Scorched Earth 2000 v1.1 是经典坦克对战游戏 Scorched Earth 的现代 JavaScript 版本，由 KAOS Software Team 开发，于 2026 年 5 月 11 日发布。游戏采用回合制炮战玩法，玩家在二维战场上操控坦克，调整角度和力度向对手发射武器，同时要考虑风力等环境因素。界面显示当前为 Player 1 的回合，已选中 Missile 武器，弹药为 999，力度设为 300，角度为 30°，风力强度为 9 。调试控制台显示了详细的游戏状态信息，包括玩家位置、现金（各 50,000），并确认当前为离线多人模式，共有四位玩家：两名人类（Player 1 和 Player 2）及两名 AI（Shooter 和 Cyborg）。

多人模式允许玩家创建或加入游戏，并可自定义分辨率（640×480 、 800×600 或 1024×768）、风力（无、微风、正常或强风）、初始现金和回合数。游戏浏览器显示多个活跃房间，配置和人数各异，有的已在进行中（例如第 2/3 回合），有的在等待玩家加入。玩家可以添加不同难度的 AI，包括 Shooter 、 Cyborg 和 Killer 变体。网络功能正常，房间显示玩家名、连接状态及游戏参数。

库存与商店系统允许玩家在回合间用赚取的现金购买武器和装备。统计系统记录每位玩家的击杀、收益及综合表现。游戏包含多种坦克型号（Tank 1 至 Tank 6），最多支持 8 名玩家，AI 数量可设为 0 到 7 。单人模式提供类似的自定义选项，便于玩家对抗 AI 练习。

开发团队名单显示这是个协作项目，由 Hei C. Ng（xixi）担任项目主管，Alexander Rasin 任开发主管，另有多位程序员、物理学家和 QA 测试人员参与。团队感谢 Wendell Hicken 创作原版 Scorched Earth，并对多位网络协助者、色彩搭配建议者表示特别感谢，甚至感谢为团队命名提供灵感的 Roger Waters 。游戏在保留经典 Scorched Earth 玩法的同时，加入了在线多人、增强图形选项和具不同行为模式的 AI 对手等现代特性。 Scorched Earth 2000 v1.1 is a modern JavaScript port of the classic tank combat game Scorched Earth, developed by the KAOS Software Team and released on May 11, 2026. The game features turn-based artillery battles where players control tanks on a 2D battlefield, adjusting angle and power to fire weapons at opponents while accounting for environmental factors like wind. The interface shows Player 1's turn with a Missile weapon selected, 999 ammo, power set to 300, angle at 30 degrees, and wind strength of 9. The debug console reveals detailed game state information including player positions, cash amounts (50,000 each), and confirms the game is currently in offline multiplayer mode with four players: two human players (Player 1 and Player 2) and two AI opponents (Shooter and Cyborg).

The multiplayer functionality allows players to create or join games with customizable settings including resolution (640x480, 800x600, or 1024x768), wind conditions (none, light, normal, or strong), initial cash, and number of rounds. The game browser shows numerous active sessions with varying player counts and configurations, some already in progress (like one in round 2/3) and others waiting for players. Players can add AI opponents with different difficulty levels including Shooter, Cyborg, and Killer variants. The networking appears functional with games displaying player names, connection status, and game parameters.

The inventory and shop system allows players to purchase weapons and equipment between rounds using earned cash. The statistics tracking shows kills, gains, and overall performance metrics for each player. The game includes various tank models (Tank 1 through Tank 6) and supports up to 8 players total, with flexible AI player counts from 0 to 7. The single-player mode offers similar customization options for players wanting to practice against AI opponents.

The development credits reveal this is a collaborative project led by Hei C. Ng (xixi) as Project Lead and Alexander Rasin as Development Lead, with contributions from multiple programmers, physicists, and QA specialists. The team acknowledges Wendell Hicken for the original Scorched Earth game and includes special thanks to various contributors for networking help, color matching hints, and even Roger Waters for the team name inspiration. The game maintains the classic Scorched Earth gameplay while adding modern features like online multiplayer, enhanced graphics options, and AI opponents with different behavioral patterns.

Meta 正在 Threads 上试行一项新功能，允许用户 @ 一个专门的 Meta AI 账号，以获取对话相关的问题答案或背景信息。该功能类似于用户在 X 上 @ xAI 的 Grok，目前在阿根廷、马来西亚、墨西哥、沙特阿拉伯和新加坡进行测试。用户可以向 AI 提问诸如"为什么大家都迷抹茶"或"‘Cannes'到底怎么读？"之类的问题。然而，Threads 用户很快发现无法屏蔽该 Meta AI 账号，因而引发强烈反弹。

无法屏蔽这一设置激起了广泛愤怒。许多人在 Meta AI 账号、 Threads 官方账号以及 Threads 负责人 Connor Hayes 的帖子下表达不满。试图屏蔽该账号的用户报告遇到错误，有报道称个人资料菜单中根本没有屏蔽选项。许多人认为，这种被强制存在的 AI 账号是一种侵入性越权行为。

此举是 Meta 更大规模、积极推进人工智能战略的一部分，公司正努力与 OpenAI 和 Google 等竞争对手抗衡。 Meta 已在 AI 领域投入数十亿美元，大量招揽顶尖人才并推出像 Muse Spark 这样的新模型。尽管用户对该功能的上线明显表示抵制，但在 Threads 上推出 Meta AI 账号仍是其将 AI 技术直接整合进应用和服务套件的又一步。 Meta is testing a new feature on Threads that allows users to tag a dedicated Meta AI account to get answers to questions or context about conversations on the platform. This feature, which is similar to how users can tag xAI's Grok on X, is currently being trialed in Argentina, Malaysia, Mexico, Saudi Arabia, and Singapore. Users can ask the AI things like "why is everyone obsessed with matcha" or "how do you actually pronounce 'Cannes'?" However, Threads users quickly discovered that they cannot block the Meta AI account, leading to significant backlash.

The inability to block the Meta AI profile has sparked widespread anger among Threads users. Many took to the platform to express their frustration in replies to posts from the Meta AI account, the main Threads account, and Threads boss Connor Hayes. Users who attempted to block the account reported encountering errors, and reports confirm that the option to block is simply not available in the profile's menu. This forced presence of the AI account is seen by many as an intrusive overreach by the company.

