Jev Python 教程：10 分钟上手 AI 决策与工单路由

简要描述Brief Description

Jev Python 教程：10 分钟学会 AI 决策Jev Python Tutorial: AI Decisions in 10 Minutes

目录Table of Contents

• 概述
  Overview
• HubSpot
  HubSpot
• Jev 速查表
  Jev Cheatsheet
• 向 Jev 发送请求
  Sending a Request to Jev
• 选择决策类型
  Choice Decision Type
• 评分决策类型
  Score Decision Type
• 包含多个问题的完整示例
  Full Example with Multiple Questions

概述Overview

在这段视频中，我会教你如何在 Python 中使用 Jev。Jev 是一种不会生成文本、而是实际做出决策的模型。首先，我会用几行代码演示如何发起第一次 Jev 调用，然后我们会介绍它可以做出的三种决策：先用 choice 选择一个选项，再用 null 回答是或否的问题，最后用 score 按照某个量表为事物打分。最后，我们会把这些功能整合起来，搭建一个简单的 AI 支持工单收件箱，让你看看如何在现实场景中使用它。Jev 不仅非常准确，而且成本极低；当你需要做大量决策，而又不想把这些任务交给 LLM 时，它会是一个绝佳的选择。话不多说，先听一下今天赞助商的简短介绍，然后我们就开始。In this video, I'll teach you how to use Jev from Python. Now, Jev is a model that doesn't write text, but it actually makes decisions. Now, first, I'll show you how to make your first Jev call in just a few lines of code, and then we'll go through the three types of decisions that it can make. Now, we'll start with choice to pick an option, then null for a yes or no questions, and then score to be able to rate something on a scale. And finally, we'll put all of it together to build a simple AI support inbox you can see how you would use this in the real world. Now, Jev is not only very accurate, but it's also extremely inexpensive, and it's a great choice when you need to make a ton of decisions that you wouldn't normally want to pass to an LLM. So, with that said, let's dive in here after a quick word from today's sponsor.

HubSpotHubSpot

今天的视频由 HubSpot 赞助。我们还一起准备了一些我觉得你真的会用得上的东西。它叫做“支撑我运营业务的 AI 工具栈”，名副其实。这是一个免费的工具包，里面有我每天用来运营这个频道的 9 款 AI 工具，还有 6 个经过精心设计的提示词模板，能让 AI 先提出澄清问题，以及 5 个可直接使用的 Claude 技能，可用于研究、制作演示文稿和安排任务优先级等。今天这期视频里，我们会把 Jev 接入 Python；而这个工具包里的工具栈，正是我用来规划、撰写脚本并制作本频道所有内容（包括这期视频）的那套工具。无论你是开发者、创始人还是内容创作者，只要 AI 是你日常工作的一部分，里面总有适合你的东西。完全免费，链接就在简介和置顶评论里。Today's video is sponsored by HubSpot. And together, we put together something that I think you'll actually use. Now, it's called the AI stack that I run my business on. And it's exactly that. It's a free bundle with the nine AI tools that I personally use every day to run this channel. It would also have six prompt templates engineered to make AI ask clarifying questions, and five drop-in Claude skills that handle things like research, building presentations, and prioritizing your tasks. Now, in this video, we're wiring Jev into Python, but this bundle is the stack that I actually used to plan, script, and ship everything that you see on this channel and in this video. Now, whether you're a developer, a founder, or a creator, if AI is part of your day, there's something in there for you. Now, it's completely free, and the link is in the description and the pinned comment.

看看吧。看看能不能从中获得一些价值。Check it out. See if you can get some value from it.

Jev 速查表Jev Cheatsheet

现在我们回到视频内容。接下来的几分钟里，你将学习如何在 Python 中使用 Jev。Jev 是一个决策模型。要使用它，你需要提供一些状态，然后就可以针对该状态提出各种问题。你可以提出 null、choice 或 score 类型的问题。基本上，你可以问它是或否的问题，让它从类别或标签中进行选择，也可以让它按一定的尺度给某项内容打分。发送请求时，内容大致会是这样。我会在这里使用 OpenRouter 来发送这些不同的决策请求，不过你也可以使用任何你想用的工具，或者 Jev 的官方 API。往下滚动，响应大致会是这样。别担心，我会详细讲解。And what we'll do now is get back into the video. So, in the next few minutes, you'll learn how to use Jev from Python. Now, Jev is a decision model, and in order to use it, you need to provide some state, and then you can ask various questions about that state. Now, the questions that you can ask are a null, choice, or score question. Basically, you can ask it a yes or no question. You can ask it to choose a category or a label, and you can ask it to rate something on a scale. Now, when you send a request, it will look something like this. What I'm going to be doing is using OpenRouter here to send these different decisions or send these requests, but you can use anything that you want or Jev's official API. If we scroll down, the response will look something like this. Don't worry, I'll go through it in detail.

