正文
获取原文件assets/motion-spec.jsonl.md
文件第一行是 sources 声明：每个上游快照在这里写一次 owner、artifact 和已接受的
artifact，后面的记录用它的 key 引用。key 用产物文件名派生的短小写名字，在本文件内稳定且唯一。
本阶段插入的片段（performance.fragment.json、
coverage-scope.fragment.json）落进同一个文件，它们用到的 key
也在这里声明。
{"record_type":"sources","schema_version":"1.0.0","sources":{"shots":{"owner":"short-drama-storyboard","artifact":"剧集/<EP>/storyboard/shots.jsonl"},"keyframes":{"owner":"short-drama-storyboard","artifact":"剧集/<EP>/storyboard/keyframes.jsonl"},"screenplay-index":{"owner":"short-drama-write","artifact":"剧集/<EP>/screenplay-index.jsonl"},"characters":{"owner":"short-drama-assets","artifact":"设定集/characters.jsonl"},"short-drama":{"owner":"creator","artifact":"short-drama.json"}}}
其后每行一个候选运动规格对象。引用写 src 加它指向的记录 record_id 或字段 field；
boundary_refs 只读，候选状态属于本运动规格，不向已经接受的上游引用传播。不要加入
duration_override、end_override 或 next_shot_write。以下字符串只说明怎样填写，
不是固定答案。
{
"motion_id": "MOTION-<stable-id>",
"status": "candidate",
"shot_ref": {
"src": "shots",
"record_id": "SHOT-<id>"
},
"keyframe_ref": {
"src": "keyframes",
"record_id": "KEY-<id>"
},
"production_profile_ref": {
"src": "short-drama",
"field": "/creator_authority/production_profile"
},
"boundary_refs": {
"duration": {
"src": "shots",
"record_id": "SHOT-<id>",
"field": "/duration_seconds",
"value_seconds": 0.0
},
"start": {
"src": "shots",
"record_id": "SHOT-<id>",
"field": "/start_boundary"
},
"primary_transition": {
"src": "shots",
"record_id": "SHOT-<id>",
"field": "/primary_transition"
},
"end": {
"src": "shots",
"record_id": "SHOT-<id>",
"field": "/end_boundary"
},
"next_start": {
"src": "shots",
"record_id": "SHOT-<next-id>",
"field": "/start_boundary",
"access": "comparison_only"
}
},
"reference_bindings": [
{
"slot_id": "REF-<stable-slot>",
"order": 1,
"artifact_ref": {
"src": "keyframes",
"record_id": "KEY-<id>"
},
"role": "start_frame | continuity_video | actual_tail_frame",
"may_control": [
"<本镜 accepted 起始构图与可见状态>"
],
"must_not_control": [
"<尚未发生的动作/终态/无权威文字>"
],
"admission_status": "unverified | creator_described | visually_inspected",
"reference_observation_ref": null,
"unresolved_risks": [
"<没有观察证据时保留的文字/水印/裁切风险>"
]
}
],
"start_anchor": {
"pose_balance": "<仅运动必需>",
"gaze": "<目标>",
"hands": {
"left": "<状态>",
"right": "<状态>"
},
"held_props": [
"<exact binding + hand>"
],
"spatial_relations": [
"<与行动对象的关系>"
]
},
"ordered_subject_motion": [
{
"order": 1,
"actor": "<asset binding>",
"trigger": "<accepted cue>",
"action": "<可见动作>",
"direction_or_path": "<方向/路径>",
"object_or_contact": "<对象/接触>",
"result": "<阶段结果>",
"timing": {
"mode": "relative | explicit",
"value": "<顺序词或秒区间>"
}
}
],
"camera": {
"behavior": "locked | move | transition",
"motivation": "reveal | pressure | alignment | relationship | transition | deliberate_stillness",
"intervals": [
{
"range": "<相对阶段或秒区间>",
"mode": "<lock/pan/tilt/dolly/handheld/follow>",
"path_tempo": "<方向/节奏>",
"endpoint": "<在 accepted framing/boundary 内>"
}
]
},
"environment_motion": [
{
"element": "<已有环境元素>",
"motion": "<有剧情意义的变化>",
"cause": "<连续性/主体动作>"
}
],
"audio": [
{
"source_ref": {
"src": "screenplay-index",
"record_id": "BLK-<EP>-<SC>-D<nn>"
},
"speaker_ref": {
"src": "characters",
"record_id": "CHAR-<id>"
},
"voice_direction_ref": {
"src": "characters",
"record_id": "CHAR-<id>",
"field": "/voice_direction"
},
"kind": "dialogue | VO | OS | SFX | ambience | music",
"exact_text": "<仅 source 有文本时逐字引用>",
"delivery_or_spatial_intent": "<不改文本的表演/声源/层级>",
"timing": "<相对阶段或秒区间>"
}
],
"timing_plan": {
"mode": "relative | explicit",
"phases": [
"<阶段、overlap 与 landing 空间>"
],
"declares_overlap": false,
"declared_total_or_endpoint_seconds": 0.0
},
"end_report": {
"projection": {
"pose": "<reported>",
"position": "<reported>",
"gaze": "<reported>",
"hands": "<reported>",
"held_props": "<reported>",
"visible_state": "<reported>"
},
"comparison": "match | mismatch | unrealized",
"differences": []
},
"reference_frame_economy": {
"frame_carries": [
"appearance",
"composition",
"base lighting"
],
"repeated_for_motion_only": [
"<hand/prop/path 等必要局部>"
]
},
"creator_overrides": [
{
"rule_id": "<VID-*>",
"choice": "<覆盖>",
"rationale": "<理由>"
}
],
"generic_prompt": "<把本规格渲染成一段只含要拍出来的画面的交付文本：从已接受起点的姿态与持物说起，逐条写动作、接触、摄影机行为与终点状态；不写镜头/记录 ID、规则 ID、状态词、工艺备注与成段否定罗列；写满的样子见 references/production-prompt-grammar.md>",
"derivation": {
"recipe_version": "<version>"
},
"provenance": "creator_project"
}
运动规格不带指回交付容器的引用。依赖方向只有一条：容器 → 运动规格 → 镜头。两端在各自
sources 里互相声明对方会形成环——依赖方向必须单向，
永远得不到可发布的稳定快照。要找某个镜头属于哪个容器，从容器记录的 members[] 反查，不在本文件里
存副本。容器记录见 delivery-container.jsonl.md。
默认 master 不重复 purpose_ref 或 coverage_scope：镜头目的、场次计划与原文覆盖从准确
shot_ref 及其上游读取。只有 pickup/alternate 才按
motion-recipe.md 增加 coverage_scope，记录相对母版的补充、保留与去向。
boundary_refs 只保留结构校验需要的 duration/start/end/next-start 精确字段投影，不再重复镜头目的、
场次计划或原文职责；删除它前必须先让 timing 与 continuity 校验器能从 shot_ref 安全解析同一快照。
普通记录省略 performance_arcs[] 与 attention_handoffs[]；存在可见表演变化或注意交接时，按
performance.fragment.json 插入完整字段。空镜、道具细节、纯空间
转场和只有物理动作的镜头不编造 arc。多参考的 slot_id 稳定且 order 唯一。
reported_end 只作比较；末镜没有真实下一镜时改用
next_start_locator。附加参考为空时使用空数组；对白说话者与声音方向只有在已接受引用存在时才填写。
母版、补拍和替代关系保留在同一规格文件内，替代决定由审查结论拥有。具体取舍按
references/motion-recipe.md 与 references/review-and-fixtures.md 判断。