[
  {
    "role": "system",
    "content": "你是 Novel2Video 项目的 DeepSeek v4 pro thinking/max 第三方咨询 agent。你只提供专家建议，不是 operator production acceptance、provider/manual evidence、accepted overlay、canonical KG、raw KG 或 replay authority。当前任务是后 P计划主线方案优化，不推进实现，不执行 provider/media/manual evidence。请严格输出 JSON，不要 markdown，不要解释性前后缀。建议必须通用、工程可落地、可验证，不能按作品名、实体名、样本文本或书名写 runtime branching。"
  },
  {
    "role": "user",
    "content": "请基于以下输入，优化 Now-32 之后的候选主线方案。重点判断 pipeline 如何利用 KG/entity links、reference assets、production packet lineage 与 DeepSeek v4 pro thinking/max，形成 direct strong 无法稳定替代的生产系统优势。请给出 Now-32/33/34/35/36 的候选主线、边界、工件、gate、DeepSeek 角色、风险与控制面更新建议。必须保持 next_entrypoint 不变，不能建议立即进入 provider execution。\n\nINPUT_JSON:\n{\n  \"version\": \"post_p_plan_now32_36_strategy_deepseek_consultation_input.v1\",\n  \"created_at\": \"2026-06-05T05:18:34.386430Z\",\n  \"task\": \"Optimize post-P计划 mainlines without advancing implementation; keep Now-32 active route and plan candidate Now-33/34/35/36 lines.\",\n  \"operator_position\": {\n    \"p_plan_closed\": true,\n    \"do_not_advance_project_implementation_now\": true,\n    \"continue_using_deepseek_as_third_party_consultant\": true,\n    \"deepseek_advice_may_substitute_for_human_expert_advice_but_not_evidence_or_acceptance\": true,\n    \"pipeline_outputs_are_good_but_have_not_maximized_pipeline_advantage_or_deepseek_v4_limits\": true,\n    \"pipeline_unique_advantage\": \"KG/entity classification/entity links can bind screenplay shots to stable entities, locations, props, motifs, reference assets, and downstream provider prompts; direct strong generation cannot guarantee this traceable production graph.\",\n    \"industry_benchmark_docs\": [\n      \"docs/resource\"\n    ]\n  },\n  \"active_control_plane\": {\n    \"next_entrypoint\": {\n      \"id\": \"platform_neutral_prompt_readiness_manual_execution_log_contract\",\n      \"title\": \"Prompt Readiness QA And Manual Execution Log Contract\",\n      \"owner_layer\": \"Pipeline Orchestrator / Video Prompt Packet & Adapter Render Agent / Video Generation Agent / Provider Adapter & Payload Agent / QA & Consistency Agent / Visualization & Review UI Agent\",\n      \"why_now\": \"Now-32 remains the active evidence-gated route; P计划 V2 creative-quality work is a sidecar and does not authorize provider/manual/feedback execution.\"\n    },\n    \"primary_goal\": \"Keep Now-32 as the evidence-gated active route after P计划 V2 final closeout. P计划 is closed as an implementation umbrella after P2V-24; future provider/manual/feedback/KG/overlay/replay/output-index work requires an explicit Now-32 or downstream evidence contract and real evidence.\",\n    \"current_implementation_priority\": \"Now-32_evidence_gate_active_after_P计划_closeout_no_provider_execution_authorized\",\n    \"current_p_plan_stage\": \"P计划_V2_closed_after_P2V24_final_closeout\",\n    \"now32_post_p_plan_recalibration\": {\n      \"active_route\": \"platform_neutral_prompt_readiness_manual_execution_log_contract\",\n      \"current_sidecar\": \"P计划_V2_closed_after_P2V24_final_closeout\",\n      \"next_recommended_entrypoint\": \"Now-32_evidence_gate_or_future_explicit_contract\",\n      \"provider_manual_feedback_boundary\": \"blocked until explicit evidence/manual/provider contract\",\n      \"report_target\": \"https://gongju.gold/novel2video/report\",\n      \"p_plan_closed\": true\n    },\n    \"blocked_by\": [\n      \"P计划 implementation umbrella is closed; no further P计划 substage is open.\",\n      \"The 240 held rows and chapter 18 held-only status remain outside complete readiness until explicit QA/repair/evidence contract.