This move is part of Meta's broader and aggressive push into artificial intelligence, as the company works to compete with rivals like OpenAI and Google. Meta has invested billions in AI, including spending heavily on hiring top AI talent and launching new models like Muse Spark. The introduction of the Meta AI account on Threads represents another step in integrating its AI technology directly into its suite of apps and services, despite clear user resistance to the feature's implementation.

2026 年 5 月 11 日，Princeton University 的教职员工投票决定自 7 月 1 日起对所有线下考试强制实施监考，结束自 1893 年以来持续了 133 年的荣誉制度。该政策以仅一票反对获得通过，要求教师在考场内作为见证人，但不得干预学生答题。监考人员将记录疑似违规行为并上报由学生组成的 Honor Committee，必要时可出庭作证。此举是在数月讨论后做出的，主要由对学术诚信日益担忧所驱动，尤其是人工智能工具和个人电子设备的普及使作弊更难被察觉。

最初的荣誉准则是在学生请愿取消监考后制定的；一个多世纪以来，Princeton 的制度依赖个人自律与同伴举报。但新政策指出，观察不当行为越来越困难，且学生因担心网上曝光或人肉搜索而不愿直接举报同伴。 2025 年高年级学生调查显示，近 30% 的受访者承认曾作弊，44.6% 的人知道违规行为却选择不举报，仅有 0.4% 的人曾举报过同伴。这些趋势，加上生成式 AI 等新挑战，促使学校转向正式监考。

该提案由学院院长 Michael Gordin 牵头，得到了包括 Honor Committee 、本科生院长办公室和 Undergraduate Student Government 在内的主要学生与教职团体的支持。虽然部分学生和教职员担心监考会削弱信任，但也有人认为这是必要的调整。 2026 年 11 月已有一项政策要求对个人和小组考试实施监考，为这次更广泛的改变奠定了基础。 Honor Committee 的章程将保持不变，但学校中关于禁止监考的规定会相应更新以反映新要求。

支持者认为，监考虽不能完全杜绝作弊，但能起到威慑作用，并减轻学生在考试期间监督同伴的负担。前 Honor Committee 主席 Nadia Makuc '26 表示，委员会长期考虑将监考人员作为额外见证人，尤其是在案件增加和 AI 带来新问题的情况下。前同伴代表联合主席 William Aepli '26 预计，听证会上呈交的证据类型将发生变化。教职员工反应不一，教授 Jill Dolan 称此举"虽遗憾，但有必要"，并承认这标志着 Princeton 文化的一次重大转变。 On May 11, 2026, Princeton University faculty voted to mandate proctoring for all in-person examinations starting July 1, ending a 133-year-old honor system that had relied on student self-governance since 1893. The policy, passed with only one opposing vote, requires instructors to be present in exam rooms as witnesses but prohibits them from interfering with students. Suspected violations will be documented by proctors and reported to the student-run Honor Committee, where they may later testify. This change follows months of deliberation driven by rising concerns over academic integrity, particularly due to the proliferation of AI tools and personal electronic devices that make cheating harder to detect.

The original Honor Code was established after a student petition to eliminate proctoring, and for over a century, Princeton's system depended on individual accountability and peer reporting. However, the new policy cites increased difficulty in observing misconduct and a growing reluctance among students to report peers directly, fueled by fears of online shaming or doxxing. A 2025 Senior Survey found that nearly 30% of respondents admitted to cheating, while 44.6% knew of violations they chose not to report, and only 0.4% had ever reported a peer. These trends, combined with challenges like generative AI, prompted the shift toward formal supervision.

The proposal, led by Dean of the College Michael Gordin, received endorsements from key student and faculty groups, including the Honor Committee, the Office of the Dean of Undergraduate Students, and the Undergraduate Student Government. While some students and faculty expressed concern that proctoring could erode trust, others supported it as a necessary adaptation. A November 2026 policy had already mandated proctoring for individual and small-group exams, setting the stage for this broader change. The Honor Committee Constitution remains unchanged, but university rules banning proctoring will be updated to reflect the new requirement.

Proponents argue that while proctoring won't eliminate cheating, it will serve as a deterrent and reduce pressure on students to monitor peers during exams. Former Honor Committee Chair Emerita Nadia Makuc '26 noted that the committee had long considered adding proctors as additional witnesses, especially given recent strains like increased case loads and AI-related challenges. William Aepli '26, former co-chair of the Peer Representatives, anticipated changes in the types of evidence presented in hearings. Faculty reactions were mixed, with Professor Jill Dolan calling the move "a shame, but necessary," acknowledging it marks a significant cultural shift for Princeton.

资深政治记者 Robert Gehrke 正在庆祝一项重要里程碑。《 The Salt Lake Tribune 》几年前曾设立付费墙，如今决定全面开放。从本周四起，sltrib.com 上的所有新闻将对全球任何读者完全免费，无论你身在 Utah 还是世界其他地方。他称此举为公众获取信息的"重大胜利"。

盖尔克回忆互联网早期，那时免费提供新闻曾被视为一场革命。他反思，这些最初的做法虽然可能是误判，却导致新闻逐渐被视为没有价值。《 The Salt Lake Tribune 》后来设立了付费墙——这是艰难但必要的商业决策，帮助报社存活下来，也确保犹他州民众仍有专门团队持续报道州议会和市政厅事务。盖尔克说，没有付费墙，这份报纸可能早已停刊。

付费墙并非永久安排。真正的转折出现在 2019 年，《 The Salt Lake Tribune 》成为全国首批转型为非营利机构的大型传统媒体之一。这一转变改变了它的使命：不再仅仅为财务存续而战，而是把提供必要报道作为一项公共服务来服务社区。目标始终是覆盖所有人，而不仅是那些能负担订阅的人，尽管维持运营的收入仍然重要。

那 2026 年发生了什么？盖尔克说，正是这些读者、订阅者和多年来支持报社的慷慨捐赠者，让免费访问成为可能。他们的资助使报社得以在 Logan 、 Moab 和 St. George 设立分社并扩展报道范围，如今也能彻底移除经济障碍。盖尔克认为，收费曾是对新闻价值的确认，而免费则是对这种价值的放大——把知识民主化，赋予每一位 Utahn 所需的信息。

他把这一步看作对当下媒体生态的直接回应：在社交媒体充斥争议和噱头十足却不可靠的信息的时代，拥有值得信赖的本地消息来源比以往都重要。他强调，正是职业且有职业伦理的新闻工作者在挖掘真相、监督权力。把这种核实与深度报道免费提供，是整个社区的胜利。

尽管为免费开放感到振奋，盖尔克也明确表示，新闻制作仍需成本。他呼吁现有订阅者和捐赠者继续支持组织。重要的本地新闻——包括法庭报道、追踪公共资金流向和调查性报道——都需要资源。他说，这一转变并不意味着新闻的价值降低；相反，作为一项免费的公共服务，它对 Utah 的未来更为重要。 Robert Gehrke, a longtime politics reporter at The Salt Lake Tribune, is celebrating a huge milestone. The newspaper, which went behind a paywall a few years ago, is opening its doors wide. Starting Thursday, all journalism on sltrib.com will be completely free for anyone to read, whether you're in Utah or anywhere else in the world. It's a decision he calls a "massive victory" for public access to information.