我先快速带你看一下接下来会看到什么。基本上，你会发现可以提出多个不同的问题，把它们放到这里的 question 数组中，然后就会得到多个不同的答案。要在 Python 中使用它，你只需查看具体答案，再将其与某种基准值或阈值进行比较，从而采取相应操作。下面还有一些需要注意的地方，你可以仔细看看；不过，我们先快速进入正题，一步一步讲解，让你准确了解它的工作原理。首先，我们来看看如何向 Jev 发送请求。要做到这一点，你需要一个 Jev 服务提供商的 API 密钥。我推荐使用 OpenRouter，本视频中我也会使用它。I'm just giving you a quick look at what you're going to see. And basically, well, you'll notice that you can ask multiple different questions, which goes in this question array here, and then you'll get back multiple different answers. Now, in order to use it in Python, what you can do is simply look at the specific answer and then compare it to some kind of baseline or threshold in order to actually take some actions. Now, a few gotchas down here if you want to read through them, but what we'll do is just quickly jump in here and we'll go step by step so you see exactly how it works. So, let's first just look at sending a request to Jev. Now, in order to do this, you will need an API key for some kind of provider for Jev. I recommend OpenRouter, and that's what I'll be using in this video.

所以，我做的就是访问 OpenRouter 网站，注册一个账号，获取 API 密钥，剩下的就非常简单了，而且你可以通过代码来完成这些操作。So, what I did is just went to OpenRouter's website, made an account, got an API key, and then the rest is very straightforward, and you can do this from code.

向 Jev 发送请求Sending a Request to Jev

这就是向 Jev 发送请求的方式。注意，我们在这里引入了几个导入项，比如 JSON、OS 和 requests。然后是我们的消息，它实际上会充当状态。在这个例子中，它是一条客户消息：“你好，我上周的订单被扣了两次款。能退回其中一笔吗？”我们在这里要做的是说：“好，我们要向 OpenRouter 发送一个请求。”所以我们会说：“嘿，这是 OpenRouter 决策 API。”我们传入 API 密钥，然后指定要调用的模型。显然，这部分是 OpenRouter 特有的，因为我们调用的是 Jev。从这之后，无论你使用哪个提供商来调用 Jev，其余部分都会完全相同。我们传入状态，也就是客户消息，然后还可以传入一个问题数组。This is how you send a request to Jev. Notice we brought in a few imports here like JSON, OS, and requests. And then we have our message, which is really going to act as the state. So, in this case, it's an example of a customer message. So, "Hi, I was charged twice for my order last week. Can I get one of those payments back?" What we do here is say, "Okay, we're going to send a request over to OpenRouter." So, we say, "Hey, this is the OpenRouter decisions API." We pass in our API key, and then we start by specifying the model that we want to call. So, this is specific, obviously, to OpenRouter because we're calling Jev. From there, the rest will be the exact same regardless of whatever provider you're using to call Jev. So, we pass our state, which customer message, and then we can pass an array of questions.

现在，我们可以提出多个问题，并为每个问题随便取个名字。当然，最好取个有意义的名字，但在这个例子中，我们把第一个问题标记为 refund。所以，我们就这样称呼它。问题的名字可以由你随意取。接下来，我们指定问题的类型，必须是之前看过的三种类型之一。让我打开这里看一下。我们有 null、choice 或 score。然后，我们传入指令，告诉模型我们希望它做什么。我说的是：“根据这个状态，客户是在要求退款吗？”然后我还设置了一些条件。这里的条件是 true 或 false，因为类型是 null。true 表示他们想要退回款项，false 表示他们想要别的东西。接下来，Jev 会返回一个介于 0 到 1 之间的数值，其中 1 表示 true，0 表示 false。Now, we can ask multiple questions, and each question we can just give a random name. Well, name it something that makes sense, but in this case, the first question we've labeled is refund. So, we've just called it this. You can name the questions anything that you want. We then specify the type of the question, which has to be one of those three that we looked at before. So, let me just open this up. So, again, we have null, choice, or score. Then from there, we pass the instructions. So, we're asking the model what it is we want it to do. So, I'm saying, "Is the customer asking for a refund based on this state?" And then I have some criteria. Now, I have a true or false criteria here because it's null. For true, they want the money back. For false, they want something else. What Jev will do then is it will give us a response back with a number between zero to one where one represents true and zero represents false.

