Don't spend three months building something nobody wants. Demand Radar validates the demand for you: real evidence from both top-down and bottom-up, cross-checked against each other, then a red-team set loose on the weak points of your idea. Instead of dumping data on you, it gives a plain-language verdict — Go / Conditional / Pivot / No-go — plus the very first thing to do next. 别花三个月，做一个没人要的东西。Demand Radar 替你把需求验一遍：自上而下和自下而上两边的真实证据交叉对账，再放一个红队专挑你想法的硬伤。最后不甩给你一堆数据，而是一句人话裁决——可以做 / 有条件做 / 转向 / 别做，外加现在第一步该干什么。
Validating demand usually means opening a dozen tabs, keeping the answers you already liked, and going all in. Demand Radar does the opposite: multi-source evidence cross-checked, a quantified score, and a red-team dead set on killing your idea. What you walk away with isn't "looks good" — it's a conclusion that survived an attempt to disprove it, and a next step you can act on.查需求，多数人就是开十几个标签页，最后只留下自己爱看的那几条，然后梭哈。Demand Radar 反过来：多源证据交叉验、量化打分、再上一个铁了心要否掉你的红队。你拿到的不是「看着挺好」，而是一个扛过了证伪的结论，和一份能照着改的下一步。
Reddit complaints aren't enough. It watches top-down (reports, market size, funding) and bottom-up (HN, app reviews, search, social) at once. Where the two don't line up — "analysts call the market huge, yet nobody in the community complains" — is usually what you most need to know.光看 Reddit 抱怨不够。它同时盯自上而下（行业报告 / 市场规模 / 融资）和自下而上（HN / 应用商店评论 / 搜索 / 社交）。两边对不上的地方——「分析师说市场巨大，可社区里没人抱怨」——往往才是你最该知道的真相。
You see your idea through rose-tinted glasses; the red-team doesn't. This agent's only job is to kill your hypothesis — fake demand, saturated, shrinking, nobody pays, can't reach them. Every "Go" has to survive it. In testing it caught a buyer mismatch and dropped a verdict from Conditional 55% to Pivot 40% in one move — so you see the fatal flaw before you write a line of code.你对自己的点子有滤镜，红队没有。这个 agent 唯一的任务就是杀死你的假设：伪需求、已饱和、在萎缩、没人付钱、根本够不着。任何「可以做」都得先扛过它。实测里，它揪出一个买家错配，一招把「有条件做 55%」打到「转向 40%」——让你在写第一行代码前就看见硬伤。
Willingness-to-pay counts double — whether it can make money comes first. A signal needs ≥2 independent sources to count; one source stays a lead. Three judges score separately and disagreements get flagged red, never averaged away. So the score you get holds up to scrutiny — not a gut call.付费意愿权重翻倍——能不能赚钱是头等大事；一个信号至少 2 个独立来源才算数，单源只能当线索。3 个评委各打各的，分歧直接标红，不和稀泥。所以你拿到的分数，是经得起追问的，不是拍脑袋。
The first screen answers three things: does your demand hold up, why, and what to do next — with action cards you can run with (switch buyer, narrow the use case, pre-sell). Scores and the full evidence trail fold in behind, every claim one click from its source. The decision-grade info is right there at a glance.报告第一屏就回答三件事：你的需求成不成立、为什么、接下来该干什么——还配上能直接照做的行动卡（换买家、收窄场景、做预售）。评分和完整证据链折在后面，每条结论点一下就溯源。该决策的信息，一眼就拿到。
The result isn't scattered links and half-finished notes — it's one self-contained HTML file you can send straight to a co-founder or your team. And gathering the evidence needs no sign-up, no OAuth, no API keys: free official endpoints plus web search do it, so you skip the config wrangling and login walls.结果不是散落的链接和半成品，是一个自包含的 HTML 文件，直接发给合伙人或团队就能看。而且取证不用注册、不用授权、不用 API key，免费官方接口加网页搜索就把证据取齐，省去折腾配置和卡在登录的麻烦。
Ask in Chinese, get a Chinese report; ask in English, get English. No bending to the tool, no translating back and forth — validating demand is something you should be able to spell out in whatever language comes easiest.中文问出中文报告，英文问出英文报告。不用迁就工具、不用来回翻译，验需求这件事，本来就该用你最顺手的语言说清楚。
Paste into Claude Code, say your idea, and minutes later you've got a verdict-first, traceable report:粘进 Claude Code，说出你的点子，几分钟后就能拿到一份带裁决、能溯源的报告：