Gehrke looks back on the early days of the internet, when giving away news for free felt like a revolution. He reflects that those initial decisions, while perhaps a miscalculation, led to a situation where news started to seem valueless. The Tribune eventually implemented a paywall. It was a tough but necessary business move, one that helped keep the paper alive and ensured those in Utah still had a dedicated team covering the Capitol and city halls. Without it Gehrke says the paper might have folded entirely.

The paywall was never meant forever. The real turning point came in 2019 when The Tribune became the first major legacy publication in the country to transition to a nonprofit. That shift changed the mission. Instead of just surviving financially the focus became serving the community by providing essential reporting as a public service. The goal was always to reach everyone not just those who could afford a subscription although keeping the light still mattered.

So what changed in 2026? According to Gehrke these readers the subscribers and the generous donors who have supported the paper over the years are the reason free access is now possible. Their financial backing has allowed The Tribune to expand its reach with bureaus in Logan, Moab, and St. George and now to remove the financial barrier entirely. He argues that charging for the news was about recognizing value, and giving it away for free is about amplifying that value. It is about democratizing knowledge and empowering every Utahn with the information they need.

Gehrke sees this move as a direct response to the current media landscape. In an era where social media is filled with controversy and "dubious sizzle" having a trusted local source is more critical than ever. He points out that professional ethical journalists are the ones digging for the truth and holding power to account. Making that verification and in-depth reporting available at no cost is a win for the entire community.

Despite the excitement over free access Gehrke is clear that the work still costs money. He urges current subscribers and donors to stay with the organization. Critical local journalism including court coverage, tax dollar tracking, and investigative pieces requires resources. This transformation doesn't mean the journalism is worth less he says. If anything providing it as a free public service makes it even more essential to the future of Utah.

Cigna 公司旗下的 EviCore by Evernorth 是这一相对隐蔽行业中的最大参与者，为超过 1 亿美国人的医疗承保决策提供外包服务，约占三分之一的参保人群。调查发现，该公司使用一种内部称为"拨盘"的人工智能算法来调整索赔被拒的概率，且一些合同通过经济激励鼓励提高拒绝率以减少医疗支出。根据内部文件和对多名前雇员及行业专家的访谈，EviCore 向保险公司承诺投资回报率为 3 比 1——即保险公司每付给该公司 1 美元，就能在医疗费用上节省 3 美元。

调查显示，EviCore 利用该算法来批准或拒绝预先授权请求，算法会根据批准的概率给请求打分。前雇员表示，公司可以调整将案件交由人工复核的门槛，从而随意提高或降低拒绝率。更多的复核往往导致更多的拒绝，因为只有医生在参考公司医疗指南后才能做出最终拒绝。这些指南多次受到医疗团体的批评，包括美国心脏病学会和美国放射肿瘤学会，批评点包括过时、僵化，导致不当拒绝或延误必要治疗，因此一些医生戏称该公司为"EvilCore"。

来自俄亥俄州的 61 岁焊工 Little John Cupp 的案例反映了这种做法的人为后果。当他出现心脏症状时，医生要求进行心脏导管检查，但 EviCore 两次以"不具医疗必要性"为由拒绝。医院改为为他做了费用更低的核成像运动负荷试验，结果显示心功能恶化。他在接受该检查后仅 36 小时便因心脏骤停去世。虽然部分医学专家对是否需要导管检查存在分歧，但也有人认为该手术可能挽救了他的生命，并称最初的拒绝是不负责任的。 Cupp 的女儿 Chris 曾对多方提起诉讼，但因针对雇主提供的健康计划的诉讼必须在联邦法院审理，而相关判例通常偏向保险公司，她不得不放弃对 UnitedHealthcare 和 EviCore 的追责。

调查还发现，EviCore 不仅通过直接拒绝来节约成本，还通过高管所称的"哨兵效应"实现节省：一旦医生知道有公司在监控，他们就可能干脆不再提出某些检查或治疗的申请。以佛蒙特州为例，心脏科的预先授权请求在一年多的时间里下降了 38% 。尽管 EviCore 及其母公司 Cigna 声称公司依循循证医学以提升质量和安全、减少不必要的支出，但像美国医学会前主席 Barbara McAneny 这样的医疗专业人士认为，这一体系把利润置于病人护理之上。 EviCore 也并非个例，竞争对手如 Elevance Health 旗下的 Carelon Medical Benefits Management 也有类似做法，并涉及已和解的"不当拒绝"诉讼。

监管机构很少对像 EviCore 这样的公司处以重大惩罚。康涅狄格州保险部门在审查 196 份档案后发现 77 项违规，仅对 EviCore 处以 1.6 万美元的罚款。该公司还获得了一些行业协会的认证，这些协会审查其是否符合行业标准。 Chris Cupp 现在靠开校车维持生计，她对保险公司在生死攸关的治疗决策中拥有过多话语权感到愤怒，尤其是对像心脏病这样可能致命的疾病。她希望通过讲述自己的遭遇，能阻止其他人因保险拒绝而遭遇同样可被预防的死亡。 EviCore by Evernorth, owned by Cigna, is the largest player in a largely hidden industry that outsources medical coverage decisions for more than 100 million Americans, about 1 in 3 insured people. A ProPublica and Capitol Forum investigation found the company uses an artificial intelligence algorithm known internally as "the dial" to adjust the likelihood of claim denials, with some contracts financially incentivizing higher denial rates to reduce health spending. The investigation, based on internal documents and dozens of interviews with former employees and industry experts, revealed that EviCore promises insurers a 3-to-1 return on investment, meaning for every dollar spent on the company, insurers pay three dollars less on medical care and costs.

