value", "evidence_ids": [ "C07" ], "measured_reader_gain": false }, "history_value": { "text": "This is a selected cross-section pointer, not evidence that we recovered the entire contemporaneous appendix or watched the reference being added.", "kind": "supporting_snapshot", "evidence_ids": [ "C07" ], "endpoint_is_published_final": false, "model_advantage_demonstrated": false }, "evidence_ids": [ "C07" ], "review_status": "curator_commentary_not_validated", "boundary": "The referenced plot and full context are unavailable. We do not invent definitions for the named metrics.", "focus_passages": [ { "evidence_id": "C07", "label": "Explicit evidence route", "text": "We further analyze the effect of the query budget in Appendix~\\ref{sec:query-budget}, sweeping from 1 to 10 model calls and also including an all-calls setting in Figure~\\ref{fig:query}.", "start_character_in_excerpt": 26, "end_character_in_excerpt": 212, "verbatim": true, "verbatim_relative_to": "source_excerpt" }, { "evidence_id": "C07", "label": "Connection to the main setting", "text": "This supports our decision to use a small default budget of three queries in the main experiments.", "start_character_in_excerpt": 364, "end_character_in_excerpt": 462, "verbatim": true, "verbatim_relative_to": "source_excerpt" } ], "reason_evidence": { "status": "analyst_interpretation", "explanation": "A proposed explanation of communicative value, not recovered writer intention.", "evidence_ids": [ "C07" ] }, "short_skill": "Link concise explanation to detailed evidence", "skill_ids": [ "WS03", "WS05" ] } }, { "id": "T3-04", "title": "A clearer opening can coexist with an unfinished revision", "primary_skill_id": "WS06", "skill_focus": { "skill_id": "WS06", "label": "Revision judgment and consistency", "subskill": "Recognize a mixed-quality intermediate draft", "evidence_linked_rationale": "An explicit definition coexists with duplicated active explanations in the same state.", "evidence_ids": [ "C06", "C08" ], "status": "curator_interpretation_not_validated" }, "reader_goal": { "id": "understand", "purpose": "Retain a useful clarification while recognizing the need for consolidation.", "evidence_ids": [ "C06", "C08" ], "kind": "proposed_reader_value", "measured": false }, "evidence_ids": [ "C06", "C08" ], "original_instruction": { "status": "not_established_for_this_reading_step", "text": null }, "quality_preference": null, "training_example_status": "not_prepared_or_evaluated", "learning_opportunity": { "kind": "newly_authored_task_sketch_not_historical_prompt", "question": "Identify what should be preserved and what repeats; do not assume later means uniformly better.", "input_evidence_ids": [ "C08" ], "view": "retrospective_source_inspection", "answers": "Open review question; no gold answer, preference or reward supplied.", "cutoff_safe_training_export": false, "training_or_evaluation_run": false, "evidence_overlap_group": "ImitSAT-single-project" }, "reading_order_in_chapter": 7, "reading_view": { "title": "A clearer opening can coexist with an unfinished revision", "reader_problem": "What should a revision keep—and what should it remove?", "plain_context": "The later appendix now defines “query budget” for the reader. But two active explanations of diminishing returns remain.", "observed_development": "Revision 2895 contains both the new definition and repeated active benefit–cost prose. This is not just an old paragraph hidden in comments.", "why_this_matters": { "text": "A revision can supply needed context and still leave repetition. Learning from history should preserve that mixed state rather than label every later draft better.", "kind": "analyst_interpretation_of_reader_value", "evidence_ids": [ "C06", "C08", "C09" ], "measured_reader_gain": false }, "history_value": { "text": "The consolidated later active paragraph does not show this coexistence. The intermediate state supplies a concrete preserve-versus-remove decision that the endpoint alone cannot reconstruct.", "kind": "mixed_intermediate_state", "evidence_ids": [ "C06", "C08", "C09" ], "endpoint_is_published_final": false, "model_advantage_demonstrated": false }, "evidence_ids": [ "C06", "C08", "C09" ], "review_status": "curator_commentary_not_validated", "boundary": "No editing session or intention is recovered. The reported trends are not verified, and the exact intervening edits are missing.", "focus_passages": [ { "evidence_id": "C08", "label": "New explicit definition", "text": "In 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.", "start_character_in_excerpt": 706, "end_character_in_excerpt": 869, "verbatim": true, "verbatim_relative_to": "source_excerpt" }, { "evidence_id": "C08", "label": "First active explanation", "text": "the 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.", "start_character_in_excerpt": 1168, "end_character_in_excerpt": 1360, "verbatim": true, "verbatim_relative_to": "source_excerpt" }, { "evidence_id": "C08", "label": "Second active explanation in the same draft", "text": "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.", "start_character_in_excerpt": 1572, "end_character_in_excerpt": 1759, "verbatim": true, "verbatim_relative_to": "source_excerpt" } ], "reason_evidence": { "status": "analyst_interpretation", "explanation": "A proposed explanation of communicative value, not recovered writer intention.", "evidence_ids": [ "C06", "C08", "C09" ] }, "short_skill": "Recognize a mixed-quality intermediate draft", "skill_ids": [ "WS04", "WS06" ] } }, { "id": "T3-09", "title": "Track the subject of a claim through revision", "primary_skill_id": "WS05", "skill_focus": { "skill_id": "WS05", "label": "Evidence and claim calibration", "subskill": "Preserve the subject and scope of a claim", "evidence_linked_rationale": "Comparing the active endpoint sentences reveals that the claim now names only the comparator.", "evidence_ids": [ "C06", "C08" ], "status": "curator_interpretation_not_validated" }, "reader_goal": { "id": "scrutinize", "purpose": "Know which method a runtime assertion actually concerns.", "evidence_ids": [ "C06", "C08" ], "kind": "proposed_reader_value", "measured": false }, "evidence_ids": [ "C06", "C08" ], "original_instruction": { "status": "not_established_for_this_reading_step", "text": null }, "quality_preference": null, "training_example_status": "not_prepared_or_evaluated", "learning_opportunity": { "kind": "newly_authored_task_sketch_not_historical_prompt", "question": "Identify the scope difference and distinguish active prose from superseded commented wording.", "input_evidence_ids": [ "C06", "C08" ], "view": "retrospective_source_inspection", "answers": "Open review question; no gold answer, preference or reward supplied.", "cutoff_safe_training_export": false, "training_or_evaluation_run": false, "evidence_overlap_group": "ImitSAT-single-project" }, "reading_order_in_chapter": 8, "reading_view": { "title": "Track the subject of a claim through revision", "reader_problem": "Does the runtime statement still apply to the same methods?", "plain_context": "One selected appendix state says runtime worsens for both methods. A later active sentence names only GQSAT, the comparator named in that passage.", "observed_development": "The active scope differs between revisions 2614 and 2895: “both our method and GQSAT” becomes “GQSAT.”", "why_this_matters": { "text": "A stylistic rewrite must not silently restore a broader assertion when the current draft states a narrower one.", "kind": "analyst_interpretation_of_reader_value", "evidence_ids": [ "C06", "C08", "C09" ], "measured_reader_gain": false }, "history_value": { "text": "The endpoints establish a scope change that the later sentence alone cannot date or explain. It already occurred by revision 2895; the consolidation at 2898 must not be credited with introducing it.", "kind": "qualifier_change_across_selected_states", "evidence_ids": [ "C06", "C08", "C09" ], "endpoint_is_published_final": false, "model_advantage_demonstrated": false }, "evidence_ids": [ "C06", "C08", "C09" ], "review_status": "curator_commentary_not_validated", "boundary": "The intervening narrowing edit, its reason and the relevant figure are absent. Narrower does not mean empirically verified.", "focus_passages": [ { "evidence_id": "C06", "label": "Earlier active claim", "text": "Moreover, 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.", "start_character_in_excerpt": 1224, "end_character_in_excerpt": 1387, "verbatim": true, "verbatim_relative_to": "source_excerpt" }, { "evidence_id": "C08", "label": "Later active claim", "text": "Moreover, as shown in Figure~\\ref{fig:wallclock-methods}, GQSAT also exhibit worse wall-clock performance as the query budget increases from 3 to 