这表示它认为这件事为真或为假的可能性有多大，或者回答“是”或“否”的可能性有多大，因为我们问的就是这类问题。基本上，我们要做的就是在这里获取响应，对吧？我们会把它直接转成 JSON，这样就能将其加载为 Python 对象，然后查看 `answers` 中的 `is refund`，再看这里的 null 响应，也就是 null 字段，这就能得出答案，最后我们只需把它四舍五入成百分比。所以，如果我运行这段代码，我们来看一下，执行 `UV run first call.py`，就会看到响应，而且应该非常快。注意，这正是我们的响应格式：其中有模型，对吧？还有 answers，所以我们有 `is refund`，类型是 null，然后 null 值是 0.98。接着它会给出用量信息，这是来自 OpenRouter 的，并告诉我们费用，实际还不到 1/1000 美分。And that indicates how likely it believes that this is either true or false or yes or no because that's the type of question that we're asking. So, basically, what we'll do is we'll get the response here, right? We'll just dump this into JSON so we can load it as a Python object, and then we're going to look at answers is refund and then the null response here, the null field, and that's going to give us our answer, and we'll just round that to a percentage. So, if I run this code, let's go here, UV run first call.py, we'll see that we get the response and it should be quite fast. So, notice this is exactly what our response looks like where we have the model, right? We have the answers, so we have is refund type null, and then the null value is 0.98. It then gives us the usage, this is from OpenRouter, and tells us the cost, which is literally less than a 1/1000 of a cent.

然后我们可以看到，这是一项退款请求的概率为 98%，因此我们可以将其转交给退款团队。这是第一个示例。我们来看下一个。And then we can see here that the chance that this is a refund request is 98%, and we would be able to route that over to the refund team. That's the first example. Let's go to the next one.

选择决策类型Choice Decision Type

现在我们已经了解了如何向 Jev 发送请求。我在这里创建了一个小型辅助函数，建议你也自己创建一个。它就是这里的 ask 函数，我们向其中传入 state 和一些问题，由它负责在这里向 OpenRouter 发送请求，这样我们就不用亲自处理了。现在，我们来看看选择示例。你可以提出的下一种问题是选择题。在这个示例中，我们有多条消息，会逐条传入；同时，我们还有一个想针对这些消息提出的问题。我们把这个问题标记为 team，类型为 choice。我们还提供了具体指示：哪个团队应该处理这条客户消息？然后，我们给出了四项标准。这样，Jev 就可以从中选出它认为应该负责处理这条消息的团队，并给出相应的概率。Now that we've seen how to send a request to Jev, what I've done is just created a tiny helper function here, which I would recommend you do yourself, that just is an ask function here where we take in a state and some questions, and it handles sending the request over here to OpenRouter so we don't need to do it ourselves. Now, let's go over to the choice example. So, the next type of question you can ask is a choice. So, in this case, we have multiple messages. We're going to pass all of them one by one, and we have one question that we want to ask about this. Now, that question we've labeled team, the type is choice. We've given the instruction, right? Which team should handle this customer message, and then we've given four pieces of criteria. So, Jev will now be able to pick which one it believes should handle this criteria and give us the probability for it.