The investigation showed that EviCore approves or denies prior authorization requests using this algorithm, which scores requests based on the probability of approval. Former employees explained that the company can adjust the threshold for sending cases for human review, thereby increasing or decreasing denial rates at will. More reviews lead to more denials since only doctors can issue final denials after consulting company medical guidelines. These guidelines have been repeatedly criticized by medical groups, including the American College of Cardiology and the American Society for Radiation Oncology, as being outdated, rigid, and resulting in inappropriate denials or delays in necessary care, leading some doctors to refer to the company as "EvilCore."

The story of Little John Cupp, a 61-year-old welder from Ohio, illustrates the human impact. After developing cardiac symptoms, his doctor requested a heart catheterization, but EviCore denied it twice as "not medically necessary." Cupp was instead given a less expensive nuclear stress test, which revealed worsening heart function. He died of cardiac arrest just 36 hours after that test. While some medical experts disagreed on whether the catheterization was necessary, others argued it could have saved his life, with one calling the original denial irresponsible. Cupp's daughter, Chris, sued multiple parties but was forced to drop UnitedHealthcare and EviCore because lawsuits against employer-filed health plans must be tried in federal court where case law favors insurers.

The investigation found that EviCore achieved cost savings not only through denials but also through what executives call the "sentinel effect," where doctors stop requesting procedures altogether once they know the company is watching. In Vermont, cardiology prior authorization requests dropped 38% in a little over a year. While EviCore and its parent company Cigna stated that the company uses evidence-based medicine to improve quality and safety while reducing unnecessary costs, medical professionals like former AMA president Barbara McAneny argued the system prioritizes profit over patient care. The company is not alone, as competitors like Carelon Medical Benefits Management, a subsidiary of Elevance Health, also engage in similar practices, including settled lawsuits alleging improper denials.

Regulators rarely impose significant penalties on companies like EviCore. Connecticut's insurance department recently fined EviCore just $16,000 for over 77 violations found in a review of 196 files. The company is also accredited by trade associations that review compliance with industry standards. Chris Cupp, who now drives a school bus to make ends meet, expressed frustration that insurers have too much say over life-saving decisions, especially regarding conditions like heart disease that can be fatal. She hopes sharing her story will prevent others from experiencing similar preventable deaths due to insurance denials.

Apple 在 2026 年 3 月推出了 MacBook Neo，定价 599 美元，成为有史以来最便宜的 Mac 笔记本。 Apple 没有采用 M-series 芯片，而是搭载了 A18 Pro——与 iPhone 16 Pro 相同的处理器，使用 TSMC 第二代 3nm 工艺（N3E）制造。该芯片包含六个 CPU 核心（两个性能核，4.04 GHz；四个能效核，2.42 GHz）、一个支持硬件光线追踪的五核 GPU 以及 16 核 Neural Engine 。尽管有人质疑"在 Mac 上用手机芯片"的做法，但基准测试结果并不支持这种否定：A18 Pro 在 Geekbench 6 上单核约 3569 分、多核约 8879 分，单核性能处于 M3 与 M4 之间，在该价位上比同档次的 Intel 和 Qualcomm 机型快约 38%–43% 。

对这款采用铝制无风扇机身的笔记本做实际热测试可以看出，Neo 更像短跑型选手而非耐久型选手。冷启动时 A18 Pro 能爆发出强劲的短时性能，单核得分约 3569 分；但在持续负载下热表现急转直下。一次五分钟的全核压力测试显示：在前 60 秒满载后，仅 15 秒内 CPU 利用率就从 570% 跌至 207%，下降约 64% 。热饱和后单核得分降至 476 分，比冷启动时下降 87% 。芯片触及 105°C 的温控上限，频率从 3.3 GHz 降到约 2.3 GHz，但机身外表温度仅为 97.6°F（36.4°C），表明 Apple 把舒适度放在了持续性能之前。

在日常使用场景下，如网页浏览、办公、媒体观看和轻量开发，Neo 依然流畅且够用。那些能在 60 秒内完成的任务（例如小项目编译或照片处理）能从桌面级单核短时爆发中受益。只有在长时间的视频转码或大型构建等持续重负载下，Neo 的局限才会明显浮现。其 GPU 在 Metal 下得分为 31,286 分，由于着色器核心较少，实际上略逊于 M1 Air，GPU 密集型任务并非 Neo 的目标用途。

架构上，A18 Pro 与 M4 有相同血统：都基于 ARMv9.2-A 、相同的定制 Everest 与 Sawtooth 核心，并采用相同的 TSMC N3E 工艺。按频率归一化后，两款芯片每 GHz 大约能提供 857 分的 Geekbench 成绩，意味着 IPC 基本一致。差别主要在系统层面：A18 Pro 的内存带宽仅为 M4 的一半（60 GB/s 对比 120 GB/s），CPU 和 GPU 核心更少，热设计也更保守。更大的 24 MB 系统级缓存在一定程度上弥补了带宽不足，使得把它称为"M4 的小号版本"来描述日常响应性并不夸张。

599 美元的定价从晶圆经济学角度看也讲得通。 A18 Pro 的裸片面积仅 105 mm²，比 M4 小约 25%，在 300 mm 晶圆上可产出约 586 片裸片，良率处于 85%–90% 。按 Apple 估算的每片晶圆成本 18,000–20,000 美元计算，每颗裸片在封装前的成本约为 34–40 美元。与此相比，M4 或 M4 Max 的单芯片成本要高得多。 Apple 每年出货约 2.3 亿部 iPhone，因此 A18 Pro 的掩膜费和研发成本早已在发布 Neo 之前分摊到数亿台设备上。 Neo 还可能使用那些在 iPhone 生产中第 6 个 GPU 核心失效后降级的裸片，以进一步提升良率。整机物料清单大约为 200–290 美元，使得 Apple 能保持约 50%–58% 的毛利率。

8 GB 内存上限是 Neo 的明显短板，但这既是工程设计的结果，也与战略时机有关。 A18 Pro 的内存控制器本就是为 iPhone 16 Pro 的 8 GB 配置设计，若要升级需要改变封装设计。更关键的是，2026 年的结构性 DRAM 短缺让这一决定在战略上更合理。 Samsung 、 SK Hynix 和 Micron 已将多达 40% 的先进晶圆产能转向用于 AI 加速器的高带宽内存（HBM），而 HBM 每 GB 消耗的晶圆面积约为标准 DDR5 的三倍。 Micron 在 2025 年 12 月完全退出了消费级内存市场。 DDR5 32 GB 套装从 2025 年第三季度的 120 美元涨到 2026 年第一季度的 350 美元，内存占 PC BOM 的比例也从 16% 上升到 23% 。 Gartner 预测到 2028 年笔记本平均售价将上涨 17%，500 美元以下的入门级 PC 市场将消失。

