Back to Details

event-production

Compare original and translation side by side

🇺🇸

Original

English
🇨🇳

Translation

Chinese

Event Production

活动制作

You own the technical execution layer: the signal path from a speaker's mouth and laptop to a room's projector, a recording file, and - sometimes - a live stream. Everything upstream arrives decided:
  • samber/dev-event-organizer-skills@event-format-selection
    decides whether the event is hybrid or streamed at all.
  • samber/dev-event-organizer-skills@event-schedule-design
    fixes which room at which hour.
  • samber/dev-event-organizer-skills@event-speaker-experience
    collects what each talk needs.
You turn those into equipment, crew, redundancy and a publication pipeline.
Five boundaries, four of them stated by the sibling from its own side.
  • Venue. 
    samber/dev-event-organizer-skills@event-venue-sourcing
    names hybrid and multi-venue technical specification as a gap it will not fill: "Do not invent an equipment list or a bandwidth figure. Put it to the venue as a question it must answer in writing" - and routes the specification here and to 
    samber/dev-event-organizer-skills@event-format-selection
    . You author that specification - guaranteed upstream bandwidth, rigging points, camera positions, uplink redundancy, control-room space - and hand it back as written questions for the venue to answer during negotiation. You never negotiate the contract, and you take the room's seated capacity and walking distances as given.
  • The day. 
    samber/dev-event-organizer-skills@event-run-of-show
    states the split: "
    samber/dev-event-organizer-skills@event-production
    engineers technical redundancy - the backup mic, the second recording path. You own what a room lead does in the ninety seconds after the redundancy fails." Build the backup path before the day starts. Never re-describe live incident escalation here.
  • Speakers. 
    samber/dev-event-organizer-skills@event-speaker-experience
    collects A/V requirements - "what the talk does, not just what the laptop is: a live demo, an audio clip, a hardware prop and an internet dependency" - and owns how that answer is asked for and acted on. You consume what it collects and you consume its consent record. You never publish a recording the consent record does not cover.
  • Captioning. 
    samber/dev-event-organizer-skills@event-accessibility-inclusion
    decides whether captioning and hearing loops are needed and to what standard. You deliver them technically, and they land in the same budget line as recording.
  • The supplier. 
    samber/dev-event-organizer-skills@event-vendor-sourcing
    carries AV and production suppliers in its own scope and states the split from its side: it "selects and contracts that supplier. Never let one skill do both jobs: a supplier picked before the spec exists writes the spec." You author the technical specification the supplier is quoted against and held to; you never run the quote comparison or sign the contract. Hand the brief over before selection starts, not after a quote arrives.
Every ranking below is a default, not a law: it shifts with context and with who executes it. After the interview, re-rank all three menus against what you know:
  • A venue with a house technician and installed cameras.
  • A community that already runs a video crew across editions.
  • An event that has never published a video.
  • An owned equipment stock.
Each of these overturns a default rung.
你负责技术执行层：从演讲者嘴边的麦克风和笔记本电脑到房间投影仪、录制文件，有时再到直播流的信号路径。上游的一切都已经决定好了：
  • samber/dev-event-organizer-skills@event-format-selection
    决定活动是否为混合形式，或是否要直播。
  • samber/dev-event-organizer-skills@event-schedule-design
    确定哪个房间在哪一个小时使用。
  • samber/dev-event-organizer-skills@event-speaker-experience
    收集每场演讲需要什么。
你把这些转化为设备、工作人员、冗余设计和发布流程。
五个边界，其中四个由同级技能从自身角度声明。
  • 场地。 
    samber/dev-event-organizer-skills@event-venue-sourcing
    将混合/多场地技术规格列为它不会填补的空白：“不要凭空编造设备清单或带宽数字。把它作为一个必须以书面形式回答的问题交给场地”——并将规格转到这里和 
    samber/dev-event-organizer-skills@event-format-selection
    。你编写该规格——保证的上行带宽、吊挂点、摄像机位置、上行链路冗余、控制室空间——并把它作为书面问题交回，供场地方在谈判期间回答。你从不谈判合同，并将房间的座位容量和行进距离视为已定条件。
  • 活动当天。 
    samber/dev-event-organizer-skills@event-run-of-show
    声明了分工：“
    samber/dev-event-organizer-skills@event-production
    设计技术冗余——备用麦克风、第二条录制路径。你负责冗余失效后的九十秒内房间负责人该做什么。”在活动开始前建立好备份路径。不要在这里重新描述现场事故升级流程。
  • 演讲者。 
    samber/dev-event-organizer-skills@event-speaker-experience
    收集音视频需求——“演讲要做什么，而不只是笔记本电脑是什么：现场演示、音频片段、硬件道具和网络依赖”——并负责如何询问和落实这一答案。你使用它收集到的内容及其同意记录。永远不会发布同意记录未覆盖的录制内容。
  • 字幕。 
    samber/dev-event-organizer-skills@event-accessibility-inclusion
    决定是否需要字幕和感应线圈，以及达到什么标准。你从技术上交付它们，它们与录制落在同一预算行中。
  • 供应商。 
    samber/dev-event-organizer-skills@event-vendor-sourcing
    在自己的范围内负责音视频和生产供应商，并从自身角度说明分工：它“选择并签约该供应商。绝不要让一个 skill 同时做两件事：在规格存在之前选定的供应商会自己编写规格。”你编写供应商据以报价并被考核的技术规格；你从不做报价比较或签合同。在比选开始前移交简报，而不是在报价到达之后。
下面的每个排序都是默认值，而不是铁律：它会随上下文和执行者而变化。完成访谈后，根据你所了解的情况重新对三个菜单排序：
  • 有驻场技术员和已安装摄像机的场地。
  • 已经跨届运营视频团队的社区。
  • 从未发布过视频的活动。
  • 自有设备储备。
其中每一条都会推翻某个默认层级。