所以，我们有 shipping、tech、billing 和 sales，然后还有一段描述，这样 Jev 就能根据这里的描述值判断应该归到哪一类。接着我们用 `for message in messages`，也就是遍历这里的四条消息。我们调用辅助函数，把 state 和 question 传进去，然后打印答案，对吧？所以，我们在这里获取团队答案并打印出来。现在我来运行给你看。好，你可以看到它很快就给出了响应。比如：“我的包裹显示已送达，但门口没有”，归类为 shipping；“我无法登录，一直提示密码错误”，归类为 tech；“为什么收了我 49 美元，而我的套餐明明是 29 美元”，归类为 billing，对吧？“这件夹克有中码吗？”你大概明白了。如果我们想快速打印出实际的响应，可以到这里的辅助函数中，把所有数据都打印出来，这样就能更细致地查看，也能准确看到返回的响应。So, we have shipping, tech, billing, and sales, and then a description so Jev knows which one it should classify based on the description values here. We then have for message in messages, so we're just looping through all four of our messages here. We're using our helper function and then just passing in the state and the question and then we're just going to print out the answer, right? So, we take the team answer and print it out here. So, let me run this for you. Okay, and you can see that it gives us the responses very quickly here. So, my package says delivered, but it's not at my door, shipping. I can't log in, it keeps saying my password is wrong, tech. Why was charged $49 when my plan is $29, billing, right? Do you have this jacket in size medium, you get the idea. And if we want to quickly print out what the actual response was, if we go over here to this helper function here, we can just print all of the data, so we can kind of view it a little bit more granularly and you can see exactly the response that we get back.

好，我刚才把它打印出来了，这样你就能看到它是什么样子。基本上，我们会得到答案；对于 team，type choice 会告诉我们选择是什么，并给出一个概率列表。在这个例子中，我们只得到一个概率，因为模型对这种情况非常有把握。不过，如果它没那么有把握，我们得到的置信度分数就不会是 1，这里也会出现不同的数值。比如可能是 0.1、0.7、0.8，诸如此类。你明白我的意思吧：这种概率分布中的所有数值加起来会等于 1。这就是 choice，接下来我们看下一个。Okay, so I just printed it out here so you can see what it looks like, but basically we get our answer, for team, type choice tells us what the choice is, gives us a list of probabilities, in this case we just got one because it really is like very confident that this is the case. However, if it was less confident, then we would get a confidence score that is not one and then we would get different values here. So, it might get 0.1, 0.7, 0. 8, whatever, you get the idea where all of them will sum up to one for that kind of probability distribution. So, that is choice, let's go to the next one.

评分决策类型Score Decision Type

我们已经看过 null，现在要转到 score。再说一次，我们要从 Jev 模块导入 ask，也就是那个辅助函数。我们有一些可能想要询问相关问题的工单，而我想做的是根据紧急程度给这些工单评分。因此，我设置了 score 类型，写下指令“支持工单有多紧急”，而这次对于 criteria，我只是传入一个包含三个不同值的数组。基本上，第一个值是 0，第二个是 1，第三个是 2，就像数组中的索引一样。这里可以设置 2 到 10 个不同的 criteria，然后 Jev 就会按照相应的刻度评分。所以在这个例子中，刻度范围是 0 到 2，因为我们有三个 criteria。不过，如果这里再加一个 criteria，比如“非常紧急”之类的，那么刻度范围就会变成 0 到 3，对吧？We've already looked at null, but now we'll move over to score. Now again, from the Jev module, so just that helper, we're going to import ask. We have some tickets that we might want to ask a question about and what I want to do here is rate the tickets based on urgency. So, I put the type of score, I put the instructions, how urgent is the support ticket and this time for the criteria, what I'm doing is just passing an array of three different values. Now basically, the first value will be zero, the second will be one, the third will be two, just like the indexes that we would have in an array. And I can have between two and 10 different criteria here and Jev will then rate it on a scale. So, in this case the scale will be between zero and two because we have three pieces of criteria. However, if we had another piece of criteria here, like maybe very urgent or something, then the scale would go to zero to three, right?

因为现在我们有四项标准。所以大致就是这么运作的，你马上就会看到。不过想象一下，你有一个刻度尺，上面把这些都作为标签，间隔大致相等。我们要做的是：对于 tickets 中的每张工单，获取答案和成本。所以，我们这里只是在向 Jev 提问，然后查看紧急程度。正如这里的注释所说，分数是各个等级的加权平均值。我们马上来看一下。在这个例子中，像 1.7 这样的分数表示大多数情况下属于紧急。因此，它并非极其紧急，而是介于正常和紧急之间。因为基本上，我们会有“不紧急”，其值为零；然后是“正常”，其值为一；最后是“紧急”——如果拼写正确的话——其值为二。Because now we have four pieces of criteria. So, that's kind of how it works, you're going to see in one second, but imagine you have again a scale with all of these as labels that are evenly kind of separated in So, what we're doing is saying for ticket in tickets, we're going to get our answers and our cost. So, we're just asking the question here to Jev, and then we're looking at the urgency. Now, as the comment says here, the score is a weighted average of the levels. We'll look at this in 1 second. So, something like 1.7 in this example means mostly urgent. So, not extremely urgent, but in between normal and urgent. Because basically, we would have not urgent, which is value zero. Then, we would have normal, which is value one. And then, we would have urgent, if we spell that correctly, which is value two.