在这样的价格环境下，随着 Windows 笔记本价格上涨 15%–20%，Apple 的 599 美元 Neo 竞争力不断增强。 2025 年中期那些出货时配备 16 GB 内存的竞争机型现在售价已在 700–750 美元之间，而 Apple 通过将内存减半，将受短缺影响的风险降了一半。 Apple 还能通过 iCloud+ 和 Apple One 从 Neo 用户处获得增值服务收入，在两年设备生命周期内带来约 240–480 美元的服务营收，使得硬件利润在生态转化中的重要性有所下降。 Neo 在前所未有的价格点上是一次令人印象深刻的工程实现，时机把握也很精准，正好利用住了市场危机带来的窗口。

A18 Pro 并非妥协品：在单线程工作中它达到 M3 至 M4 级别，由成熟的 iPhone 芯片大规模量产打造，且无需额外研发支出。 8 GB 内存上限会随着时间显得捉襟见肘且不可升级，因此第二代 Neo 推出 12 GB 或 16 GB 版本几乎是必然。但就当前形势看，随着 500 美元以下 PC 的消失和笔记本平均价格上涨 17%，一款售价 599 美元、具备 M3 级单核性能、铝制无风扇机身与约 16 小时续航的 MacBook，对于 Apple 来说极具战略意义——不是因为它是最好的 Mac，而是因为它在行业危机中抢占了入门级市场。 Apple unveiled the MacBook Neo in March 2026 at $599, making it the most affordable Mac laptop ever. Instead of using an M-series chip, Apple equipped it with the A18 Pro, the same processor found in the iPhone 16 Pro, fabricated on TSMC's second-generation 3nm process (N3E). The chip features six CPU cores (two performance at 4.04 GHz and four efficiency at 2.42 GHz), a five-core GPU with hardware ray tracing, and a 16-core Neural Engine. While critics were quick to dismiss the "phone chip in a Mac" approach, the benchmarks tell a different story. The A18 Pro posts Geekbench 6 scores of around 3,569 single-core and 8,879 multi-core, placing its single-core performance between the M3 and M4, and beating Intel and Qualcomm competitors at this price tier by 38-43%.

Real-world thermal testing on the fanless aluminum chassis reveals the Neo's nature as a sprinter rather than a marathon runner. In cold-start conditions, the A18 Pro delivers impressive burst performance, with single-core scores around 3,569. However, after sustained load, the thermal story changes dramatically. A five-minute all-core stress test shows CPU utilization crashing 64% in just 15 seconds after the first 60 seconds of full boost, dropping from 570% utilization down to 207%. Post-thermal soak single-core scores plummet to 476, an 87% reduction from the cold-start baseline. The chip hits its 105°C thermal limit and drops from 3.3 GHz to approximately 2.3 GHz, yet the external chassis barely exceeds body temperature at 97.6°F (36.4°C), showing Apple prioritized comfort over sustained power.

For everyday tasks like web browsing, office work, media consumption, and light development, the Neo feels responsive and capable. Tasks completing within 60 seconds, such as compiling small projects or processing photos, benefit from the desktop-class single-core burst performance. The laptop's limitations become apparent only during sustained heavy workloads like long video encoding or large builds. GPU performance at 31,286 (Metal) actually trails the M1 Air slightly due to having fewer shader cores, making GPU-intensive work clearly outside the Neo's target use case.

Architecturally, the A18 Pro shares DNA with the M4, both using ARMv9.2-A, the same custom Everest and Sawtooth cores, and the same TSMC N3E process. When normalized by clock speed, both chips deliver approximately 857 Geekbench points per GHz, meaning the IPC is essentially identical. The key differences are system-level: the A18 Pro has half the memory bandwidth of the M4 (60 GB/s versus 120 GB/s), fewer CPU and GPU cores, and a much lower thermal envelope. The larger 24 MB System Level Cache partially compensates for the bandwidth gap, making "Baby M4" a reasonable shorthand for everyday responsiveness.

The $599 price point becomes understandable through wafer economics. The A18 Pro's die measures just 105 mm², 25% smaller than the M4, yielding approximately 586 gross dies per 300mm wafer with 85-90% maturity yields. At Apple's estimated wafer cost of $18,000-$20,000, each die costs roughly $34-40 before packaging. Compare this to the M4 or M4 Max, which cost significantly more per chip. Apple ships 230 million iPhones annually, meaning all mask costs and R&D for the A18 Pro were amortized across hundreds of millions of units before a single Neo shipped. The Neo may also use binned dies that failed their sixth GPU core during iPhone production, further improving yields. The full bill of materials runs about $200-290, giving Apple healthy gross margins around 50-58%.

The 8GB RAM ceiling represents the Neo's defining limitation, but it reflects both engineering constraints and strategic timing. The A18 Pro's memory controller was designed for the iPhone 16 Pro's 8GB configuration, and upgrading would require different packaging. More importantly, a structural DRAM shortage in 2026 has made this decision strategically sound. Samsung, SK Hynix, and Micron have reallocated up to 40% of advanced wafer output to High Bandwidth Memory for AI accelerators, with HBM consuming three times the wafer area per gigabyte compared to standard DDR5. Micron exited the consumer memory market entirely in December 2025. DDR5 32GB kits that cost $120 in Q3 2025 hit $350 by Q1 2026, and memory's share of PC bill of materials rose from 16% to 23%. Gartner projects a 17% rise in average laptop prices and the disappearance of the sub-$500 entry-level PC segment by 2028.

This pricing environment makes Apple's $599 Neo increasingly competitive as Windows laptop prices climb 15-20%. While competitors shipping with 16GB in mid-2025 now cost $700-750, Apple's choice to halve the RAM halves its exposure to the shortage. The company can also monetize Neo buyers through iCloud+ and Apple One subscriptions generating $240-480 in services revenue over a two-year device lifecycle, making hardware margin less critical for ecosystem conversion. The Neo is a genuinely impressive engineering achievement at an unprecedented Apple price point, strategically timed to exploit a market in crisis. The A18 Pro is not a compromise chip but M3-to-M4 class in single-threaded work, built from mature iPhone silicon at massive scale with zero incremental R&D cost. The 8GB memory ceiling will age poorly and is not upgradeable, making the second-generation Neo with 12GB or 16GB the obvious follow-up. But right now, with sub-$500 PCs disappearing and average laptop prices climbing 17%, a $599 MacBook with M3-class single-core performance, an aluminum fanless chassis, and 16 hours of battery life represents the most strategically significant product Apple has shipped in years, not because it is the best Mac, but because it captures the entry-level market during an industry crisis.