5.", "start_character_in_excerpt": 1974, "end_character_in_excerpt": 2122, "verbatim": true, "verbatim_relative_to": "source_excerpt" } ], "reason_evidence": { "status": "analyst_interpretation", "explanation": "A proposed explanation of communicative value, not recovered writer intention.", "evidence_ids": [ "C06", "C08", "C09" ] }, "short_skill": "Preserve the subject and scope of a claim", "skill_ids": [ "WS05", "WS06" ] } }, { "id": "T3-05", "title": "Consolidate the explanation without widening the claim", "primary_skill_id": "WS06", "skill_focus": { "skill_id": "WS06", "label": "Revision judgment and consistency", "subskill": "Consolidate without silently broadening a claim", "evidence_linked_rationale": "The later state comments out duplicate prose and keeps the runtime observation specific to GQSAT.", "evidence_ids": [ "C08", "C09" ], "status": "curator_interpretation_not_validated" }, "reader_goal": { "id": "scrutinize", "purpose": "Follow the rationale while retaining the limited scope of its supporting assertion.", "evidence_ids": [ "C08", "C09" ], "kind": "proposed_reader_value", "measured": false }, "evidence_ids": [ "C08", "C09" ], "original_instruction": { "status": "not_established_for_this_reading_step", "text": null }, "quality_preference": null, "training_example_status": "not_prepared_or_evaluated", "learning_opportunity": { "kind": "newly_authored_task_sketch_not_historical_prompt", "question": "Compare active and commented text and check whether the scoped runtime statement is preserved.", "input_evidence_ids": [ "C08", "C09" ], "view": "retrospective_source_inspection", "answers": "Open review question; no gold answer, preference or reward supplied.", "cutoff_safe_training_export": false, "training_or_evaluation_run": false, "evidence_overlap_group": "ImitSAT-single-project" }, "reading_order_in_chapter": 9, "reading_view": { "title": "Consolidate the explanation without widening the claim", "reader_problem": "Can the draft become clearer while preserving its definitions and claim boundaries?", "plain_context": "The next selected appendix state keeps the budget definition, unifies the benefit–cost explanation, and retains the GQSAT-specific runtime statement.", "observed_development": "Duplicate prose is commented out. The remaining active paragraph connects the trade-off to the default setting without restoring “both methods.”", "why_this_matters": { "text": "Good revision includes controlled preservation: remove repetition while keeping the definition and the current claim scope.", "kind": "analyst_interpretation_of_reader_value", "evidence_ids": [ "C08", "C09", "C06" ], "measured_reader_gain": false }, "history_value": { "text": "The sequence exposes an intermediate drafting problem, a separately observed narrowing, and later consolidation. Later active prose does not establish this sequence; retained comments provide some older wording, not the full ordering.", "kind": "historical_comparison", "evidence_ids": [ "C08", "C09", "C06" ], "endpoint_is_published_final": false, "model_advantage_demonstrated": false }, "evidence_ids": [ "C08", "C09", "C06" ], "review_status": "curator_commentary_not_validated", "boundary": "GQSAT’s reported behavior does not by itself establish the best budget for the proposed method. Neither the figures nor a measured reader gain are supplied.", "focus_passages": [ { "evidence_id": "C09", "label": "Consolidated explanation", "text": "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.", "start_character_in_excerpt": 1361, "end_character_in_excerpt": 1570, "verbatim": true, "verbatim_relative_to": "source_excerpt" }, { "evidence_id": "C09", "label": "The default and the retained claim scope", "text": "Based on this trade-off, we adopt a default budget of 3 queries. This choice is further supported by the wall-clock results in Figure~\\ref{fig:wallclock-methods}, where GQSAT’s runtime worsens noticeably as the query budget increases from 3 to 5.", "start_character_in_excerpt": 1572, "end_character_in_excerpt": 1818, "verbatim": true, "verbatim_relative_to": "source_excerpt" } ], "reason_evidence": { "status": "analyst_interpretation", "explanation": "A proposed explanation of communicative value, not recovered writer intention.", "evidence_ids": [ "C08", "C09", "C06" ] }, "short_skill": "Consolidate without silently broadening a claim", "skill_ids": [ "WS06", "WS04", "WS05" ] } }, { "id": "T3-10", "title": "Give the experimental evidence a navigable map", "primary_skill_id": "WS03", "skill_focus": { "skill_id": "WS03", "label": "Argument and structure", "subskill": "Organize evidence across the paper", "evidence_linked_rationale": "The roadmap places setup, principal comparisons and supplementary studies in a navigable structure.", "evidence_ids": [ "C10" ], "status": "curator_interpretation_not_validated" }, "reader_goal": { "id": "build_on", "purpose": "Locate intended evidence paths when inspecting or extending the paper.", "evidence_ids": [ "C10" ], "kind": "proposed_reader_value", "measured": false }, "evidence_ids": [ "C10" ], "original_instruction": { "status": "not_established_for_this_reading_step", "text": null }, "quality_preference": null, "training_example_status": "not_prepared_or_evaluated", "learning_opportunity": { "kind": "newly_authored_task_sketch_not_historical_prompt", "question": "Map the announced analyses to their stated locations; mark which supporting artifacts are unavailable.", "input_evidence_ids": [ "C10" ], "view": "retrospective_source_inspection", "answers": "Open review question; no gold answer, preference or reward supplied.", "cutoff_safe_training_export": false, "training_or_evaluation_run": false, "evidence_overlap_group": "ImitSAT-single-project" }, "reading_order_in_chapter": 10, "reading_view": { "title": "Give the experimental evidence a navigable map", "reader_problem": "How should readers find the studies relevant to each claim?", "plain_context": "A later experimental introduction announces setup, main comparisons and additional analyses, with appendix references.", "observed_development": "The review-macro passage organizes the experimental account by study type and location.", "why_this_matters": { "text": "A roadmap can help readers locate support instead of treating the experiments as a list of disconnected results.", "kind": "analyst_interpretation_of_reader_value", "evidence_ids": [ "C10" ], "measured_reader_gain": false }, "history_value": { "text": "This is a structure example from one selected state. The named studies are not extra source datasets delivered here, and the planning process is not recovered.", "kind": "single_state_control", "evidence_ids": [ "C10" ], "endpoint_is_published_final": false, "model_advantage_demonstrated": false }, "evidence_ids": [ "C10" ], "review_status": "curator_commentary_not_validated", "boundary": "Compiled visibility, most supporting analyses and the C10 history-group timestamp are unavailable.", "focus_passages": [ { "evidence_id": "C10", "label": "Setup and principal comparison", "text": "We first describe the experimental setup, including baselines, solver implementation, datasets, model, and evaluation metrics. We then compare ImitSAT with SATformer and Graph-Q-SAT on random 3-SAT test sets and on a range of structured SAT families.", "start_character_in_excerpt": 91, "end_character_in_excerpt": 341, "verbatim": true, "verbatim_relative_to": "source_excerpt" }, { "evidence_id": "C10", "label": "Supplementary-analysis roadmap", "text": "Finally, we summarize several additional analyses whose full results are deferred to the appendices, including a query budget ablation, a GNN augmented variant, a Top-K masking versus fallback study, and integrations with more advanced solvers", "start_character_in_excerpt": 593, "end_character_in_excerpt": 836, "verbatim": true, "verbatim_relative_to": "source_excerpt" } ], "reason_evidence": { "status": "analyst_interpretation", "explanation": "A proposed explanation of communicative value, not recovered writer intention.", "evidence_ids": [ "C10" ] }, "short_skill": "Organize evidence across the paper", "skill_ids": [ "WS03" ] } } ] } ], "evidence": { "A01": { "id": "A01", "title": "Sequence-formulation context", "source_file": "introduction.tex", "content_kind": "manuscript_prose", "composition_attribution": "Unknown. The supplied records do not establish who or what wrote each passage.", "source_data_synthetic": false, "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.", "source_ref": { "side": "after", "revision_semantics": "Recorded Overleaf revision interval. The revision number identifies a manuscript state, not an author or the time spent editing.", "text_file": "sources/A01.tex", "text_sha256": "3c563569a464806b863e25684389da27e5b06c30bf79f221801f55e670e13456", "revision_start": 