Interview

访谈

Ask one question at a time, multiple-choice where possible. Questions 7-9 exist because the menus diverge sharply on time-to-effect, durability and effort - those orderings cannot be picked for the user.
  1. Is anything captured at all, and in what form: nothing, recorded for later, or streamed live? This is the question that reorders every menu below. Ask it first and get a real answer rather than an aspiration.
  2. What did the format decision actually commit to, and to whom? A promised livestream is a commitment; a hoped-for one is not.
  3. What is the grid - how many rooms run at once, how long is the shortest changeover between sessions, and which sessions are plenary?
  4. What did speaker collection return: live demos, audio playback, hardware props, internet dependencies, non-standard connectors? Which of those has no answer yet?
  5. What is the recording and publication consent posture per session format, and where is that record held? You cannot publish past it.
  6. What does the venue actually give you - a house technician or not, installed projection and sound or not, rigging points, a control-room space, an internet connection with a guaranteed upstream figure or a shrug?
  7. What is the last date the venue or the supplier can still add something? That deadline, not the event date, is what these menus are ranked against.
  8. One-off edition or a recurring event? A recurring event promotes every rung whose payoff lands after this edition ends: a trained crew, a written brief, owned equipment, a repeatable pipeline.
  9. What is the effort ceiling - who runs this on the day, how many people can be doing only production rather than also doing something else, and does anything have to be reversible after the event?
  10. Has accessibility decided on captioning or hearing loops? That decision lands in your budget line and your signal path, and it is not yours to make.
  11. What already exists that should re-rank the menus: cameras the community owns, a crew that has done this four times, a venue that streams its own events, last edition's recordings and what went wrong with them?
  12. Is the audience genuinely distributed, such that reaching a remote half is the point of the event rather than a bonus?
一次只问一个问题，尽可能提供多项选择。问题 7-9 之所以存在，是因为这些菜单在见效时间、持久性和投入上分歧很大——这些排序不能替用户挑选。
  1. 是否要采集任何内容，以什么形式：不采集、录制供以后观看，还是直播？ 这个问题会重新排列下面所有菜单。先问这个问题，并得到一个真实答案，而不是一个愿望。
  2. 形式决策实际承诺了什么，向谁承诺？承诺过的直播是义务；只是希望有直播则不是。
  3. 日程网格是什么——同时运行多少个房间，相邻场次之间最短转场时间是多少，哪些场次是全体会议（plenary）？
  4. 演讲者信息收集返回了什么：现场演示、音频播放、硬件道具、网络依赖、非标准连接器？其中哪些还没有答案？
  5. 每种场次形式的录制和发布同意立场是什么，记录存放在哪里？你不能越过它发布。
  6. 场地实际能提供什么——有没有驻场技术员，有没有已安装的投影和音响，有没有吊挂点、控制室空间，有没有带保证上行带宽数字的网络连接，还是只是耸耸肩？
  7. 场地或供应商还能追加东西的最后日期是哪天？这些菜单是相对于这个截止日期排序的，而不是活动日期。
  8. 单届活动还是周期性活动？周期性活动会提升每一级收益在本届结束之后才显现的层级：受过训练的团队、书面简报、自有设备、可重复的流程。
  9. 投入上限是什么——当天由谁运营，有多少人可以只做制作而不兼任其他工作，活动后是否需要可逆性？
  10. 无障碍团队是否已经决定要做字幕或感应线圈？这个决定会落在你的预算行和信号路径中，但不该由你来做。
  11. 已经存在哪些应该重新排序菜单的东西：社区自有的摄像机、已经做过四次的团队、为自己活动做直播的场地、上一届的录制成果以及它们出了什么问题？
  12. 观众是否真的分布在不同地点，以至于触达远程的一半是活动的意义所在，而不是附加好处？

What actually splits this work

真正切分这项工作的是什么

This collection usually replaces the B2B/B2C split with community-run versus vendor-run. Test it here rather than assuming it: it splits the crew and almost nothing else.
It holds for crew, which splits into two poles:
  • Community pole. Staffs production with volunteers. The sourced training ladder is built exactly for that pole: a full introduction deck "for events without 'experienced' A/V angels," a brief refresher plus hands-on training "for experienced angels" at a recurring event (C3VOC).
  • Vendor pole. Contracts a supplier. The source addresses it as an external party: "your audiovisual company should be able to provide a camera or cameras" (DevOpsDays).
Those are genuinely different jobs: one is recruiting and training people, the other is writing a brief and holding a supplier to it.
Capture posture is the axis that reorders everything else. Nothing captured, recorded for later, and streamed live are three different events technically, and the difference dwarfs who is paying:
  • Nothing captured. Needs a working room and no pipeline at all.
  • Recorded for later. Needs a clean signal, a reliable file and somewhere to put it. Forgives almost every failure that lasts under a minute.
  • Streamed live. Forgives nothing, needs rehearsal and a dedicated operator (DevOpsDays), carries a takedown risk on published platforms, and consumes an uplink nobody has measured.
State which of the three a recommendation assumes whenever they differ.
When asked to execute a hybrid or streamed format, report what the format costs technically before it is locked, then stop. The format decision belongs upstream; your job is to make sure it is taken with that number visible.
The clearest published finances available say hybrid "AV support" ran "more than double the cost of a physical-only conference," alongside a $484,886 loss in the year that conference ran fully hybrid (ACM SIGCHI, published Nov 2023). It is an academic conference, not a developer one: cite the ratio, never the figure, and never attribute it to a developer conference.
这套分工通常用“社区运营”对“供应商运营”来取代 B2B/B2C 的划分。在这里验证，而不是假设：它切分的是团队，几乎没别的。
这对团队成立，团队分为两极：
  • 社区极。 用志愿者为制作配备人员。引用的培训阶梯正是为这一极构建的：一套完整的入门讲解，“面向没有‘经验丰富的 A/V 天使（angels）’的活动”；在周期性活动（C3VOC）中，“为有经验的天使”提供简短复习加动手培训。
  • 供应商极。 签约供应商。来源把它当作外部方来写：“你的音视频公司应该能够提供一台或多台摄像机”（DevOpsDays）。
这确实是两种不同的工作：一种是招募和培训人员，另一种是编写简报并让供应商按简报执行。
采集姿态是重新排列一切的主轴。不采集、录制后观看和直播，在技术上是三种不同的活动，而且差异远超谁付钱的问题：
  • 不采集。 需要一个可用的房间，完全不需要流程管线。
  • 录制后观看。 需要干净的信号、可靠的文件和一个存放位置。几乎可以原谅所有持续不到一分钟的故障。
  • 直播。 不容忍任何故障，需要彩排和一名专职操作员（DevOpsDays），在已发布平台上存在下架风险，并消耗一条没人测量过的上行链路。
只要三者有差异，就要说明某条建议假设的是哪一种。
当被要求执行混合或直播形式时，在锁定之前报告该形式在技术上的成本，然后停止。形式决策属于上游；你的工作是确保在决策时能看到那个数字。
现有最清晰的公开财务数据表明，混合形式的“AV 支持”是“纯线下会议成本的两倍多”，而该会议完全采用混合形式的那一年还出现了 484,886 美元的亏损（ACM SIGCHI，2023 年 11 月发布）。这是一个学术会议，不是开发者会议：引用比例，不要引用具体数字，也不要把它归因于开发者会议。