Haiku 是一个受 BeOS 启发的开源操作系统，面向个人计算，强调速度、简洁和强大功能。项目拥有活跃的开发社区，定期发布更新，核心开发者和参与 Google Summer of Code (GSoC) 的贡献者持续推动进展。 2026 年，Haiku 再次入选 GSoC，选定三名学生开展旨在提升系统功能与现代化组件的项目。

本次 GSoC 2026 入选项目聚焦 Haiku 的关键改进方向。 Leo Rouleau 将扩展 Devices 应用，把它打造为一款完善的硬件管理器，让用户能在图形界面中查看详细规格并直接执行管理操作。 Mohammed R. Attia 将致力于现代化 Haiku 的蓝牙协议栈，完成 HCI（Host Controller Interface）改进并实现 HID（Human Interface Device）配置文件支持，从而提升对键盘、鼠标等输入设备的兼容性。 Vighnesh Sawant 则将在蓝牙协议栈中实现 HFP（Hands‑Free Profile）支持，以实现音频传输和免提通话，并为未来功能扩展奠定基础。

这些工作延续了 Haiku 在渐进式改进和社区参与方面的传统。项目定期发布活动与合同报告，跟踪内核开发、驱动支持和用户界面优化等各子系统的进展。近期报告强调了对 ARM64 架构支持的工作，表明 Haiku 正向 x86 以外的平台扩展。与此同时，团队通过 "Gerrit code review iceberg" 博客系列等举措，持续清理长期积压的代码审查，旨在减少待处理变更并促进开发者协作。

Haiku 的经费来自捐赠和周边产品销售，筹款活动有助于维持开发工作。项目还提供每晚构建版本供测试，让用户体验最新功能与修复（但这些版本可能不稳定）。在经验丰富的导师团队引导下，新贡献者得以参与，项目也有明确的技术路线图。作为一个轻量、高效且用户友好的操作系统，Haiku 在继承 BeOS 传统的同时，持续适配当代计算的需求。 Haiku is an open-source operating system inspired by BeOS, designed specifically for personal computing with a focus on speed, simplicity, and power. The project continues to maintain an active development community, with regular updates and contributions from both core developers and participants in programs like Google Summer of Code (GSoC). In 2026, Haiku was accepted into GSoC once again, selecting three students to work on projects aimed at improving the system's functionality and modernizing its components.

The selected GSoC 2026 projects highlight key areas of development for Haiku. Leo Rouleau will expand the capabilities of the Devices application, transforming it into a comprehensive hardware manager that allows users to view detailed specifications and perform administrative tasks directly from the GUI. Mohammed R. Attia will focus on modernizing Haiku's Bluetooth stack by completing HCI (Host Controller Interface) improvements and implementing support for the HID (Human Interface Device) profile, enabling better compatibility with input devices like keyboards and mice. Vighnesh Sawant will implement support for the HFP (Hands-Free Profile) in the Bluetooth stack, allowing audio streaming and hands-free voice calls, while also laying groundwork for future enhancements.

These efforts build on Haiku's ongoing commitment to incremental improvement and community engagement. The project regularly publishes activity and contract reports that track progress across various subsystems, including kernel development, driver support, and user interface refinements. Recent reports have highlighted work on ARM64 architecture support, indicating Haiku's expansion beyond traditional x86 platforms. Additionally, the team continues to address long-standing code reviews through initiatives like the "Gerrit code review iceberg" blog series, which aims to reduce the backlog of pending changes and encourage collaboration among developers.

Haiku remains financially supported through donations and merchandise sales, with fundraising campaigns helping sustain development efforts. The project also provides nightly builds for testing purposes, allowing users to experiment with the latest features and bug fixes, though these versions may be unstable. With a dedicated mentor team guiding new contributors and a clear roadmap for technical advancement, Haiku continues to evolve as a lightweight, efficient, and user-friendly operating system rooted in the legacy of BeOS but adapted for contemporary computing needs.

S-100 虚拟工作台是一个基于网页的模拟器，用于重现 20 世纪 70 、 80 年代的 S-100 总线计算机系统。用户可以通过拖放卡片库向虚拟 S-100 机箱装配硬件，配置并运行诸如 Altair 8800 、 IMSAI 8080 和 Processor Technology SOL-20 等经典系统。平台提供多种预设配置（例如运行 CP/M 2.2 并配置 64K RAM 的 Altair 8800），也可在可用插槽中添加 CPU 、内存、 I/O 和存储卡来自行组建自定义系统。

模拟器支持包括 Intel 8080 和 Zilog Z80 在内的多种 CPU，提供精确的指令集仿真和可调的时钟速率。内存选项包括可配置地址范围的静态 RAM 卡和可加载内置或自定义二进制映像的 ROM 卡，并通过虚拟 I/O 端口支持影子 ROM 行为。串行 I/O 由可配置的 UART 卡处理（如 MITS 88-2SIO 和通用 SIO 设备），为运行 CP/M 等操作系统提供控制台输入输出。

存储仿真涵盖硬扇区和软扇区软盘控制器。 MITS 88-DCDD 控制器支持早期 Altair 使用的 IBM 3740 原始格式，而基于 WD1793 的 FDC 则兼容 IMSAI 、 Cromemco 及其他采用标准平面磁盘映像的非 MITS 平台。驱动器 A 到 D 可分别挂载磁盘映像文件，用户可从真实的 CP/M 磁盘映像启动并运行遗留软件。

图形与显示选项包括 Cromemco Dazzler 卡（从系统 RAM 中软件定义的帧缓冲生成复合视频输出，支持多种彩色与单色分辨率）和 Processor Technology VDM-1（提供内存映射文本显示，16 行 ×64 列 ASCII 字符、硬件滚屏和反显支持）。这些扩展使依赖图形或增强文本输出的软件能够进行直观交互。

界面还包含用于输入输出的终端模拟器、显示实时地址、数据信号与控制信号活动的总线分析仪，以及用于调试的总线跟踪日志。用户可以将 Intel HEX 文件直接加载到内存，使用运行、单步、重置和重启等命令控制执行并检查内存内容。系统精确仿真底层总线行为，包括 MEMR 、 MEMW 、 IOR 、 IOW 和 HOLD/HLDA 仲裁，为历史软件保存与教学提供了忠实的 S-100 总线环境再现。 This is a description of the S-100 Virtual Workbench, a web-based emulator for vintage S-100 bus computer systems from the 1970s and 1980s. The platform allows users to configure and run classic hardware setups, including the Altair 8800, IMSAI 8080, and Processor Technology SOL-20, using a drag-and-drop card library to populate a virtual S-100 chassis. Users can select from a range of preset configurations, such as an Altair 8800 running CP/M 2.2 with 64K RAM, or build custom systems by adding CPU, memory, I/O, and storage cards to available slots.