253, "revision_end": 256 }, "integrity_check": { "algorithm": "SHA-256", "text_sha256": "3c563569a464806b863e25684389da27e5b06c30bf79f221801f55e670e13456", "character_count": 456 }, "manuscript_revision": 256, "contributor_privacy": { "labels_replaced": false, "method": "No contributor-label substitution was needed in this excerpt.", "marker_ids": [], "marker_note": "No private marker replacement in this excerpt.", "actor_id": null, "authorship_known": false } }, "A09": { "id": "A09", "title": "Earlier expert-derived supervision", "source_file": "introduction.tex", "content_kind": "manuscript_prose", "composition_attribution": "Unknown. The supplied records do not establish who or what wrote each passage.", "source_data_synthetic": false, "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.", "source_ref": { "side": "after", "revision_semantics": "Recorded Overleaf revision interval. The revision number identifies a manuscript state, not an author or the time spent editing.", "text_file": "sources/A09.tex", "text_sha256": "4b72f3dad06de7b0d9f0ee8a7acaa7d6dfdb62d60f6a9dc86fd866b7406455d7", "revision_start": 253, "revision_end": 256 }, "integrity_check": { "algorithm": "SHA-256", "text_sha256": "4b72f3dad06de7b0d9f0ee8a7acaa7d6dfdb62d60f6a9dc86fd866b7406455d7", "character_count": 302 }, "manuscript_revision": 256, "contributor_privacy": { "labels_replaced": false, "method": "No contributor-label substitution was needed in this excerpt.", "marker_ids": [], "marker_note": "No private marker replacement in this excerpt.", "actor_id": null, "authorship_known": false } }, "A02": { "id": "A02", "title": "Explicit positioning note", "source_file": "introduction.tex", "content_kind": "working_note", "composition_attribution": "Unknown. The supplied records do not establish who or what wrote each passage.", "source_data_synthetic": false, "text": "Use imitation learning to compare with graphq", "source_ref": { "side": "after", "revision_semantics": "Recorded Overleaf revision interval. The revision number identifies a manuscript state, not an author or the time spent editing.", "text_file": "sources/A02.tex", "text_sha256": "c576b1f6d1559dd8c497d453c750aebcedcbbc6361f315bd47021ef55b79f88f", "revision_start": 437, "revision_end": 439 }, "integrity_check": { "algorithm": "SHA-256", "text_sha256": "c576b1f6d1559dd8c497d453c750aebcedcbbc6361f315bd47021ef55b79f88f", "character_count": 45 }, "manuscript_revision": 439, "contributor_privacy": { "labels_replaced": false, "method": "No contributor-label substitution was needed in this excerpt.", "marker_ids": [], "marker_note": "No private marker replacement in this excerpt.", "actor_id": null, "authorship_known": false } }, "A03": { "id": "A03", "title": "Technical articulation", "source_file": "Methods.tex", "content_kind": "manuscript_prose", "composition_attribution": "Unknown. The supplied records do not establish who or what wrote each passage.", "source_data_synthetic": false, "text": "\\subsection{Autoregressive Imitation}\n\nA learner is now introduced to imitate the expert keytrace.\nThe CDCL solver requests one branch at a time, so the learner must map the formula and a prefix of the expert trace to the next signed variable under a small computational budget.\nAn Autoregressive (AR) Model approach fits this need, since it conditions on a prefix and predicts the next element in a sequence.\nThe presentation proceeds in three steps.\nWe first build a compact serialization that the learner can read, and then introduce the next decision AR model.\nFinally, we describe online use inside CDCL.\n\n", "source_ref": { "side": "after", "revision_semantics": "Recorded Overleaf revision interval. The revision number identifies a manuscript state, not an author or the time spent editing.", "text_file": "sources/A03.tex", "text_sha256": "74a951effe8fced0298b765c486a50131322cac6c27db7f547b566524556a3a0", "revision_start": 517, "revision_end": 518 }, "integrity_check": { "algorithm": "SHA-256", "text_sha256": "74a951effe8fced0298b765c486a50131322cac6c27db7f547b566524556a3a0", "character_count": 611 }, "manuscript_revision": 518, "contributor_privacy": { "labels_replaced": false, "method": "No contributor-label substitution was needed in this excerpt.", "marker_ids": [], "marker_note": "No private marker replacement in this excerpt.", "actor_id": null, "authorship_known": false } }, "A04": { "id": "A04", "title": "Expert-learning identity", "source_file": "introduction.tex", "content_kind": "manuscript_prose", "composition_attribution": "Unknown. The supplied records do not establish who or what wrote each passage.", "source_data_synthetic": false, "text": "Recent work brings learning into SAT but does not fully match the control flow of CDCL. SATformer~\\citep{shi2023satformer} learns instance-level signals and adjusts initial variable activities, then stops acting inside the branching loop. Graph-Q-SAT ~\\citep{kurin2020can} learns a state-centric agent that is queried online, yet it does not directly model the executed branch sequence, and it forgoes clear decision-level supervision. These gaps raise a simple question. Can we learn from an expert that CDCL itself reveals, one that is nearly conflict-free and that reduces propagation by avoiding detours?\n\n\nWe introduce \\emph{[modelname]}, an imitation learner for CDCL branching. From a full run, we collapse backtracks into a short expert keytrace that retains the surviving decisions. Replaying this keytrace on the same instance is nearly conflict-free and removes redundant propagation, which yields clean training targets.\nThen, the learner reads the instance together with a prefix of the keytrace and predicts the next signed variable as the next branch decision.\nAt solve time, the learner acts as a drop-in branching rule under a small query budget, and the solver falls back to the native heuristic when the model is uncertain. \nAll other parts of CDCL remain unchanged, so completeness and robustness are preserved.\n", "source_ref": { "side": "after", "revision_semantics": "Recorded Overleaf revision interval. The revision number identifies a manuscript state, not an author or the time spent editing.", "text_file": "sources/A04.tex", "text_sha256": "72700532246eb244a7e9adcba26aac21c9ad649f38c3d8d4ae8041df0d603090", "revision_start": 561, "revision_end": 567 }, "integrity_check": { "algorithm": "SHA-256", "text_sha256": "72700532246eb244a7e9adcba26aac21c9ad649f38c3d8d4ae8041df0d603090", "character_count": 1332 }, "manuscript_revision": 567, "contributor_privacy": { "labels_replaced": false, "method": "No contributor-label substitution was needed in this excerpt.", "marker_ids": [], "marker_note": "No private marker replacement in this excerpt.", "actor_id": null, "authorship_known": false } }, "A10": { "id": "A10", "title": "Comparator context at the same manuscript revision 1072 checkpoint", "source_file": "introduction.tex", "content_kind": "review_macro", "composition_attribution": "Unknown. The supplied records do not establish who or what wrote each passage.", "source_data_synthetic": false, "text": "\\reviewmarkA{Classical branching heuristics, however, are hand-crafted and limited in their adaptability. Recent work has sought to improve solver performance by integrating learning-based guidance. For example, SATformer~\\citep{shi2023satformer} learns instance-level signals to adjust variable activities during initialization, but it exerts no influence once the branching loop begins. Graph-Q-SAT~\\citep{kurin2020can} introduces an online agent within CDCL, yet it relies on reinforcement learning (RL), which requires extensive exploration and can be unstable due to sparse rewards and delayed feedback.}", "source_ref": { "side": "after", "revision_semantics": "Recorded Overleaf revision interval. The revision number identifies a manuscript state, not an author or the time spent editing.", "text_file": "sources/A10.tex", "text_sha256": "83f511c00a12b6fc332f445eae23eccd5e4cd6c8b145dabe52487dce82ccc897", "revision_start": 1066, "revision_end": 1072 }, "integrity_check": { "algorithm": "SHA-256", "text_sha256": "83f511c00a12b6fc332f445eae23eccd5e4cd6c8b145dabe52487dce82ccc897", "character_count": 609 }, "manuscript_revision": 1072, "contributor_privacy": { "labels_replaced": true, "method": "A contributor-specific review command was replaced; its scientific text is unchanged.", "marker_ids": [ "review_marker_A" ], "marker_note": "Marker A preserves a repeated markup label only; it does not establish one person or the author of either passage.", "actor_id": null, "authorship_known": false } }, "A06": { "id": "A06", "title": "Explicit IL-versus-RL rationale", "source_file": "introduction.tex", "content_kind": "review_macro", "composition_attribution": "Unknown. The supplied records do not establish who or what wrote each passage.", "source_data_synthetic": false, "text": "\\reviewmarkA{In contrast, we adopt imitation learning, which learns directly from expert traces. By training on KeyTraces, i.e., collapsed sequences of surviving decisions from full solver runs, our approach provides clear, dense supervision at every branching step. This allows the learner to reproduce high-quality decisions without costly exploration, yielding faster convergence, more stable training, and a natural alignment with the prefix-conditioned nature of branching.