\",\n      \"Production acceptance, provider/media execution, manual/provider/operator evidence, feedback/replay, canonical KG/raw KG, accepted/operator overlay writeback, adapter render, outputs index mutation, and old B7 continuation remain blocked until explicit downstream/evidence contract.\"\n    ],\n    \"operator_runtime_preferences\": {\n      \"background_wait_poll_interval_minutes\": 10,\n      \"scope\": \"background commands and long-running provider-independent tasks that require waiting, where tooling permits\",\n      \"boundary\": \"runtime preference only; does not authorize provider execution, media generation, manual/provider/operator evidence, feedback intake, KG/overlay/replay, outputs index mutation, or production acceptance\"\n    },\n    \"phase_objective\": \"Keep Now-32 evidence-gated after P计划 V2 final closeout; P计划 is closed as an implementation umbrella and no provider/manual/feedback execution is authorized without explicit evidence contract.\",\n    \"workflow_index_classification_rules\": [\n      \"Active route comes only from execution_state.json.next_entrypoint.\",\n      \"Now-32 is the formal active route authority and provider/manual evidence gate; it is not a P计划 substage.\",\n      \"P计划 V2 creative-quality redesign sidecar is closed after P2V-24 final closeout.\",\n      \"Future work proceeds only through Now-32 evidence gate or a future explicit contract; no P计划 substage is open.\",\n      \"B计划/北辰计划 foundation/entity closure is complete as a gate state.\",\n      \"DeepSeek v4 pro thinking/max may substitute for human advice/expert consultation under operator delegation, but not production acceptance or evidence writeback.\",\n      \"Do not create new md unless artifact-backed contract, retrospective, decision record, report, or handoff is required.\",\n      \"Detailed history belongs in workflow.json and patchnote.md, not this index.\"\n    ]\n  },\n  \"p_plan_final_metrics\": {\n    \"p2v24_run\": \"outputs/p_plan_v2_24_p_plan_final_closeout_20260605_0742\",\n    \"chapter_packages\": 20,\n    \"neutral_draft_rows\": 1638,\n    \"held_rows\": 240,\n    \"fully_ready_partial_or_held_chapters\": \"13/7\",\n    \"debt_register_rows\": 247,\n    \"chapter_18_status\": \"held_only_no_ready_drafts\",\n    \"provider_prompt_syntax_count\": 0,\n    \"validator_failure_count\": 0,\n    \"production_lane_accepted_count\": 0\n  },\n  \"resource_summary\": {\n    \"seedance_resource\": {\n      \"zip_path\": \"docs/resource/【对客】Doubao-Seedance-2.0：企业级业务场景指南.zip\",\n      \"zip_sha256\": \"675c2969fb1f253084d5242de3948284b854b3bdc3d04765079bd5c9981bd026\",\n      \"md_member\": \"【对客】Doubao-Seedance-2.0：企业级业务场景指南.md\",\n      \"md_sha256\": \"58c1b2c67db6cbe460ab81a4f321906eb0cb006b40d5a2a78ba09a3756731daf\",\n      \"observed_industry_constraints\": [\n        \"Prompt formula is subject + action + camera language + reference assets + style/aesthetics + audio/SFX + restrictions.\",\n        \"Reference assets are core: images, videos, and audio should be explicitly addressed and reused across generation, editing, extension, and replacement tasks.\",\n        \"Provider-facing prompts increasingly support complex multi-shot camera choreography, transitions, physical realism, micro-expressions, sound synchronization, and story continuity.\",\n        \"Story generation value is not only a single shot; forward/backward extension and stitching between shots are first-class tasks.\",\n        \"Negative constraints and safety/legal boundaries around human likeness/reference assets must be carried as explicit control data.