Workflow

工作流程

  1. Establish capture posture. Nothing, recorded, or streamed. Write it down. Everything downstream is a consequence of this line.
  2. Read the inputs rather than re-collecting them: the grid, the per-speaker A/V needs, the consent record per format, the accessibility decisions, and the venue's own answers.
  3. Pick the coverage tier (see below) per room, not for the event as a whole. A plenary room and a third breakout room rarely deserve the same tier.
  4. Draw the signal path before ordering anything: where audio is captured, where it is mixed, where the room's display feed splits from the recorded feed, and where the crew watches what is actually going out. See references/signal-path-and-redundancy.md.
  5. Engineer redundancy against named failures, not against a general fear of failure (see below). The failure list is sourced; use it rather than improvising one.
  6. Assign crew roles and pick training depth (see below). Name people, not functions.
  7. Write the venue technical specification as questions and hand it back to venue sourcing to put in front of the venue in writing. Do not state a bandwidth figure you do not have - ask for a guaranteed upstream number and get it in the contract. Template in references/av-supplier-brief.md.
  8. Write the supplier brief and hand it to whoever contracts the supplier. See references/av-supplier-brief.md.
  9. Hand the Media cost line to 
    samber/dev-event-organizer-skills@event-budget
     with its tiering already split (see below). Supply the content; the budget decides the shape.
  10. Rehearse, if anything is live. Rehearsal is sourced as a requirement for streaming, not a nicety, and it is the step that gets cut first. It is also the only place a signal path gets tested end to end before an audience is in the room.
  11. Run the post-event pipeline: verify every file exists and is playable before the crew disperses, then edit, check against the consent record, publish, and send each speaker their own link.
Present the production plan section by section for validation, before anything is ordered:
  1. Capture posture and coverage tier.
  2. Signal path and redundancy.
  3. Crew and training.
  4. Venue spec and supplier brief.
  5. Budget and pipeline.
After the order goes in, changes cost money rather than a conversation.
If your harness has persistent memory, record per edition:
  • The venue's real answers on bandwidth, rigging and house equipment.
  • Which coverage tier each room actually got.
  • Every failure that happened and which redundancy caught it.
  • How long the pipeline took from event to published video.
  • What the supplier actually delivered against the brief.
A venue and a supplier used twice are worth not re-learning.
  1. 确定采集姿态。 不采集、录制或直播。把它写下来。下游的一切都是这一行的结果。
  2. 读取输入而不是重新收集： 日程网格、每位演讲者的音视频需求、每种形式的同意记录、无障碍决策，以及场地方自己的回答。
  3. 按房间选择采集覆盖层级（见下文），而不是按整个活动。一个全会厅和一个第三分会场很少应该用同一层级。
  4. 在订购任何东西之前画出信号路径： 音频在哪里采集，在哪里混音，房间显示信号与录制信号在哪里分开，工作人员在哪里观看实际输出的内容。参见 references/signal-path-and-redundancy.md。
  5. 针对已指明的故障设计冗余，而不是针对笼统的故障恐惧（见下文）。故障清单是有来源的；使用它，而不是临时编一个。
  6. 分配工作人员角色并选择培训深度（见下文）。任命具体的人，而不是职能。
  7. 把场地技术规格写成问题，交回给场地采购，以便以书面形式提交给场地方。不要给出你没有的带宽数字——要求一个保证的上行数字，并写进合同。模板见 references/av-supplier-brief.md。
  8. 编写供应商简报，交给负责签约供应商的人。参见 references/av-supplier-brief.md。
  9. 把 Media 成本行交给 
    samber/dev-event-organizer-skills@event-budget
    ，并已按层级拆分好（见下文）。提供内容；预算决定形态。
  10. 如果有任何直播内容，进行彩排。 彩排在引用来源中是直播的要求，而不是可有可无的加分项，但它也是第一个被砍掉的步骤。它也是唯一能在观众到场之前对信号路径进行端到端测试的地方。
  11. 运行活动后流程： 在团队解散前验证每个文件存在且可播放，然后剪辑、对照同意记录检查、发布，并把每个演讲者自己的链接发给他们。
在订购任何东西之前，逐节呈现制作计划以供确认：
  1. 采集姿态和覆盖层级。
  2. 信号路径和冗余。
  3. 工作人员和培训。
  4. 场地规格和供应商简报。
  5. 预算和流程。
订单下达后，变更花的是钱，而不是一次对话。
如果你的运行环境有持久记忆，请记录每届活动的：
  • 场地关于带宽、吊挂和场内设备的真实回答。
  • 每个房间实际获得的覆盖层级。
  • 发生过的每个故障，以及哪条冗余抓住了它。
  • 流程从活动到发布视频花了多长时间。
  • 供应商实际对照简报交付了什么。
使用过两次的场地和供应商，没必要重新学习一遍。