}", "source_ref": { "side": "after", "revision_semantics": "Recorded Overleaf revision interval. The revision number identifies a manuscript state, not an author or the time spent editing.", "text_file": "sources/A06.tex", "text_sha256": "fbf991cdc3cd81aaac2cab586ce86a8e404891ab9e18cc5cf0046e30c7bcda88", "revision_start": 1066, "revision_end": 1072 }, "integrity_check": { "algorithm": "SHA-256", "text_sha256": "fbf991cdc3cd81aaac2cab586ce86a8e404891ab9e18cc5cf0046e30c7bcda88", "character_count": 479 }, "manuscript_revision": 1072, "contributor_privacy": { "labels_replaced": true, "method": "A contributor-specific review command was replaced; its scientific text is unchanged.", "marker_ids": [ "review_marker_A" ], "marker_note": "Marker A preserves a repeated markup label only; it does not establish one person or the author of either passage.", "actor_id": null, "authorship_known": false } }, "A07": { "id": "A07", "title": "Contribution promoted", "source_file": "introduction.tex", "content_kind": "manuscript_prose", "composition_attribution": "Unknown. The supplied records do not establish who or what wrote each passage.", "source_data_synthetic": false, "text": "We summarize the contributions of this work as below.\n\\begin{itemize}\n \\item We propose ImitSAT, the first branching policy for CDCL solvers based on imitation learning. \nUnlike prior methods that rely on reinforcement learning, ImitaSAT leverages dense, decision-level supervision from expert traces.\n\\item We cast branching as a sequential modeling problem by collapsing solver runs into compact sequences of surviving decisions. These sequences serve as clean, conflict-free training targets and align naturally with prefix-conditioned autoregressive modeling.\n\n\\end{itemize}", "source_ref": { "side": "after", "revision_semantics": "Recorded Overleaf revision interval. The revision number identifies a manuscript state, not an author or the time spent editing.", "text_file": "sources/A07.tex", "text_sha256": "988a3ad275c539741a6d3711213f91016710a32c859d299ea1010089d4c9d3d3", "revision_start": 1547, "revision_end": 1554 }, "integrity_check": { "algorithm": "SHA-256", "text_sha256": "988a3ad275c539741a6d3711213f91016710a32c859d299ea1010089d4c9d3d3", "character_count": 581 }, "manuscript_revision": 1554, "contributor_privacy": { "labels_replaced": false, "method": "No contributor-label substitution was needed in this excerpt.", "marker_ids": [], "marker_note": "No private marker replacement in this excerpt.", "actor_id": null, "authorship_known": false } }, "A08": { "id": "A08", "title": "Literature relationship developed", "source_file": "related works.tex", "content_kind": "manuscript_prose", "composition_attribution": "Unknown. The supplied records do not establish who or what wrote each passage.", "source_data_synthetic": false, "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}.", "source_ref": { "side": "after", "revision_semantics": "Recorded Overleaf revision interval. The revision number identifies a manuscript state, not an author or the time spent editing.", "text_file": "sources/A08.tex", "text_sha256": "241387f79147cf211eb2162a15051cb4946a63ed85f769293b9f0b4c1ed37670", "revision_start": 1583, "revision_end": 1585 }, "integrity_check": { "algorithm": "SHA-256", "text_sha256": "241387f79147cf211eb2162a15051cb4946a63ed85f769293b9f0b4c1ed37670", "character_count": 1281 }, "manuscript_revision": 1585, "contributor_privacy": { "labels_replaced": false, "method": "No contributor-label substitution was needed in this excerpt.", "marker_ids": [], "marker_note": "No private marker replacement in this excerpt.", "actor_id": null, "authorship_known": false } }, "B02": { "id": "B02", "title": "Organize the SAT literature", "source_file": "related works.tex", "content_kind": "manuscript_prose", "composition_attribution": "Unknown. The supplied records do not establish who or what wrote each passage.", "source_data_synthetic": false, "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.}", "source_ref": { "side": "after", "revision_semantics": "Recorded Overleaf revision interval. The revision number identifies a manuscript state, not an author or the time spent editing.", "text_file": "sources/B02.tex", "text_sha256": "214a40d95caf6e9de312dce02260293e33fb53b3aaec31906f4534399b08ffeb", "revision_start": 1570, "revision_end": 1576 }, "integrity_check": { "algorithm": "SHA-256", "text_sha256": "214a40d95caf6e9de312dce02260293e33fb53b3aaec31906f4534399b08ffeb", "character_count": 1501 }, "manuscript_revision": 1576, "contributor_privacy": { "labels_replaced": false, "method": "No contributor-label substitution was needed in this excerpt.", "marker_ids": [], "marker_note": "No private marker replacement in this excerpt.", "actor_id": null, "authorship_known": false } }, "B04": { "id": "B04", "title": "Connect the literature to this paper", "source_file": "related works.tex", "content_kind": "manuscript_prose", "composition_attribution": "Unknown. The supplied records do not establish who or what wrote each passage.", "source_data_synthetic": false, "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.", "source_ref": { "side": "after", "revision_semantics": "Recorded Overleaf revision interval. The revision number identifies a manuscript state, not an author or the time spent editing.", "text_file": "sources/B04.tex", "text_sha256": "d3bd52fcc95805ef9a0bd17792caa3651a840a5d4964738529325687d26e5915", "revision_start": 1591, "revision_end": 1597 }, "integrity_check": { "algorithm": "SHA-256", "text_sha256": "d3bd52fcc95805ef9a0bd17792caa3651a840a5d4964738529325687d26e5915", "character_count": 533 }, "manuscript_revision": 1597, "contributor_privacy": { "labels_replaced": false, "method": "No contributor-label substitution was needed in this excerpt.", "marker_ids": [], "marker_note": "No private marker replacement in this excerpt.", "actor_id": null, "authorship_known": false } }, "B05": { "id": "B05", "title": "Repair the argument flow", "source_file": "related works.tex", "content_kind": "manuscript_prose", "composition_attribution": "Unknown. The supplied records do not establish who or what wrote each passage.", "source_data_synthetic": false, "text": "Motivated by the limitations of neural guidance for SAT detailed above, and drawing inspiration from imitation learning, we propose \\modelname{}. This branching policy for CDCL clones a near conflict-free KeyTrace distilled from solver runs. To achieve this, we cast branching as prefix-conditioned sequence prediction and train an autoregressive next-decision model on compact sequences of surviving decisions. This provides dense, decision-level supervision at low per-query cost. These design choices help reduce propagation and improve wall-clock time under small query budgets.", "source_ref": { "side": "after", "revision_semantics": "Recorded Overleaf revision interval. The revision number identifies a manuscript state, not an author or the time spent editing.", "text_file": "sources/B05.tex", "text_sha256": "03c0a3de2d7905c4da6f2cdb3578908d1bc4a8ba460494c1a441cb9f9f9fb637", "revision_start": 1597, "revision_end": 1605 }, "integrity_check": { "algorithm": "SHA-256", "text_sha256": "03c0a3de2d7905c4da6f2cdb3578908d1bc4a8ba460494c1a441cb9f9f9fb637", "character_count": 582 }, "manuscript_revision": 1605, "contributor_privacy": { "labels_replaced": false, "method": "No contributor-label substitution was needed in this excerpt.", "marker_ids": [], "marker_note": "No private marker replacement in this excerpt.", "actor_id": null, "authorship_known": false } }, "B06": { "id": "B06", "title": "A local wording adjustment", "source_file": "related works.tex", "content_kind": "manuscript_prose", "composition_attribution": "Unknown. The supplied records do not establish who or what wrote each passage.", "source_data_synthetic": false, "text": "Motivated by the limitations of neural guidance for SAT detailed above, and drawing inspiration from imitation learning, we propose \\modelname{}. The branching policy for CDCL