\"\n      ],\n      \"prompt_guide_excerpt_refs\": [\n        \"lines 137-157: prompt formula and component definitions\",\n        \"lines 161-174: camera scale/angle/motion/style vocabulary\",\n        \"lines 185-214: film examples for seamless cuts, one-take choreography, continuous action, micro-expression\",\n        \"lines 231-245: story continuity and video extension/stitching examples\"\n      ]\n    },\n    \"microglow_resource\": {\n      \"zip_path\": \"docs/resource/火山剧创优秀短剧作品赏析--动画短片 PV 剧本：《微光巢穴》.zip\",\n      \"zip_sha256\": \"dd834bff65d1b0ad587aca100cfba6b40c9bf78209515493aae85aa8906b7614\",\n      \"md_member\": \"火山剧创优秀短剧作品赏析--动画短片 PV 剧本：《微光巢穴》.md\",\n      \"md_sha256\": \"a5dde1154ee085e2b60acb7173fd8bd4815b0df2e01d272978e3b50d137434b8\",\n      \"observed_industry_constraints\": [\n        \"Top sample combines three layers: high-level style/tempo premise, scene/dialogue writing, and time-coded shot-by-shot PV execution plan.\",\n        \"The time-coded layer uses duration windows, picture/action, dialogue, subtitles, sound effects, music, action rhythm, and title cards.\",\n        \"Character, prop, monster, and scene settings include asset images; the production document treats visual references as part of the script package, not afterthoughts.\",\n        \"Combat/action shots emphasize physical tactics, weak-point logic, speed changes, smoke, lighting, impact SFX, and final hook.\",\n        \"The deliverable is not just prose; it is a reusable asset-linked production bible for video generation and iteration.\"\n      ],\n      \"structure_refs\": [\n        \"lines 3-14: duration/type/scene/role/tempo/color overview\",\n        \"lines 30-218: scene-level dramatic writing and dialogue\",\n        \"lines 230-394: time-coded PV shot plan\",\n        \"lines 405-420: character/prop/scene asset tables\"\n      ]\n    }\n  },\n  \"planning_questions\": [\n    \"How should the project define candidate Now-33/34/35/36 mainlines after Now-32 without activating them yet?\",\n    \"How should the pipeline exploit KG/entity links, asset/reference bindings, and traceability so it is not merely competing with direct DeepSeek creative writing?\",\n    \"How should DeepSeek v4 pro thinking/max be used as a standing consultant and score/review mechanism without becoming fake acceptance or evidence?\",\n    \"How should industry benchmark constraints from docs/resource be translated into future schema/gate/QA requirements?\",\n    \"What should remain explicitly blocked until real provider/manual evidence exists?\"\n  ],\n  \"required_response_schema\": {\n    \"overall_verdict\": \"string\",\n    \"strategic_diagnosis\": [\n      \"string\"\n    ],\n    \"candidate_mainlines\": [\n      {\n        \"id\": \"Now-32|Now-33|Now-34|Now-35|Now-36\",\n        \"name\": \"string\",\n        \"objective\": \"string\",\n        \"pipeline_advantage_to_unlock\": [\n          \"string\"\n        ],\n        \"allowed_work\": [\n          \"string\"\n        ],\n        \"blocked_work\": [\n          \"string\"\n        ],\n        \"required_artifacts\": [\n          \"string\"\n        ],\n        \"gate_criteria\": [\n          \"string\"\n        ],\n        \"deepseek_consultant_role\": [\n          \"string\"\n        ]\n      }\n    ],\n    \"pipeline_advantage_model\": [\n      \"string\"\n    ],\n    \"industry_benchmark_requirements\": [\n      \"string\"\n    ],\n    \"deepseek_operating_model\": [\n      \"string\"\n    ],\n    \"recommended_control_plane_updates\": [\n      \"string\"\n    ],\n    \"risks_and_countermeasures\": [\n      \"string\"\n    ],\n    \"do_not_do\": [\n      \"string\"\n    ],\n    \"confidence\": \"high|medium|low\"\n  }\n}"
  }
]