Coverage tier

覆盖层级

Recording is the floor and streaming is the layer above it: that ordering is sourced, not a preference. Published guidance states it plainly - "whether or not you're able to livestream, it's important to record all the talks" (DevOpsDays) - and the same guide's own tiered-budget mechanic makes streaming the first thing a lean tier drops while recording survives.
Ranking (default, not a law - Q1, Q7, Q8 and Q11 re-rank it):
  • effort (equipment, people on the day, rehearsal, coordination with the venue's network): 
    live stream on top of two cameras > two cameras with vision mixing > single operated camera > fixed camera on the board feed > smartphone floor > no capture
  • value (a talk that outlives the day, and a speaker with something to show for it): 
    live stream on top of two cameras > two cameras with vision mixing == single operated camera > fixed camera on the board feed > smartphone floor > no capture
    - argued tie: both put a watchable, audible talk in the archive, which is the whole of the value for an audience watching later. The second camera buys slide detail and reaction shots, and the operator buys framing; neither buys a talk that did not otherwise exist.
  • compliance cost (the review it triggers, and what stops being reversible): 
    live stream on top of two cameras > two cameras with vision mixing == single operated camera == fixed camera on the board feed == smartphone floor > no capture
    - argued tie: every recorded rung needs the same thing, a consent record covering that session's format before anything is published, and none of them is harder to obtain than the others. Streaming is the rung that changes the review, because a live feed is published before anyone can check it. A speaker's opt-out has to be honoured in the room rather than in an edit, and background music in a demo can take the stream down on the platform's own initiative (DevOpsDays). A published stream cannot be recalled from the people already watching it.
  • efficiency: 
    fixed camera on the board feed > smartphone floor > single operated camera > two cameras with vision mixing > live stream on top of two cameras > no capture
  • Fixed camera on the board feed - the default and the sourced recommendation: a camera and a recording computer taking audio off the mixing board rather than the camera's own microphone. Near-zero on the day once set, and the board feed is what makes the result listenable. Move up one rung when slides carry the talk and a wide shot cannot read them, or when a plenary is the recording anyone will actually watch.
  • Smartphone floor - the sourced absolute minimum: "if you have no budget for recording talks, a smartphone is better than nothing." It sits second on efficiency rather than first because a phone on a tripod at the back of a room captures room audio, and a talk you cannot hear is not an archive. Use it for a room you were going to leave uncovered, never as the plan for the room that matters.
  • Single operated camera - a person framing the shot, following the speaker, and holding on the slides when the slides are the point. Buys watchability; costs one trained person per room per session.
  • Two cameras with vision mixing - a second angle plus somebody switching between feeds. The camera operator and the vision mixer are two separate named roles in sourced practice, not one person with more work.
  • Live stream on top of two cameras - the starved option: tops value and tops effort, so efficiency never picks it. It also needs the two things the source names explicitly and every organizer under-provisions: rehearsal ahead of time, and at least one person whose only job during the event is running it. Promotion conditions, keyed to Q2, Q2, Q12 and Q6/Q11: an audience that genuinely cannot travel and was promised access, a sponsor commitment that names streaming, a distributed community whose remote half is the point of the event, or a venue that already streams its own events and hands you the uplink.
  • Delete, do not demote: streaming with no rehearsal and no dedicated operator. It reads as the cheap version of streaming and it is not a rung on this ladder - it is the same cost with the failure built in, discovered live, in front of the exact audience the stream existed to serve. Parked at the bottom of a menu it reappears as "we'll just point the laptop at it."
录制是底线，直播是它上面的一层：这个排序有来源，而不是偏好。公开的指南说得很清楚——“无论你是否能够直播，录制所有演讲都很重要”（DevOpsDays）——同一份指南自己的分层预算机制也让直播成为精简层级第一个砍掉的东西，而录制保留。
排序（默认值，不是铁律——Q1、Q7、Q8 和 Q11 会重新排序）：
  • 投入（设备、当天人员、彩排、与场地网络的协调）：
    双机位加直播 > 双机位加视频切换 > 单机位人工操作 > 固定机位拍摄调音台信号 > 手机地面机位 > 不采集
  • 价值（一段比当天活得更久的演讲，以及演讲者有拿得出手的东西）：
    双机位加直播 > 双机位加视频切换 == 单机位人工操作 > 固定机位拍摄调音台信号 > 手机地面机位 > 不采集
    - 有论证的并列：两者都在档案中留下一个可观看、可收听的演讲，这对以后观看的观众来说就是全部价值。第二台摄像机带来幻灯片细节和反应镜头，操作员带来构图；两者都不会买来一个本不存在的演讲。
  • 合规成本（它触发的审查，以及什么不再可逆）：
    双机位加直播 > 双机位加视频切换 == 单机位人工操作 == 固定机位拍摄调音台信号 == 手机地面机位 > 不采集
    - 有论证的并列：每一级录制层级都需要同一样东西，即在发布任何内容之前有一份覆盖该场次形式的同意记录，而且没有哪一级比其它更难获得。直播才是改变审查流程的那一级，因为直播信号在任何人检查之前就已经发布了。演讲者的退出选择必须在现场得到尊重，而不是在剪辑中；演示中的背景音乐可能让平台自行下线直播（DevOpsDays）。已经发布出去的直播无法从正在观看的人那里召回。
  • 效率：
    固定机位拍摄调音台信号 > 手机地面机位 > 单机位人工操作 > 双机位加视频切换 > 双机位加直播 > 不采集
  • 固定机位拍摄调音台信号——默认且来自引用来源的推荐：一台摄像机和一台录制电脑，音频取自调音台而不是摄像机自带麦克风。设置好后当天几乎零成本，调音台信号是让结果听得清的关键。当演讲由幻灯片承载而全景画面读不清幻灯片时，或者当全会议程才是大家真正会看的录制内容时，向上移一级。
  • 手机地面机位——引用来源中的绝对最低标准：“如果没有预算录制演讲，智能手机总比没有好。”它在效率上排第二而不是第一，因为放在房间后面的三脚架上的手机会采集房间环境音，而听不见的演讲不构成归档。用于你原本打算不覆盖的房间，绝不要把它作为重要房间的方案。
  • 单机位人工操作——一个人负责取景构图、跟随演讲者，并在幻灯片是重点时稳住画面。买来的是可看性；成本是每个房间每场次一名受训人员。
  • 双机位加视频切换——第二个角度，加一个在信号之间切换的人。在引用来源的实践中，摄像师和视频切换员是两个独立的具名角色，而不是一个人承担更多工作。
  • 双机位加直播——预算紧张时最先被牺牲的选项：价值和投入都最高，所以效率永远不会选它。它还需要引用来源明确列出、而每个组织者都配置不足的两样东西：提前彩排，以及至少一名活动期间唯一工作就是运行它的人。晋升条件，对应 Q2、Q2、Q12 和 Q6/Q11：一群确实无法到场且被承诺可以观看的观众、一项明确提到直播的赞助承诺、一个远程一半是活动意义的分布式社区，或者一个已经为自己活动做直播并把上行链路交给你的场地。
  • 删除，不要降级：没有彩排、没有专职操作员的直播。 它看起来像直播的廉价版本，但它不是这个阶梯上的一级——它是同样的成本，只是内置了故障，并在直播原本要服务的观众面前现场暴露。把它放在菜单底部，它会以“我们就把笔记本电脑对准它吧”的形式重新出现。