clones a near conflict-free KeyTrace distilled from solver runs. To achieve this, we cast branching as prefix-conditioned sequence prediction and train an autoregressive next-decision model on compact sequences of surviving decisions. This provides dense, decision-level supervision at low per-query cost. These design choices help reduce propagation and improve wall-clock time under small query budgets.", "source_ref": { "side": "after", "revision_semantics": "Recorded Overleaf revision interval. The revision number identifies a manuscript state, not an author or the time spent editing.", "text_file": "sources/B06.tex", "text_sha256": "fe1cb43071ff2d15c4e227b9509dec27a6072e300b5c99cbd545b47c4326a8cc", "revision_start": 1714, "revision_end": 1716 }, "integrity_check": { "algorithm": "SHA-256", "text_sha256": "fe1cb43071ff2d15c4e227b9509dec27a6072e300b5c99cbd545b47c4326a8cc", "character_count": 581 }, "manuscript_revision": 1716, "contributor_privacy": { "labels_replaced": false, "method": "No contributor-label substitution was needed in this excerpt.", "marker_ids": [], "marker_note": "No private marker replacement in this excerpt.", "actor_id": null, "authorship_known": false } }, "C03": { "id": "C03", "title": "State what the timer includes", "source_file": "experiments.tex", "content_kind": "manuscript_prose", "composition_attribution": "Unknown. The supplied records do not establish who or what wrote each passage.", "source_data_synthetic": false, "text": "\\subsection{Wall-clock Time}\nTo measure practical impact, end‑to‑end solve time is recorded for each instance. The timer starts when the CDCL solve loop begins and stops when the instance is solved; CNF parsing and simplification are not counted. All model inference costs are included. \\textsc{Arbiter} and Graph‑Q‑SAT receive 3 calls per instance to match compute budgets. SATformer adjusts VSIDS variable scores once at initialization.", "source_ref": { "side": "after", "revision_semantics": "Recorded Overleaf revision interval. The revision number identifies a manuscript state, not an author or the time spent editing.", "text_file": "sources/C03.tex", "text_sha256": "ab4d583e5d6b2dbea4b532fc66a791a1a928cbba93e2166dcb2a6bf3aac83583", "revision_start": 782, "revision_end": 794 }, "integrity_check": { "algorithm": "SHA-256", "text_sha256": "ab4d583e5d6b2dbea4b532fc66a791a1a928cbba93e2166dcb2a6bf3aac83583", "character_count": 438 }, "manuscript_revision": 794, "contributor_privacy": { "labels_replaced": false, "method": "No contributor-label substitution was needed in this excerpt.", "marker_ids": [], "marker_note": "No private marker replacement in this excerpt.", "actor_id": null, "authorship_known": false } }, "C05": { "id": "C05", "title": "Draft the budget analysis", "source_file": "appendix.tex", "content_kind": "manuscript_prose", "composition_attribution": "Unknown. The supplied records do not establish who or what wrote each passage.", "source_data_synthetic": false, "text": "\\section{Query Budget Analysis}\nThis section details our method across different query budgets (1-10) and all model calls.\nAs shown in Figure~\\ref{fig:query}, the performance gain diminishes as the query budget increases. It shows the largest gain from the first 3 queries; later queries, especially after 6 calls, show a smaller gain. \nWe treat the gain from the model, which reduces the propagation number and query times, as a trade-off. So to balance the gain and cost, we set a default query budget of 3. In addition, as shown in Figure~\\ref{fig:wallclock-methods}, both our method and Graph-Q-SAT exhibit worse time performance as the query size increases from 3 to 5.", "source_ref": { "side": "after", "revision_semantics": "Recorded Overleaf revision interval. The revision number identifies a manuscript state, not an author or the time spent editing.", "text_file": "sources/C05.tex", "text_sha256": "876432f79f69a0c284e43364891a872423f927df4f50829ebd80a3f2657c86c0", "revision_start": 2608, "revision_end": 2611 }, "integrity_check": { "algorithm": "SHA-256", "text_sha256": "876432f79f69a0c284e43364891a872423f927df4f50829ebd80a3f2657c86c0", "character_count": 674 }, "manuscript_revision": 2611, "contributor_privacy": { "labels_replaced": false, "method": "No contributor-label substitution was needed in this excerpt.", "marker_ids": [], "marker_note": "No private marker replacement in this excerpt.", "actor_id": null, "authorship_known": false } }, "C06": { "id": "C06", "title": "Separate benefit from cost", "source_file": "appendix.tex", "content_kind": "mixed_prose_and_comments", "composition_attribution": "Unknown. The supplied records do not establish who or what wrote each passage.", "source_data_synthetic": false, "text": "\\section{Query Budget Analysis}\n% This section details our method across different query budgets (1-10) and all model calls.\n% As shown in Figure~\\ref{fig:query}, the performance gain diminishes as the query budget increases. It shows the largest gain from the first 3 queries; later queries, especially after 6 calls, show a smaller gain. \n% We treat the gain from the model, which reduces the propagation number and query times, as a trade-off. So to balance the gain and cost, we set a default query budget of 3. In addition, as shown in Figure~\\ref{fig:wallclock-methods}, both our method and Graph-Q-SAT exhibit worse time performance as the query size increases from 3 to 5.\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.", "source_ref": { "side": "after", "revision_semantics": "Recorded Overleaf revision interval. The revision number identifies a manuscript state, not an author or the time spent editing.", "text_file": "sources/C06.tex", "text_sha256": "798b4f47187458fa64e6c2036d5f4d4f124499b2738336538895cca64bff12fa", "revision_start": 2611, "revision_end": 2614 }, "integrity_check": { "algorithm": "SHA-256", "text_sha256": "798b4f47187458fa64e6c2036d5f4d4f124499b2738336538895cca64bff12fa", "character_count": 1387 }, "manuscript_revision": 2614, "contributor_privacy": { "labels_replaced": false, "method": "No contributor-label substitution was needed in this excerpt.", "marker_ids": [], "marker_note": "No private marker replacement in this excerpt.", "actor_id": null, "authorship_known": false } }, "C08": { "id": "C08", "title": "Define the term; overlapping prose remains", "source_file": "appendix.tex", "content_kind": "mixed_prose_and_comments", "composition_attribution": "Unknown. The supplied records do not establish who or what wrote each passage.", "source_data_synthetic": false, "text": "\\section{Query Budget Analysis}\n\\label{sec:query-budget}\n% This section details our method across different query budgets (1-10) and all model calls.\n% As shown in Figure~\\ref{fig:query}, the performance gain diminishes as the query budget increases. It shows the largest gain from the first 3 queries; later queries, especially after 6 calls, show a smaller gain. \n% We treat the gain from the model, which reduces the propagation number and query times, as a trade-off. So to balance the gain and cost, we set a default query budget of 3. In addition, as shown in Figure~\\ref{fig:wallclock-methods}, both our method and Graph-Q-SAT exhibit worse time performance as the query size increases from 3 to 5.\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\n% We 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}, \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.", "source_ref": { "side": "after", "revision_semantics": "Recorded Overleaf revision interval. The revision number identifies a manuscript state, not an author or the time spent editing.", "text_file": "sources/C08.tex", "text_sha256": "2977a8090fad4c8f08af643bce95278a35db9390add20a7a02cef4799343d91c", "revision_start": 2892, "revision_end": 2895 }, "integrity_check": { "algorithm": "SHA-256", "text_sha256": "2977a8090fad4c8f08af643bce95278a35db9390add20a7a02cef4799343d91c", "character_count": 2122 }, "manuscript_revision": 2895, "contributor_privacy": { "labels_replaced": false, "method": "No contributor-label substitution was needed in this excerpt.", "marker_ids": [], "marker_note": "No private marker replacement in this excerpt.", "actor_id": null, "authorship_known": false } }, "C09": { "id": "C09", "title": "Consolidate and retain the narrowed claim", "source_file": "appendix.tex", "content_kind": "mixed_prose_and_comments", "composition_attribution": "Unknown. The supplied records do not establish who or what wrote each passage.", "source_data_synthetic": false, "text": "\\section{Query Budget Analysis}\n\\label{sec:query-budget}\n% This section details our method across different query budgets (1-10) and all model calls.