The emulator supports multiple CPU types, including the Intel 8080 and Zilog Z80, with accurate instruction set emulation and configurable clock rates. Memory options include static RAM cards with configurable address ranges and ROM cards that can load built-in or custom binary images, with support for shadow ROM behavior via phantom I/O ports. Serial I/O is handled through configurable UART cards like the MITS 88-2SIO and generic SIO devices, enabling console input and output for running operating systems like CP/M.

Storage emulation includes both hard-sector and soft-sector floppy disk controllers. The MITS 88-DCDD controller supports the original IBM 3740 format used by early Altair systems, while the WD1793-based FDC provides compatibility with IMSAI, Cromemco, and other non-MITS platforms using standard flat disk images. Disk drives A through D can be assigned disk image files, allowing users to boot and run legacy software from authentic CP/M disk images.

Graphics and display options include the Cromemco Dazzler card, which generates composite video output from a software-defined frame buffer in system RAM, supporting multiple color and monochrome resolutions. The Processor Technology VDM-1 provides memory-mapped text display with 16 rows by 64 columns of ASCII characters, hardware scrolling, and inverse video support. These cards enable visual interaction with software that relies on graphical or enhanced text output.

The interface includes a terminal emulator for user input and output, a bus analyzer showing real-time address, data, and control signal activity, and a bus trace log for debugging. Users can load Intel HEX files directly into memory, control execution with run, step, reset, and reboot commands, and inspect memory contents. The system accurately emulates low-level bus behavior, including MEMR, MEMW, IOR, IOW, and HOLD/HLDA arbitration, providing a faithful recreation of the S-100 bus environment for historical software preservation and education.

Kickstarter 更新了内容指南，禁止多类 NSFW 内容——包括"暗示性行为"、"MILF/DILF"、"暗示裸露"，以及对"女性乳头 / 乳晕、生殖器"和"肛门"的描绘，甚至连"臀部"也被列为禁项。 Kickstarter 和其支付处理商 Stripe 暂未就此置评。 Stripe 部分由 Palantir 董事长 Peter Thiel 和 X 所有者 Elon Musk 持股，这一变化似乎受到其影响。

据艺术家兼作家 Mike Wolfer 所述，Kickstarter 最早在 2026 年 3 月就开始向创作者发送邮件，告知 Stripe 将对任何包含成人或 NSFW 内容的项目进行独立审查。报道还称，Stripe 有权在众筹项目进行中乃至项目成功融资后将其下架。这一突变尤其引人注意：Kickstarter 在 2025 年 9 月刚推出旨在展示成人向优质项目的通讯"Kickstarter After Dark"。

如果这些变化确系由 Stripe 推动，那并非支付处理商首次左右数字平台的内容政策。 2025 年，在 Visa 和 Mastercard 等银行合作伙伴与支付处理商的压力下，Steam 和 Itch.io 被迫下架大量 NSFW 游戏。澳大利亚反色情组织 Collective Shout 宣称对此次审查负责，但更大背景是金融公司在收紧对线上买卖与分享内容的监管范围。

Kickstarter 此次的限制较其此前仅禁止"色情内容"的规定发生了显著变化。新规在禁止条目上更为具体、范围更广，远超传统意义上的色情内容，这使依赖平台为成人向项目筹资的创作者，尤其是漫画和游戏领域的从业者深感担忧。

这一事态凸显了支付处理商对在线内容的日益影响力。作为金融中介，Stripe 、 Visa 和 Mastercard 等公司通过威胁切断支付通道，实际上能决定数字平台上什么可以买卖、什么不可以。这迫使平台要么接受金融公司的要求，要么冒着无法运营的风险。

对创作者而言，即便平台愿意托管成人内容，负责处理交易的支付方可能仍不同意。这种局面对创作表达产生了寒蝉效应：创作者不仅要遵守所用平台的内容政策，还要应对那些往往不透明且不断变化的金融标准。结果是，成人内容越来越被边缘化，原因更多是几家强势金融中介的决策，而非法律或公众需求。 Kickstarter has updated its content guidelines to prohibit various forms of NSFW content, including "implied sex acts," "MILF/DILF" content, "implied nudity," and depictions of "female nipples/areolas, genitalia," and "anuses." The platform has even banned content featuring "buttocks." This shift appears to be driven by pressure from its payment processor, Stripe, which is partially owned by Palantir Chairman Peter Thiel and X proprietor Elon Musk. Kickstarter and Stripe did not immediately respond to requests for comment.

According to artist and writer Mike Wolfer, Kickstarter began emailing creators as early as March 2026 to inform them that Stripe would conduct its own review of any projects featuring adult or NSFW content. Stripe reportedly reserves the right to shut down crowdsourced projects while they are live or even after they have been successfully funded. This sudden shift is particularly notable given that Kickstarter launched its "Kickstarter After Dark" newsletter in September 2025, which was designed to showcase the best adult-oriented projects on the platform.

If Stripe is indeed behind these changes, it would not be the first time payment processors have influenced content policies on digital platforms. In 2025, both Steam and Itch.io were forced to remove numerous NSFW games from their platforms following pressure from banking partners and payment processors like Visa and Mastercard. The Australian anti-porn group Collective Shout claimed credit for the censorship, but the move appears to be part of a broader trend among financial companies to crack down on what can be bought, sold, and shared online.

The restrictions on Kickstarter represent a significant change from the platform's previous guidelines, which simply prohibited "pornographic content." The new rules are far more specific and expansive in what they ban, covering a wide range of content that goes well beyond traditional pornography. This has raised concerns among creators who rely on the platform to fund adult-oriented projects, particularly in the comics and gaming communities.