Redundancy depth

冗余深度

Redundancy here means a backup path engineered before the day, not a person improvising during it. The sourced failure list is what it protects against, and the load-bearing lesson is that a backup system has its own failure modes.
At one large conference (C3VOC, 34C3), the backup existed and was still unusable:
  • 128 recorded audio tracks with no labels.
  • A recorder with no clock sync.
  • Drives a person had to physically carry to an unloading station and back.
  • A bulk copy that spiked the network on a predictable cadence.
  • Multi-gigabyte files nobody could seek through afterwards.
The redesign written to replace it writes chunked, named files synced continuously to a storage server. Full detail in references/signal-path-and-redundancy.md.
Ranking (default, not a law - Q1, Q6, Q8 and Q9 re-rank it):
  • effort (hardware, wiring time, setup per room, someone who understands it): 
    duplicate encode path > continuous chunked sync to storage > three-destination signal split > independent second audio recording > spare mic and spare cable
  • value (a usable file exists after a failure nobody noticed at the time): 
    continuous chunked sync to storage > independent second audio recording > duplicate encode path > three-destination signal split > spare mic and spare cable
  • value (the session in the room keeps running): 
    spare mic and spare cable > three-destination signal split > duplicate encode path == independent second audio recording == continuous chunked sync to storage
    - argued tie: all three sit downstream of the split that feeds the room, so none of them is audible or visible to the audience. They protect the archive and the stream, and the room never knows either way.
  • efficiency: 
    spare mic and spare cable > independent second audio recording > three-destination signal split > continuous chunked sync to storage > duplicate encode path
  • Spare mic and spare cable - the default: a second microphone within reach, spare batteries, a spare cable for every run, and a spare adapter for every connector the speaker collection surfaced. Near-zero, and it is the only rung that saves the session happening right now. Move up one rung the moment any session is not repeatable - a keynote, a launch announcement, a speaker who flew in for one talk.
  • Independent second audio recording - a recorder that is not in the live mix, capturing the same source. The point is independence: a failure in the mixing or encoding chain does not reach it. Audio-only, because audio is what makes a recording usable and video without it is not.
  • Three-destination signal split - the sourced architecture pattern: one input, split to the room's projector with minimal delay, to the encoder that produces the recorded or streamed output, and to a monitor the crew watches. Those are three destinations, not three copies of one pipeline, and a failure in the encode path does not have to take down what the room sees or what the crew is watching. Its second payoff is diagnostic: without the crew's own monitor, nobody finds out the recording has been black for forty minutes.
  • Continuous chunked sync to storage - the starved option: recordings written as short, named, clock-synced chunks and synced continuously to a server rather than collected in bulk afterwards. It tops the archive-value axis, because every failure mode it fixes is one where the recording nominally succeeded and was useless anyway, and it tops effort, needing infrastructure and someone who maintains it. Promotion conditions: a multi-room event where files are collected from several rooms at once, a recurring event that will amortise the setup, an archive somebody has committed to publishing on a deadline, or a previous edition where a recording was lost or unusable.
  • Duplicate encode path - a full second encoding chain in parallel. Real redundancy, real cost, and it protects the layer that is already the least likely to fail silently.
  • Delete, do not demote: an unlabelled backup nobody has tried to open. A drive full of unnamed tracks with no time sync is not a backup, and the moment it is needed is the wrong moment to discover that. Every backup rung above owes a labelling convention and one test playback before the event.
这里的冗余是指在活动前设计好的备份路径，而不是活动期间有人即兴发挥。它保护的是来自引用来源的故障清单，而关键教训是备份系统也有自己的故障模式。
在一个大型会议（C3VOC，34C3）上，备份存在但仍然不可用：
  • 128 条没有标签的录音音轨。
  • 一台没有时钟同步的录音机。
  • 需要有人把硬盘实体往返搬运到卸载站。
  • 一次以可预测节奏冲击网络的批量复制。
  • 之后没人能快速定位播放的多个 GB 级文件。
为替代它而编写的重新设计会写入分块、带名称的文件，并持续同步到存储服务器。完整细节见 references/signal-path-and-redundancy.md。
排序（默认值，不是铁律——Q1、Q6、Q8 和 Q9 会重新排序）：
  • 投入（硬件、布线时间、每个房间的设置、一个懂它的人）：
    重复编码路径 > 持续分块同步到存储 > 三路信号分发 > 独立第二路录音 > 备用麦克风和备用线缆
  • 价值（在没人当时注意到的故障之后仍然存在可用文件）：
    持续分块同步到存储 > 独立第二路录音 > 重复编码路径 > 三路信号分发 > 备用麦克风和备用线缆
  • 价值（房间里的这场议程继续运行）：
    备用麦克风和备用线缆 > 三路信号分发 > 重复编码路径 == 独立第二路录音 == 持续分块同步到存储
    - 有论证的并列：这三者都位于为房间供信号的分发点下游，所以观众既听不到也看不到它们。它们保护的是归档和直播，房间里的观众永远不知道有这回事。
  • 效率：
    备用麦克风和备用线缆 > 独立第二路录音 > 三路信号分发 > 持续分块同步到存储 > 重复编码路径
  • 备用麦克风和备用线缆——默认：一个触手可及的备用麦克风、备用电池、每段走线的备用线缆，以及演讲者信息收集暴露出的每种连接器的一个备用转接头。几乎零成本，也是唯一能救下正在进行的这场议程的层级。只要任何一场议程不可重来——主旨演讲、发布公告、飞过来只为讲一场的演讲者——就向上移一级。
  • 独立第二路录音——一台不在直播混音链路中的录音机，录制同一音源。重点在于独立：混音或编码链路中的故障不会波及它。只录音频，因为音频是一段录制可用性的关键，没有音频的视频不算可用。
  • 三路信号分发——来自引用来源的架构模式：一路输入，分发到房间投影仪（最小延迟）、到产生录制或直播输出的编码器、以及到工作人员观看的监视器。这是三个目的地，而不是同一条管线的三份拷贝；编码路径中的故障不必连带摧毁房间看到的内容或工作人员正在监视的内容。它的第二个收益是诊断性的：没有工作人员自己的监视器，没人会发现录制已经黑屏了四十分钟。
  • 持续分块同步到存储——预算紧张时最先被牺牲的选项：录音被写成短的、带名称的、时钟同步的分块，并持续同步到服务器，而不是事后批量收集。它在归档价值轴上最高，因为它修复的每一种故障模式都是“录制名义上成功但反正没用”的情况；它在投入上也最高，需要基础设施和一个维护它的人。晋升条件：需要同时从多个房间收集文件的多房间活动、可以摊平这套设置的周期性活动、有人承诺按期发布的归档，或者上一届出现过录制丢失或不可用的活动。
  • 重复编码路径——一条并行的完整第二编码链路。真正的冗余、真正的成本，它保护的正是已经最不可能静默出问题的那一层。
  • 删除，不要降级：没人尝试打开过的无标签备份。 一个装满未命名音轨、没有时间同步的硬盘不是备份；等到需要它的那一刻才去发现这一点是最糟糕的时机。上面每一级备份层级都需要一套命名规范，并在活动前做一次试播放。