\n% As shown in Figure~\\ref{fig:query}, the performance gain diminishes as the query budget increases. It shows the largest gain from the first 3 queries; later queries, especially after 6 calls, show a smaller gain. \n% We treat the gain from the model, which reduces the propagation number and query times, as a trade-off. So to balance the gain and cost, we set a default query budget of 3. In addition, as shown in Figure~\\ref{fig:wallclock-methods}, both our method and Graph-Q-SAT exhibit worse time performance as the query size increases from 3 to 5.\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\n% We 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}, \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. \nBased on this trade-off, we adopt a default budget of 3 queries. This choice is further supported by the wall-clock results in Figure~\\ref{fig:wallclock-methods}, where GQSAT’s runtime worsens noticeably as the query budget increases from 3 to 5.\n\n% 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.\n% We view the reduction in propagations achieved by the model as a benefit that must be traded off against the computational cost of additional queries.\n% To balance this trade-off, we set a default query budget of 3.\n% Moreover, as shown in Figure~\\ref{fig:wallclock-methods}, GQSAT also exhibit worse wall-clock performance as the query budget increases from 3 to 5.", "source_ref": { "side": "after", "revision_semantics": "Recorded Overleaf revision interval. The revision number identifies a manuscript state, not an author or the time spent editing.", "text_file": "sources/C09.tex", "text_sha256": "b657f2859c7d6b88481258742093429632df66d1d813a4c23a6e7638b47c91ce", "revision_start": 2895, "revision_end": 2898 }, "integrity_check": { "algorithm": "SHA-256", "text_sha256": "b657f2859c7d6b88481258742093429632df66d1d813a4c23a6e7638b47c91ce", "character_count": 2378 }, "manuscript_revision": 2898, "contributor_privacy": { "labels_replaced": false, "method": "No contributor-label substitution was needed in this excerpt.", "marker_ids": [], "marker_note": "No private marker replacement in this excerpt.", "actor_id": null, "authorship_known": false } }, "C04": { "id": "C04", "title": "Explain when savings exceed overhead", "source_file": "experiments.tex", "content_kind": "manuscript_prose", "composition_attribution": "Unknown. The supplied records do not establish who or what wrote each passage.", "source_data_synthetic": false, "text": "Across random 3‑SAT test sets and structured families, \\textsc{Arbiter} traces the lowest curves under these budgets, which means more instances are solved in less time, as shown in Figure~\\ref{fig:wallclock-methods}.\nLearning model-based branching introduces query overhead, so wall‑clock gains appear only once propagation savings exceed this cost. With 3 calls spent early in the run, \\textsc{Arbiter} already surpasses a native‑branching CDCL on the 100 variable range and on structured families, as seen in Fig.~\\ref{fig:wallclock-minisat}. On smaller variable ranges, the curves are close since the model cost is comparable to the available propagation savings. ", "source_ref": { "side": "after", "revision_semantics": "Recorded Overleaf revision interval. The revision number identifies a manuscript state, not an author or the time spent editing.", "text_file": "sources/C04.tex", "text_sha256": "03e956764af3097f5440bc9b1cf2c2cdb291de669911b25edeee7264bbf5aa20", "revision_start": 799, "revision_end": 803 }, "integrity_check": { "algorithm": "SHA-256", "text_sha256": "03e956764af3097f5440bc9b1cf2c2cdb291de669911b25edeee7264bbf5aa20", "character_count": 668 }, "manuscript_revision": 803, "contributor_privacy": { "labels_replaced": false, "method": "No contributor-label substitution was needed in this excerpt.", "marker_ids": [], "marker_note": "No private marker replacement in this excerpt.", "actor_id": null, "authorship_known": false } }, "C07": { "id": "C07", "title": "Link the analysis into the main paper", "source_file": "experiments.tex", "content_kind": "manuscript_prose", "composition_attribution": "Unknown. The supplied records do not establish who or what wrote each passage.", "source_data_synthetic": false, "text": "\\paragraph{Query budget.}\nWe further analyze the effect of the query budget in Appendix~\\ref{sec:query-budget}, sweeping from 1 to 10 model calls and also including an all-calls setting in Figure~\\ref{fig:query}.\nMRPP and $W_{1\\%}$ improve as the budget increases, with most of the gain coming from the first three calls and only marginal improvements beyond six.\nThis supports our decision to use a small default budget of three queries in the main experiments.", "source_ref": { "side": "after", "revision_semantics": "Recorded Overleaf revision interval. The revision number identifies a manuscript state, not an author or the time spent editing.", "text_file": "sources/C07.tex", "text_sha256": "56676695878ffe70ef69389836938efde6da54fbab95ee143647a348a2747b63", "revision_start": 2717, "revision_end": 2727 }, "integrity_check": { "algorithm": "SHA-256", "text_sha256": "56676695878ffe70ef69389836938efde6da54fbab95ee143647a348a2747b63", "character_count": 462 }, "manuscript_revision": 2727, "contributor_privacy": { "labels_replaced": false, "method": "No contributor-label substitution was needed in this excerpt.", "marker_ids": [], "marker_note": "No private marker replacement in this excerpt.", "actor_id": null, "authorship_known": false } }, "C10": { "id": "C10", "title": "Place the analysis in the experimental account", "source_file": "experiments.tex", "content_kind": "review_macro", "composition_attribution": "Unknown. The supplied records do not establish who or what wrote each passage.", "source_data_synthetic": false, "text": "\\rev{\nIn this section, we evaluate ImitSAT as a learned branching policy for CDCL solvers. We first describe the experimental setup, including baselines, solver implementation, datasets, model, and evaluation metrics. We then compare ImitSAT with SATformer and Graph-Q-SAT on random 3-SAT test sets and on a range of structured SAT families. Wall clock behaviour is studied both in the main plots and through a direct comparison with pure MiniSAT in Appendix~\\ref{sec:wall-clock-minisat}, and we also examine generalization to industrial benchmarks from SATCOMP in Appendix~\\ref{sec:satcopm}. Finally, we summarize several additional analyses whose full results are deferred to the appendices, including a query budget ablation, a GNN augmented variant, a Top-K masking versus fallback study, and integrations with more advanced solvers such as CaDiCaL~\\citep{BiereFallerFazekasFleuryFroleyks-CAV24} and Kissat~\\citep{BiereFallerFazekasFleuryFroleyksPollitt-SAT-Competition-2024-solvers}, detailed in Appendices~\\ref{sec:query-budget}, \\ref{sec:gnn}, \\ref{sec:topk}, and \\ref{sec:more-solvers}.}", "source_ref": { "side": "after", "revision_semantics": "Recorded Overleaf revision interval. The revision number identifies a manuscript state, not an author or the time spent editing.", "text_file": "sources/C10.tex", "text_sha256": "ab9cc938656e502f822465042f022b2b7b57d32f52371e387445af9c1784fcff", "revision_start": 2953, "revision_end": 2954 }, "integrity_check": { "algorithm": "SHA-256", "text_sha256": "ab9cc938656e502f822465042f022b2b7b57d32f52371e387445af9c1784fcff", "character_count": 1095 }, "manuscript_revision": 2954, "contributor_privacy": { "labels_replaced": false, "method": "No contributor-label substitution was needed in this excerpt.", "marker_ids": [], "marker_note": "No private marker replacement in this excerpt.", "actor_id": null, "authorship_known": false } } }, "plain_language_context": { "overview": "The selected manuscript describes a learned component that suggests decisions inside a solver. The writing questions concern how to position that component, explain its relationship to prior work, and account for the computational cost of consulting it.", "status": "Source-grounded editorial paraphrase, not replacement source text or independent technical verification.", "glossary": [ { "term": "Branching decision", "explanation": "A next decision requested by the solver; the selected method passage describes predicting one at a time.", "evidence_ids": [ "A03" ] }, { "term": "Expert trace / KeyTrace", "explanation": "The manuscript describes a shortened sequence of solver decisions used to form example next-decision targets. Near-conflict-free properties are source claims.", "evidence_ids": [ "A09", "A06" ] }, { "term": "Imitation learning", "explanation": "In these passages, learning choices from expert demonstrations rather than only naming a model architecture.", "evidence_ids": [ "A08", "A06" ] }, { "term": "Query