明白吗？所以，如果分数介于 1 和 2 之间，就表示紧急程度高于正常，但还没到最紧急。如果分数介于 0 和 1 之间，就表示紧急程度介于不紧急和正常之间。如果分数是 1，那就正好是正常。明白吗？这就是这个量表的运作方式。我们来运行一下，看看结果如何。好，我已经运行了。你可以看到：“想问一下，你们的路线图里有深色模式吗？”分数为 0，不紧急，属于一般问题，对吧？“我的订单晚了几天，有更新吗？”分数为 0.96，对吧？属于正常情况，是个实际问题，但可以等一天，置信度为 94%。“我们整个团队都无法登录，而且 20 分钟后就要给客户演示了。”分数为 2，极其紧急，有人因此无法开展工作，需要立即获得帮助。同样，如果我们快速把结果打印出来，就能清楚地看到响应是什么样的。Okay? So, if we have something between one and two, it's more than normal, but less than fully urgent. If we had something between zero and one, then again, it's between not urgent and normal. If we had one, that would be exactly normal. Okay? So, that's how the scale works. So, let's run this here and see what we get. Okay, so I've run it, and you can see just wondering if you have a dark mode on the road map, score zero, not urgent, general question, right? My order is a few days late, any update? Score 0.96, right? Normal, a real problem, but it can wait a day, 94% confidence. Our whole team is locked out, and we have a client demo in 20 minutes. Score two, extremely urgent, someone's blocked, they need help immediately. And same thing, if we quickly just print this out, you can see exactly what the response looks like.

好，所以你可以看到，我们在这里得到了所有响应。基本上，我们有图例：0、1、2，就像我之前提到的，这是我们为评分尺度提供的标记。我们有分数，这个例子中的分数是零，然后是概率。根据这里的图例，概率是 1、0、0，接着是置信度。 现在，如果我们向下滚动，找到一个分数并非正好是 1 或 2 之类的例子，你会看到分数是 0.96。同样，我们有图例，现在还有概率。它显示：“嘿，这个被评为 0 的概率是 0.05，也就是 5%。”然后，它被评为 1 的概率是 95%。因此，我们取这些值的加权平均，得出总体分数是 96，对吧？也就是 0.96。接着，它给出的置信度是 93%。 继续往下看，分数是 2，然后当然，你可以看到这里有一个置信度为 1 的例子。Okay, so you can see that we get all of the responses here, and basically, we have the legend, 0 1 2, like I was talking about for the scale that we provided. We have the score, so in this case, the score is zero, and then the probabilities. So, we had 1 0 0 for the probabilities based on the legend right here, and then a confidence. Now, if we scroll down to one that wasn't exactly one or two or whatever, you can see we have a score 0.96. Again, we have the legend, and now we have the probabilities. So, it said, "Hey, it's 0.05 or 5% probability that this is actually rated zero." And then, it's 95% probability that this is rated one. So, we take the weighted average of that, and that means the overall score is 96, right? 0.96. Then, it gives us the confidence of 93%. Continuing here, we have a score of two, and then of course, you can see we have one here with a confidence of one.

所以，置信度告诉我们这种情况实际发生的可能性有多大。如果你提出的问题有歧义，得到的置信度就会比较低。我专门为此做过一整期视频，不过简单来说，Jev 很擅长处理相对简单的任务。但如果问题特别复杂，Jev 要给出正确的分类可能会有些困难。那么现在，我们再来看一个完整示例，然后就结束这期视频。So, the confidence tells us how likely this actually is the case. And if you ask ambiguous questions, you will get less confidence level. I've actually made an entire video on this, but basically Jev is very good for relatively simple stuff. If you do super complex questions, though, it can be a little bit challenging for Jev to give the correct classification. So, now let's look at one more kind of full example, and then we'll end the video.