Crew and training

工作人员和培训

Sourced practice names six distinct crew roles, not one "AV person" (C3VOC):
  • Stage manager.
  • Audio operator.
  • Camera operator.
  • Video/vision-mixing operator - a different person from the camera operator.
  • AV-technical lead, above the operator roles.
  • General crew.
Which of them exist on your event is the staffing decision below. What each one does minute to minute is yours to define; see references/crew-roles-and-training.md for the role split and the training ladder.
Ranking (default, not a law - Q1, Q8, Q9 and Q11 re-rank it):
  • effort (people recruited, hours trained, coordination on the day, money): 
    full role split per room > contracted supplier plus a named counterpart > audio and video split > one named production owner
  • value (the technical failure that does happen gets caught by somebody whose job it was): 
    full role split per room > contracted supplier plus a named counterpart > audio and video split > one named production owner
  • efficiency: 
    one named production owner > audio and video split > contracted supplier plus a named counterpart > full role split per room
Dominance check: 4 rungs, 6 pairs, zero strict-dominance relations. Clean only by construction, and by-construction is never a pass. Value and effort are the same list, so one mechanism blocks all six pairs: the staffing that catches more failures costs strictly more people, hours, and money.
No third axis is printed to block or rescue a pair. The check catches no misordering; the efficiency line rests on the arguments below.
  • One named production owner - the default: a single person who owns the signal path, has walked it, and is not also MCing, speaking or running registration. Near-zero cost, and it converts "the AV" from a hope into somebody's job. Move up one rung as soon as more than one room runs at once, or as soon as anything is live.
  • Audio and video split - two people, because the failures are unrelated and they happen simultaneously. Audio is where the archive is won or lost; video is where the room's attention is.
  • Contracted supplier plus a named counterpart - the vendor pole. The supplier brings the crew; you still name one organizer who owns the brief, checks delivery against it, and is reachable on the day. A supplier with no counterpart delivers what it usually delivers, not what you asked for.
  • Full role split per room - the starved option: stage manager, audio, camera, vision mixing and a technical lead, per room. Tops value and tops effort together, so efficiency never picks it, and at community scale the headcount it needs is the whole organizing team. Promotion conditions: multi-camera coverage in any room, a live stream, a recurring event with a returning crew that already knows the roles, or a room where a failure is unrecoverable.
  • Delete, do not demote: the venue's house technician as the whole plan. It is not a cheap rung, it is an unowned one - a house technician runs the house system, and nobody in that arrangement owns your recording, your consent record or your files. Keep the technician; add an owner.
Training depth follows a sourced two-tier ladder rather than a single onboarding.
  • Run the full introduction when the host community has no experienced crew: a first edition, a new team, a venue nobody has worked.
  • Run the brief refresher plus hands-on for newcomers when a returning crew already knows the drill. This is the recurring-event case, and the reason a trained crew is one of the highest-return investments a repeating event makes.
Both are named practice. The hands-on tier splits into per-topic self-training decks - basics, camera, video mixer, talk timeline - and the operating detail behind each role (camera framing rules, mixer console moves, the phase-by-phase talk timeline) is in references/crew-roles-and-training.md.
来自引用来源的实践列出了六个不同的工作人员角色，而不是一个“音视频人员”（C3VOC）：
  • 舞台经理。
  • 音频操作员。
  • 摄像师。
  • 视频/视频切换操作员——与摄像师不是同一个人。
  • 音视频技术负责人，位于操作员角色之上。
  • 普通工作人员。
你的活动需要其中哪些角色，就是下面的人员配置决策。每个人每分钟做什么由你来定义；角色分工和培训阶梯见 references/crew-roles-and-training.md。
排序（默认值，不是铁律——Q1、Q8、Q9 和 Q11 会重新排序）：
  • 投入（招募的人、培训的工时、当天的协调、金钱）：
    每个房间完整角色分工 > 签约供应商加一名具名对接人 > 音频和视频分开 > 一名具名制作负责人
  • 价值（确实发生的技术故障被某个职责所在的人抓住）：
    每个房间完整角色分工 > 签约供应商加一名具名对接人 > 音频和视频分开 > 一名具名制作负责人
  • 效率：
    一名具名制作负责人 > 音频和视频分开 > 签约供应商加一名具名对接人 > 每个房间完整角色分工
支配关系检查：4 个层级，6 对，零个严格支配关系。 只有靠构造才干净，而“靠构造”从来不是通过的理由。价值和投入是同一个列表，所以同一个机制阻挡了全部六对：能抓住更多故障的人员配置，严格地需要更多人员、工时和金钱。
没有第三个轴被列出来阻挡或挽救某一对。检查没有发现排序错误；效率这条线依赖下面的论证。
  • 一名具名制作负责人——默认：一个人拥有信号路径、亲自走查过它，并且不是同时担任主持人、演讲者或签到工作。成本几乎为零，它把“音视频”从一个愿望变成某个人的工作。只要同时运行超过一个房间，或有任何直播内容，就向上移一级。
  • 音频和视频分开——两个人，因为这两类故障互不相关而且会同时发生。音频决定归档的成败；视频决定房间里观众的注意力。
  • 签约供应商加一名具名对接人——供应商这一极。供应商带来团队；你仍然要指定一名组织者负责简报、对照简报检查交付，并在活动当天可以联系到。没有对接人的供应商只会交付它通常交付的东西，而不是你要求的东西。
  • 每个房间完整角色分工——预算紧张时最先被牺牲的选项：每个房间都有舞台经理、音频、摄像、视频切换和技术负责人。价值和投入同时最高，所以效率永远不会选它；在社区规模下，它需要的人数_就是_整个组织团队。晋升条件：任意房间的多机位覆盖、直播、有已经熟悉这些角色的回归团队的周期性活动，或者某个房间的故障不可恢复。
  • 删除，不要降级：把场地驻场技术员当作全部方案。 它不是一个廉价层级，而是一个无人负责的层级——驻场技术员运行的是场地系统，而在这种安排中没有人对你的录制、你的同意记录或你的文件负责。保留技术员；增加一个负责人。
培训深度 遵循来自引用来源的两层阶梯，而不是一次性的入职培训。
  • 当主办社区没有经验丰富的团队时，运行完整入门培训：比如首届活动、新团队、没人合作过的场地。
  • 当回归团队已经熟悉流程时，运行简短复习加新手实操。这就是周期性活动的情况，也是受过培训的团队成为重复举办活动回报率最高投资之一的原因。
两者都是有名字的实践。实操层分为按主题划分的自培训材料——基础、摄像、视频切换台、演讲时间线——每个角色背后的操作细节（取景规则、切换台操作动作、分阶段的演讲时间线）见 references/crew-roles-and-training.md。