budget", "explanation": "How many times the solver is allowed to consult the model during search.", "evidence_ids": [ "C08" ] }, { "term": "Propagation savings", "explanation": "For this writing example, less work inside the solver. The source contrasts this benefit with the cost of model queries.", "evidence_ids": [ "C04", "C06" ] }, { "term": "Wall-clock time", "explanation": "Elapsed solve time under the stated measurement window. In the selected protocol, model inference is counted; parsing and simplification are excluded.", "evidence_ids": [ "C03" ] }, { "term": "Historical names and placeholders", "explanation": "The excerpts contain ImitSAT, ImitaSAT, Arbiter, [modelname], \\modelname{}, graphq, Graph-Q-SAT and GQSAT. Native spelling is preserved. Do not infer every naming equivalence or silently standardize it.", "evidence_ids": [ "A02", "A04", "A07", "C03", "C06" ] } ] }, "trajectory_cases": [ { "id": "H1", "chapter_id": "T1", "title": "From an existing method to an explicit contribution", "plain_setup": "The technical account already describes learning next decisions from solver examples. The writing problem is how to make that contribution recognizable and justify its relationship to alternatives.", "milestones": [ { "label": "Supervision is described", "revision": 256, "evidence_ids": [ "A01", "A09" ] }, { "label": "A note requests the comparison", "revision": 439, "evidence_ids": [ "A02" ] }, { "label": "The introduction names the formulation", "revision": 567, "evidence_ids": [ "A04" ] }, { "label": "Review text develops a rationale", "revision": 1072, "evidence_ids": [ "A10", "A06" ] }, { "label": "The contribution list foregrounds it", "revision": 1554, "evidence_ids": [ "A07" ] } ], "later_text_only": { "evidence_ids": [ "A07" ], "can_show": "The paper claims an imitation-learning contribution with expert-derived supervision.", "cannot_establish": "That stepwise targets were already described earlier, that a comparison note intervened, or how the framing developed across these recorded states." }, "historical_information": "The order distinguishes development of a scientific account from invention of the underlying method.", "why_it_matters": "An AI writing collaborator should preserve the technical account while improving its positioning, rather than manufacture novelty or unsupported comparative advantages.", "reason_status": "Only the working note at revision 439 is a direct recorded request. Later rationale is expressed in manuscript text; historical motivation and causality remain unknown.", "hard_question": { "question": "Did the comparison note precede the earliest supplied next-decision supervision description?", "answer": "No. The supervision description is in revision 256; the note is in revision 439. This is a claim about the supplied text, not the date of invention.", "evidence_ids": [ "A09", "A02" ] }, "core_step_ids": [ "T1-01", "T1-02", "T1-04", "T1-05", "T1-06" ], "task_id": "WA1", "limitations": [ "Selected introduction states and paired review-macro passages; no complete intermediate revisions or published final paper.", "No established algorithm switch, implementation change, per-action authorship or causal effect of the note." ] }, { "id": "H2", "chapter_id": "T2", "title": "From an unfinished structure to a connected argument", "plain_setup": "A paper needs more than a list of relevant references. The selected history shows solver-related work and a learning-method strand being organized and connected to the proposal.", "milestones": [ { "label": "Roles emerge; a heading stays empty", "revision": 1576, "evidence_ids": [ "B02" ] }, { "label": "The conceptual strand is filled", "revision": 1585, "evidence_ids": [ "A08" ] }, { "label": "An imperfect bridge appears", "revision": 1597, "evidence_ids": [ "B04" ] }, { "label": "The bridge is repaired", "revision": 1605, "evidence_ids": [ "B05" ] } ], "later_text_only": { "evidence_ids": [ "B05", "B06" ], "can_show": "A paragraph connects limitations of solver guidance and inspiration from imitation learning to the proposal.", "cannot_establish": "The earlier empty conceptual heading, the order in which the two strands were supplied, or the fragmented opening that was repaired." }, "historical_information": "The selected draft has an unfinished conceptual strand. Later records supply it and repair a connection to the proposal while retaining the underlying relationship.", "why_it_matters": "A writing model can be asked to connect prior work to a proposal rather than merely supply citations or imitate the surface of a finished paragraph.", "reason_status": "No original revision instruction is recovered. The writing-function interpretation is new analyst commentary.", "hard_question": { "question": "Is the imitation-learning strand filled in the first supplied organized-prose state?", "answer": "The supplied revision 1576 has an empty imitation-learning heading. A later selected revision 1585 contains the strand. Earlier unseen material cannot be ruled out.", "evidence_ids": [ "B02", "A08" ] }, "core_step_ids": [ "T2-02", "T2-03", "T2-04", "T2-05" ], "task_id": "WA2", "limitations": [ "Descriptions of cited work were not checked against the papers.", "These are selected states, not complete adjacent edit or reading-process telemetry.", "The initial citation scratchpad is not included in this preview; no inference about its contents is required." ] }, { "id": "H3", "chapter_id": "T3", "title": "Clarify a trade-off without broadening the evidence", "plain_setup": "The model can save solver work, but consulting it costs computation. The appendix must explain the trade-off, define the budget and preserve exactly which method a runtime assertion concerns.", "milestones": [ { "label": "The appendix explanation is ambiguous", "revision": 2611, "evidence_ids": [ "C05" ] }, { "label": "Benefit and cost are separated", "revision": 2614, "evidence_ids": [ "C06" ] }, { "label": "A definition, duplication and narrower scope coexist", "revision": 2895, "evidence_ids": [ "C08" ] }, { "label": "The active explanation is consolidated", "revision": 2898, "evidence_ids": [ "C09" ] } ], "earlier_context": { "evidence_ids": [ "C04" ], "note": "Revision 803 in the main text already states that savings must exceed model overhead. The appendix sequence is not evidence of first discovering the trade-off." }, "later_text_only": { "evidence_ids": [ "C09" ], "text_view": "active_prose_excluding_full_line_comments", "native_comment_boundary": "The complete later source retains some old wording in comments. The contrast is with its active explanation, not a claim that its full native source contains no historical clues.", "can_show": "A defined query budget, a consolidated benefit–cost explanation, a default of three calls and a GQSAT-specific runtime claim.", "cannot_establish": "The historical order of the wording changes, the coexistence of two active explanations in revision 2895, or that the claim had already narrowed before consolidation." }, "historical_information": "Separate changes can be inspected: explanation repair, a mixed-quality state, a scope difference, and later consolidation.", "why_it_matters": "A writing collaborator should improve coherence without resurrecting an older broader claim. It also needs to distinguish a readable rationale from adequate experimental support.", "reason_status": "The manuscript states trade-off rationales. The actual reasons for editing or narrowing are not recovered; no performance trend is independently verified.", "hard_question": { "question": "Did the last consolidation introduce the GQSAT-only scope?", "answer": "No. The active sentence is already GQSAT-only in revision 2895 and is retained in 2898. An earlier selected state at 2614 refers to both methods.", "evidence_ids": [ "C06", "C08", "C09" ] }, "core_step_ids": [ "T3-02", "T3-03", "T3-04", "T3-09", "T3-05" ], "task_id": "WA3", "limitations": [ "The exact narrowing edit and referenced figures are missing.", "Consolidation can be attempted from the current draft alone; history advantage is a testable question, not an assumption.", "The comparator's reported behavior alone does not establish an optimal budget for the proposed method." ] } ], "skill_profiles": [ { "id": "WS01", "label": "Scientific framing", "reader_purpose": "Make the contribution recognizable and its significance assessable.", "primary_step_ids": [ "T1-01", "T1-02", "T1-04", "T1-06" ], "related_step_ids": [ "T1-01", "T1-02", "T1-03", "T1-04", "T1-05", "T1-06", "T1-09", "T2-03" ], "related_card_count": 8, "coverage_status": "selected_examples_not_validated", "demonstrated_source_coverage": "Contribution identity, comparison framing and contribution-list emphasis.", "not_established": "No