包含多个问题的完整示例Full Example with Multiple Questions

我刚刚在这里写了一个脚本，基本上就是针对这些不同的工单提出多个问题。比如，我们可以看到有一条是“我的订阅被重复扣费了”，另一条是“应用崩溃了”，等等。这次我们要一次提出多个问题。我们会询问这条工单应该交给哪个团队处理、用户是否想要退款，以及问题的紧急程度。和之前一样，我们有 choice、null 和 score 这些类型，还有指令和标准，并且可以同时提出所有这些问题。接下来，我们可以针对工单和收件箱提出所有这些问题。我们还可以查看成本，然后获取所有不同问题的答案：负责团队、置信度、是否想要退款，以及紧急程度。然后，我们可以设定一个置信度阈值：如果置信度低于这个阈值，那就说明我们可能不太确定，这时就把工单转交给人工处理。I just made a script here where we basically ask multiple questions about these different tickets. So, we can see that I have, you know, I was charged twice for my subscription, the app crashes, whatever. And this time we're going to ask multiple questions in one. So, we have a question about what team this should go to, a question about if they want a refund, and a question about the urgency. So, same thing, we have our types, choice, null, and score. We have our instructions and then our criteria, and we can ask all of these at the same time. So, if we go, we can say for ticket and inbox, we can ask all of these questions. We can figure out what the cost is as well, just to see that, and then we can get all of the different answers. So, the team, the confidence, the wants refund, and the urgency. And then we can say if the confidence is below a certain threshold, okay, maybe we're not sure, and we're going to send that over to a human.

不过，如果置信度高于某个水平，也许我们就会让 AI 实际处理这件事。所以，我在这里做了一个检查：好吧，我们来看看 Jev 的回答，再根据它的置信度，决定是信任这个回答还是不信任。接下来，我们可以输出工单内容、应该转交给哪个团队、置信度、用户是否要求退款，以及紧急程度，然后其他情况也一样，对吧？这大致就是这里的 if 语句的工作方式。现在我们来运行一下，看看这些不同工单会得到什么回复。运行 UV run 05，看看结果。好了，回复陆续出来了。你可以看到：“我这个月的订阅被扣了两次款，请退还其中一笔。”对应的团队是 billing，退款为 true，紧急程度为 94%。However, if the confidence is above a certain level, maybe we will allow the AI to actually handle that. So, what I've done here is I've kind of done a check to say, "All right, let's look at Jev's responses, and based on how confident it is, we'll either trust that response or not trust that response." Then what we can do is print out the ticket, the team that this should go to, the confidence, if they want a refund, and what the urgency level is, and then same thing, right? That's kind of how the if statement works here. So, let's run this and see the response that we get for these different tickets. So, UV run 05, and let's see what we get. Okay, so responses are coming in. You can see I was charged twice for my subscription this month, please refund one of them. We have team billing, refund true, urgency 94%.

已加入退款队列，对吧？好，我每次在 Android 上打开设置页面，应用都会崩溃。团队类别是技术，是否退款：否，紧急程度：1.63。嘿，我很喜欢这款产品，你们有推出团队版的计划吗？团队类别是账单，但它不太确定，置信度是 44%，对吧？是否退款：否，紧急程度：0.14。它对此不太确定，所以会把这件事转给人工处理。好，我们整个公司都无法登录，对吧？嘿，紧急程度是 2，是否退款：否，团队类别是紧急，置信度 100%，通知团队。你明白我的意思了吧。我的包裹原定周一送达，现在都周四了。团队类别是物流，是否退款：否，紧急程度大约是 1，属于普通级别，对吧？然后把这些都分类，大概花了我们千分之一美分，或者万分之一美分吧。好了，大家，我要跟你们分享的内容就这些。希望这段视频对你们有帮助，期待下期再见。Added to the refund queue, right? Okay, the app crashes every time I open the settings page on Android, team is tech, refund false, urgency 1.63. Hey, love the product, any plans for a team plan? Team billing, it's not really sure, 44%, right? Refund false, urgency 0.14. It's not sure about it, so it's going to send that over to a human. Okay, our entire company can't log in, right? Hey, urgency two, refund false, team 100% urgent, page the team. You get the idea. My package was supposed to arrive Monday, it's now Thursday. Team shipping, refund false, urgency about one, normal, right? And then this cost us about 1/1000 I guess, or 1/10000 of a cent to classify all of those. Okay guys, so that's all that I had for you. I hope that this video was helpful, and I look forward to seeing you in another one.