Media budget line

Media 预算行

Hand 
samber/dev-event-organizer-skills@event-budget
a Media line, not a lump sum, and hand it already split by what survives a lean tier. The sourced taxonomy bundles live streaming, recording and captioning into one Media category, with Signage and Internet as separate categories you neither own nor absorb - attendee internet provisioning is a different line from the production pipeline's own uplink.
Split the line three ways so the budget can cut it without asking you:
  • Recording - survives every tier down to the smartphone floor.
  • Captioning - its necessity is an accessibility decision you deliver rather than decide.
  • Live streaming - the first item a lean tier drops, and the one that takes rehearsal and a dedicated operator down with it.
Give each part equipment, crew and supplier quote separately. A single Media number gets cut whole.
交给 
samber/dev-event-organizer-skills@event-budget
的是一个 Media 预算行，而不是一笔总金额，并且在交接时已经按什么能撑过精简层级来拆分。来自引用来源的分类法把直播、录制和字幕归入一个 Media 类别，**Signage（标识）和Internet（网络）**是你既不拥有也不吸收的独立类别——参会者网络接入与制作管线自身的上行链路是不同的一行。
把这一行拆成三份，让预算方不需要问你就能削减：
  • 录制——从最高层到手机地面机位都能存活。
  • 字幕——是否必要由无障碍决策决定，你负责交付而不是决定。
  • 直播——精简层级第一个砍掉的项目，它还会把彩排和专职操作员一起带走。
分别给出每个部分的设备、人员和供应商报价。如果只给一个 Media 总额，它会被整个砍掉。

On-site podcast setups

现场播客设置

Treat an on-site podcast as talk recording at smaller scale, with three differences that actually matter:
  • It needs a genuinely quiet room rather than a quiet corner.
  • It needs one microphone per participant rather than one shared.
  • Its consent question differs from a talk's: a conversational recording captures people who never agreed to be on stage.
Get consent per participant, in writing, before the first session, not from the schedule. Everything else is the same signal path and the same redundancy ladder above.
把现场播客当作小规模的演讲录制，有三个真正重要的区别：
  • 它需要一个真正安静的房间，而不是一个安静的角落。
  • 它需要每位参与者一支麦克风，而不是共用一支。
  • 它的同意问题不同于演讲：一段对话式录制会录下从未同意上台的人。
在第一场录制前，获得每位参与者书面同意，而不是从日程表里推断。其它一切都与上面相同的信号路径和相同的冗余阶梯。

Failure modes

故障模式

  • Streaming with no rehearsal and no dedicated operator. Published streaming guidance names both as requirements. An unrehearsed stream run by whoever is free fails live, in front of the audience it existed for.
  • Recording off the camera's own microphone. The sourced chain takes audio off the mixing board. Room audio from a camera at the back produces a file nobody finishes watching, and it is discovered weeks later during editing.
  • A backup with no labels and no clock sync. Sourced, at scale: recordings that existed and could not be used. Name files, sync clocks, and open one before the event.
  • Bulk-copying files over the event network. A backup process that saturates the network on a predictable cadence has become an incident of its own. Sync continuously in chunks, or copy when nothing else needs the network.
  • Deciding the format instead of reporting its cost. Hybrid roughly doubles AV cost in the clearest published finances available. Report that before the decision, then execute whatever was decided.
  • Publishing past the consent record. A camera in the room is not agreement. The consent record names the session format and the opt-out answer; anything not covered stays unpublished, including a stream where the opt-out has to be honoured in the room rather than in an edit.
  • Background music in a recorded demo. A concrete, sourced platform risk: a stream carrying incidental music can be taken down by the platform. Warn speakers before the day rather than filing an appeal after it.
  • Specifying a bandwidth figure you invented. Ask the venue for a guaranteed upstream figure in writing and put it in the contract instead of naming a number yourself.
  • A pipeline with no owner after the event. Files are collected, the crew disperses, and the videos are published five months later or never. Name the owner and the date before the event, and verify every file plays before anyone leaves the building.
  • 没有彩排、没有专职操作员的直播。 已发布的直播指南把两者都列为要求。由随便哪个有空的人来运行的未经彩排的直播，会在它原本要服务的观众面前现场失败。
  • 用摄像机自带麦克风录制。 来自引用来源的信号链从调音台取音频。房间后方摄像机录到的环境声会产生一个没人会看完的文件，而且这个问题要到几周后剪辑时才会被发现。
  • 没有标签、没有时钟同步的备份。 来自引用来源的大规模教训：录音存在却无法使用。给文件命名、同步时钟，并在活动前打开其中一个试试。
  • 在活动网络上批量复制文件。 一个以可预测节奏饱和网络的备份流程本身已经成为一起事故。持续分块同步，或在没有其它东西需要网络时再复制。
  • 代替报告成本去决定形式。 在现有最清晰的公开财务数据中，混合形式大约让 AV 成本翻倍。在决策前报告这一点，然后执行任何已决定的方案。
  • 超出同意记录范围发布。 房间里有一台摄像机不代表同意。同意记录写明场次形式和退出选择；任何未覆盖的内容都保持不发布，包括必须在现场而不是在剪辑中尊重退出选择的直播。
  • 录制演示中的背景音乐。 一个具体的、有来源的平台风险：带有附带音乐的直播流可能被平台下架。在活动前提醒演讲者，而不是事后申诉。
  • 报出一个你编造的带宽数字。 要求场地以书面形式给出保证的上行数字并写进合同，而不是自己报一个数。
  • 活动后没有负责人的流程。 文件收集完，团队解散，视频五个月后才发布或永远不会发布。在活动前指定负责人和日期，并在任何人离开场地前验证每个文件都能播放。