complete account of novelty assessment, problem selection, or idea invention.", "counts_overlap": true }, { "id": "WS02", "label": "Literature positioning", "reader_purpose": "Help a reader see why prior work matters to the present argument.", "primary_step_ids": [ "T1-05", "T2-02", "T2-03" ], "related_step_ids": [ "T1-02", "T1-05", "T2-02", "T2-03", "T2-04" ], "related_card_count": 5, "coverage_status": "selected_examples_not_validated", "demonstrated_source_coverage": "Reference organization, the added conceptual strand and paper-specific comparison.", "not_established": "Cited literature is not independently checked; notes are insufficient to reconstruct all later claims.", "counts_overlap": true }, { "id": "WS03", "label": "Argument and structure", "reader_purpose": "Make the account coherent and navigable.", "primary_step_ids": [ "T1-07", "T2-04", "T3-08", "T3-10" ], "related_step_ids": [ "T1-07", "T1-05", "T1-06", "T2-02", "T2-04", "T2-05", "T3-08", "T3-10" ], "related_card_count": 8, "coverage_status": "selected_examples_not_validated", "demonstrated_source_coverage": "Selected evidence includes an incomplete section structure, a method roadmap, and experimental organization.", "not_established": "Not a complete record of outline planning, the whole manuscript, or all revision sessions.", "counts_overlap": true }, { "id": "WS04", "label": "Technical explanation", "reader_purpose": "Make methods and their conditions understandable and usable.", "primary_step_ids": [ "T1-03", "T1-08", "T3-02", "T3-03" ], "related_step_ids": [ "T1-03", "T1-07", "T1-04", "T1-08", "T2-05", "T3-01", "T3-07", "T3-02", "T3-03", "T3-04", "T3-05" ], "related_card_count": 11, "coverage_status": "selected_examples_not_validated", "demonstrated_source_coverage": "Input/output explanation, bounded integration, term definition and benefit–cost exposition.", "not_established": "No full equation, algorithm, implementation or reproducibility audit.", "counts_overlap": true }, { "id": "WS05", "label": "Evidence and claim calibration", "reader_purpose": "Let readers judge what the evidence supports and where it stops.", "primary_step_ids": [ "T1-09", "T3-01", "T3-06", "T3-07", "T3-09" ], "related_step_ids": [ "T1-08", "T1-09", "T3-01", "T3-06", "T3-07", "T3-02", "T3-03", "T3-08", "T3-09", "T3-05" ], "related_card_count": 10, "coverage_status": "selected_examples_not_validated", "demonstrated_source_coverage": "Measurement scope, comparison controls, conditional performance and priority/claim review.", "not_established": "No independently verified experiments, novelty review or measured reader benefit.", "counts_overlap": true }, { "id": "WS06", "label": "Revision judgment and consistency", "reader_purpose": "Improve a document without losing valid content or introducing contradictions.", "primary_step_ids": [ "T3-04", "T3-05" ], "related_step_ids": [ "T3-04", "T3-09", "T3-05" ], "related_card_count": 3, "coverage_status": "selected_examples_not_validated", "demonstrated_source_coverage": "Mixed-quality drafting, duplication, consolidation and claim-scope preservation.", "not_established": "No complete reviewer-response cycle, original AI dialogue or causal revision intent.", "counts_overlap": true }, { "id": "WS07", "label": "Language and presentation", "reader_purpose": "Improve readability and precision without distorting the research.", "primary_step_ids": [ "T2-05", "T2-06" ], "related_step_ids": [ "T2-05", "T2-06" ], "related_card_count": 2, "coverage_status": "selected_examples_not_validated", "demonstrated_source_coverage": "Sentence repair, explanation segmentation and a narrow local-expression control.", "not_established": "No full figure/layout history or independent linguistic-quality rating.", "counts_overlap": true } ], "writing_action_tasks": { "input_artifact": "writing-task-inputs.json", "reviewer_artifact": "writing-task-review.json", "tasks": [ { "id": "WA1", "case_id": "H1", "title": "Make the formulation explicit without manufacturing novelty", "action": "Revise scientific positioning", "skill_ids": [ "WS01", "WS05", "WS06" ], "history_advantage_status": "The supplied history establishes a pre-existing account and a real comparison request. Improvement over a compact current-context prompt is unmeasured." }, { "id": "WA2", "case_id": "H2", "title": "Connect literature strands without rewriting the science", "action": "Draft an argument bridge", "skill_ids": [ "WS02", "WS03", "WS07", "WS05" ], "history_advantage_status": "The older related-work context may help semantic integration. Local fragment repair is also possible from the current paragraph alone." }, { "id": "WA3", "case_id": "H3", "title": "Consolidate a mixed draft and preserve the current scope", "action": "Revise without regression", "skill_ids": [ "WS04", "WS05", "WS06", "WS07" ], "history_advantage_status": "This is a useful current-draft revision task and a test of harmful reuse from older context. It is not assumed to require history or benefit from more history." } ], "source_role": "Derived uses of the same source records, not additional historical sessions or independent samples." }, "preprocessing": { "description": "Known private contributor labels have been replaced with generic review markers. Scientific wording, citations, comments and historical spellings are preserved.", "paper_identity_retained": true, "scientific_citations_retained": true, "authorship": "A repeated review marker does not establish a person or who wrote a passage.", "chronology": "Recorded manuscript revisions and selected ordering are retained. No time spent or editing duration is inferred.", "privacy_limit": "The paper remains identifiable. Replacing known contributor labels is not a guarantee that every possible identifier has been found." }, "schema": "project-experience.writing-sample.v1", "version": "1.0", "created_date": "2026-09-16", "status": "sample_for_discussion", "origin": { "description": "Selected writing history of ImitSAT, an AI research paper.", "coverage": "Selected passages and writing questions, not the complete manuscript history.", "verification_scope": "Text has been checked against the supplied excerpt records, not the original complete manuscript archive." }, "release": { "purpose": "Discussion of a possible data partnership.", "notice": "For discussing a possible data partnership. Training, redistribution and other uses require a separate agreement." }, "counts": { "independent_projects": 1, "developmental_threads": 3, "source_derived_excerpts": 21, "distinct_file_revision_states": 19, "source_files": 5, "review_questions": 24, "guided_reading_steps": 14, "overlapping_skill_views": 7, "optional_constructed_tasks": 3, "counting_note": "Counts describe overlapping views of one selected manuscript. Reading questions are not independent edits or training samples; yield per paper is not established." }, "scope": { "included_evidence_ids": [ "A01", "A09", "A02", "A03", "A04", "A10", "A06", "A07", "A08", "B02", "B04", "B05", "B06", "C03", "C05", "C06", "C08", "C09", "C04", "C07", "C10" ], "included_evidence_count": 21, "reading_step_count": 24, "included_material": "Selected manuscript passages, working notes, revisions, commentary, and optional writing tasks.", "not_included": [ "Published final paper", "Full manuscript history", "Original AI conversations", "Verified per-action authorship", "Referenced figures or experiment logs", "Model outputs or performance results" ], "preview_omission": "The argument history starts with a partly organized draft. Earlier citation notes are not included." }, "limits": [ "Source statements, including scientific priority and performance claims, are not independently verified.", "Skill associations and potential reader value are curator interpretations; they have not been independently validated.", "Historical revisions are observations, not automatically good targets.", "Chronological order does not establish authorship, editing intent, causal influence, or time spent.", "Human–AI collaboration and broader research/development coverage are purposes, not recovered activities in this sample.", "Replacing known private contributor labels does not make the paper anonymous or guarantee that every possible identifier has been found." ], "reading_paths": { "guided": { "step_ids": [ "T1-01", "T1-02", "T1-04", "T1-05", "T1-06", "T2-02", "T2-03", "T2-04", "T2-05", "T3-02", "T3-03", "T3-04", "T3-09", "T3-05" ], "label": "Follow three histories" }, "all": { "step_ids": [ "T1-01", "T1-02", "T1-03", "T1-07", "T1-04", "T1-08", "T1-05", "T1-06", "T1-09", "T2-02", "T2-03", "T2-04", "T2-05", "T2-06", "T3-01", "T3-06", "T3-07", "T3-02", "T3-03", "T3-08", "T3-04", "T3-09", "T3-05", "T3-10" ], "label": "All 24 writing questions" } } }