Measurement

衡量指标

Three completeness gates, checkable before doors open and free of any threshold to argue about:
  • Every room's signal path has been walked end to end, with one test recording played back - not checked as a level meter, played back.
  • Every published-recording plan has a consent record covering that session's format, and every session outside it is marked unpublishable.
  • Every camera, recorder and encoder has a named owner for every block, and the livestream has a person whose only job it is.
Everything below is a metric you set rather than a benchmark to hit. Say so when you present it.
  • Capture completeness (self-set): sessions with a usable recording, over sessions scheduled. Count "recorded but unusable" as a failure, since that is the failure the sourced case study is about.
  • Failures caught by a redundancy versus failures that reached the audience (self-set). The ratio is what justifies the redundancy rung next edition.
  • Time from event end to published video (self-set), plus the point in the pipeline where it stalled. This is the number that decides whether a recurring event needs a pipeline or a volunteer.
  • Supplier delivery against the brief (self-set): line items delivered as specified, over line items in the brief. It only exists if the brief exists.
Pick two or three, write down the change each would trigger, and record them before the event rather than after.
三个完整性门禁，可以在开场前检查，而且没有任何需要争论的阈值：
  • 每个房间的信号路径都已端到端走查过，并播放过一次测试录音——不是当作电平表检查，而是真正播放出来。
  • 每个已发布录制计划都有覆盖该场次形式的同意记录，所有不在其中的场次都被标记为不可发布。
  • 每台摄像机、录音机和编码器在每个时段都有具名负责人，直播流也有一个唯一职责就是负责它的人。
下面这些都是由你设定的指标，而不是需要达成的基准。在呈现时说明这一点。
  • 采集完整性（自设）：有可用录制的场次 ÷ 已排定的场次。把“已录制但不可用”算作失败，因为那正是引用案例研究中的失败。
  • 被冗余抓住的故障 对比 到达观众的故障（自设）。这个比例是下一届证明冗余层级的依据。
  • 从活动结束到视频发布的时间（自设），以及流程中卡住的位置。这个数字决定周期性活动需要的是流程管线还是一名志愿者。
  • 供应商对照简报的交付情况（自设）：按规格交付的条目 ÷ 简报中的条目。只有简报存在，这个指标才存在。
选择两三个，写下每个指标会触发的变更，并在活动前而不是活动后记录它们。

Invocation examples

调用示例

  • "We're recording all the talks at a two-room conference. What do we actually need?"
  • "A sponsor wants us to livestream the keynote. What does that add?"
  • "Write the AV brief we send to suppliers for quotes."
  • "The venue asked what our technical requirements are. What do we send them?"
  • "What's our backup plan if the mic dies during the keynote?"
  • "How many people do we need on the video crew, and what does each of them do?"
Expected output: a production plan presented section by section for validation, before anything is ordered.
  1. The capture posture stated in one line.
  2. The coverage tier per room with its reasoning.
  3. The signal path drawn end to end.
  4. The redundancy rung chosen and the named failure each part protects against.
  5. Crew roles assigned by name plus the training depth.
  6. The venue technical specification written as questions for venue sourcing to put in writing.
  7. The supplier brief.
  8. The Media budget line split three ways.
  9. The rehearsal plan.
  10. The post-event pipeline with its owner and date.
  • “我们要在一个双房间会议录制所有演讲。我们到底需要什么？”
  • “赞助商想让我们直播主旨演讲。这会增加什么？”
  • “编写我们发给供应商询价的 AV 简报。”
  • “场地问我们的技术需求是什么。我们该发什么给他们？”
  • “如果主旨演讲期间麦克风坏了，我们的备用计划是什么？”
  • “视频团队需要多少人，每个人做什么？”
预期输出：在订购任何东西之前，逐节呈现制作计划以供确认。
  1. 用一句话说明采集姿态。
  2. 每个房间的覆盖层级及其理由。
  3. 端到端画出的信号路径。
  4. 选择的冗余层级，以及每个部分防御的具名故障。
  5. 按姓名分配的工作人员角色，加上培训深度。
  6. 写成问题的场地技术规格，供场地采购以书面形式提交。
  7. 供应商简报。
  8. 拆成三份的 Media 预算行。
  9. 彩排计划。
  10. 活动后流程，及其负责人和日期。

References

参考资料

  • references/signal-path-and-redundancy.md - the three-destination split in full, the documented failure-mode list and the redesign that answered it, what carries over to a smaller event and what does not, and a positive/negative pair of redundancy plans.
  • references/av-supplier-brief.md - the constructed supplier-brief template and the venue technical specification as written questions, with the bandwidth figure left for the venue to state in writing.
  • references/crew-roles-and-training.md - the six sourced role names and what each covers, the two-tier training ladder with the promotion condition, the one-camera and two-camera coverage difference, and the rehearsal walkthrough.
  • samber/dev-event-organizer-skills@event-format-selection
    - decides whether the event is hybrid or streamed; this skill reports what that costs technically and then executes it.
  • samber/dev-event-organizer-skills@event-speaker-experience
    - collects per-speaker A/V needs and the recording consent record, both of which are inputs here and neither of which is re-collected.
  • samber/dev-event-organizer-skills@event-venue-sourcing
    - gathers the venue-side capacity answers this skill builds against; it routes the technical specification here and this skill hands it back as written questions.
  • samber/dev-event-organizer-skills@event-budget
    - receives the Media cost line already split into recording, captioning and streaming, and decides how it fits the overall shape.
  • samber/dev-event-organizer-skills@event-vendor-sourcing
    - selects and contracts the AV supplier against the technical specification authored here; never let one skill do both jobs.
  • samber/dev-event-organizer-skills@event-schedule-design
    - fixes the grid the crew works inside; a room unfit for a format is reported upstream, never solved by dropping a session.
  • references/signal-path-and-redundancy.md - 三路分发完整说明、记录的故障模式清单和应对它的重新设计、哪些可以沿用到更小的活动、哪些不能，以及一组正/反冗余计划示例。
  • references/av-supplier-brief.md - 构建好的供应商简报模板，以及作为书面问题的场地技术规格，带宽数字留给场地方书面说明。
  • references/crew-roles-and-training.md - 来自引用来源的六个角色名称及各角色职责、带晋升条件的两层培训阶梯、单机位与双机位覆盖差异，以及彩排走查。
  • samber/dev-event-organizer-skills@event-format-selection
    - 决定活动是否混合或直播；本 skill 报告其技术成本，然后执行。
  • samber/dev-event-organizer-skills@event-speaker-experience
    - 收集每位演讲者的 A/V 需求和录制同意记录，两者都是这里的输入，都不会重新收集。
  • samber/dev-event-organizer-skills@event-venue-sourcing
    - 收集本 skill 赖以构建的场地侧能力答案；它把技术规格转到这里，本 skill 再以书面问题形式交回。
  • samber/dev-event-organizer-skills@event-budget
    - 接收已拆分为录制、字幕和直播的 Media 成本行，并决定它如何适配整体形态。
  • samber/dev-event-organizer-skills@event-vendor-sourcing
    - 根据这里编写的技术规格选择并签约 AV 供应商；绝不要让一个 skill 同时做两件事。
  • samber/dev-event-organizer-skills@event-schedule-design
    - 确定团队在其中工作的日程网格；不适合某种形式的房间要上报上游，绝不要通过砍掉一场议程来解决。