The situation highlights the growing influence that payment processors have over online content. As financial intermediaries, companies like Stripe, Visa, and Mastercard have the power to effectively dictate what can and cannot be sold on digital platforms by threatening to cut off their ability to process payments. This creates a situation where platforms must either comply with the demands of these financial companies or risk losing their ability to operate.

For creators, this means that even if a platform like Kickstarter is willing to host adult content, the payment processors that handle the transactions may not be. This creates a chilling effect on creative expression, as creators must navigate not only the content policies of the platforms they use but also the often opaque and shifting standards of the financial companies that power them. The result is a landscape where adult content is increasingly marginalized, not necessarily because of legal requirements or public demand, but because of the decisions made by a handful of powerful financial intermediaries.

这份宣言主张：在公司工作的开源软件维护者，不应再为修复和维护其雇主已经依赖的开源代码而事事请示许可。核心观点是，维护开源基础设施本身就是一项正当的工作，而不是只能挤到个人时间去做的爱好。公司每天都在从开源中获益，维护者在工作时间做这类关键性基础设施工作是合理的，不应被要求走繁琐的审批或寻求上级"赐予"的许可。作者把这种做法称为"默默抵抗"——在上班时间专业地完成必要工作，把它当作任何技术债务或基础设施任务一样对待。

宣言提出三条原则。第一，去做该做的事：审查 pull request 、更新依赖、在你的工作范围内发布修复。第二，保护自己：核实雇佣合同、将公司机密隔离出来，并确保你对所贡献的开源代码拥有相应的知识产权或许可。第三，保持平衡：这不是要把全部工作时间都投入开源或以此冒险丢掉饭碗，而是在日常工作中可持续地把必要的维护工作纳入其中。

作者承认存在其它路径，例如 Open Source Pledge（公司向维护者付费）或 Open Source Friday（公司拨出时间）。这些都是积极的进展，但本宣言主张在管理层拒绝把开源维护正式视为工作时，个人可以采取更直接的行动。重点放在个人能控制的事，比如如何安排自己的工作时间，而不是被动等待公司的预算或政策变化。

针对常见反对意见，作者反驳这是"偷窃"的说法：公司几十年来一直在享受免费的开源红利，维护关键基础设施并不是在掠夺股东利益。要求维护者每次都要请示的做法只是维持一种权力不平衡，把人当成必须额外牺牲私人时间的资源。这与"安静辞职"不同：这里做出的工作是必要的基础设施维护，只是被公司拒绝承认为正式工作。优秀的雇主已经允许这种做法，本宣言针对的是那些不允许的公司。

作者 Mike McQuaid 以他作为 Homebrew 长期维护者和 GitHub 员工的经历为例说明实践可行性。他在每份工作中都谈妥了知识产权协议以合法化开源贡献，自从有了孩子之后，他超过 90% 的开源贡献都是在工作时间完成的。他强调，没有人应被要求为企业依赖的工作无偿贡献私人时间，维护者需要让自己的工作方式可持续，而不是等待他人来解决这个问题。

宣言同时声明这不是法律建议，个人在具体问题上应咨询专业律师。雇佣合同、公司政策和发明归属条款可能会主张对雇佣期间完成工作的权利，因此维护者应在开始前认真阅读合同并争取书面的开源例外或知识产权约定。作者建议以书面形式要求开源豁免，并以 GitHub 的 Balanced Employee IP Agreement 作为参考范本。保密与安全也很关键：维护者不得泄露公司机密、不得绕过安全控制。

作者也承认这种做法的局限性：在按客户计费的时间、初级工程师缺乏谈判筹码或受监管行业中，这种做法风险更大；相反，对于为雇主修复其正在使用的公开依赖的资深维护者，这种做法最为可行。最理想的情形是把开源维护作为工程工作的一部分——维护你已经参与的项目、改进与你工作相关的共享工具。

总体目标不是"从工作中偷走时间"，而是平衡企业从开源中获得的利益与你能回馈的贡献。未向公司披露并不意味着鲁莽行事；在政策或合同发生变化时，维护者应当凭判断评估风险。与其为一个随时可能用新人替代你的公司耗尽精力，不如通过可持续的方式把开源维护纳入职业生涯中，这对个人和社区都更健康。 This manifesto argues that open source software maintainers who work at companies should stop asking for permission to fix and maintain the open source code their employers already depend on. The core idea is that maintaining open source infrastructure is legitimate work, not a hobby to be done only on personal time. Companies extract value from open source every day, and it is reasonable for maintainers to use work hours for this essential infrastructure work without needing formal approval, paperwork, or a manager's blessing. The author frames this as "quiet resistance" — doing the necessary work professionally and on company time, treating it like any other technical debt or infrastructure task.

The manifesto outlines three main principles. First, do the work: review pull requests, update dependencies, and ship fixes where your work already involves open source. Second, protect yourself by verifying your employment contract, keeping confidential information separate, and ensuring you own the open source IP you contribute. Third, maintain balance: this is not about spending 100% of work hours on open source or getting fired, but about integrating necessary maintenance into your regular work life sustainably.

The author acknowledges alternatives like the Open Source Pledge, where companies pay maintainers, and Open Source Friday, where companies donate time. These are seen as positive steps, but the manifesto positions itself as the next evolution: taking direct action when management refuses to formally recognize open source maintenance as work. The focus is on what individuals can control, like how they spend their working time, rather than waiting for company budget decisions.

Addressing objections, the author counters that this is not theft, as companies have benefited from free open source subsidies for decades. Maintaining critical infrastructure is not stealing from shareholders. The argument that maintainers should ask permission is dismissed as preserving a power imbalance, comparing it to not needing permission for basic necessities. This is distinguished from quiet quitting, as it produces essential infrastructure work that companies refuse to classify as legitimate. Good employers already allow this practice, and the manifesto targets those that do not.

The author, Mike McQuaid, shares his personal experience as a long-time Homebrew maintainer and GitHub employee. He has negotiated IP agreements at every employer to legitimize his open source work and now does over 90% of his open source contributions during work hours since having children. He emphasizes that nobody is entitled to unpaid personal time for work that businesses depend on, and maintainers must make their work sustainable without waiting for others to solve the problem.

The manifesto includes important caveats: it is not legal advice, and individuals should consult qualified professionals for their specific situations. Employment contracts, policies, and invention assignment clauses can claim work created during employment, so maintainers should read their contracts and negotiate IP agreements before starting. The author recommends asking for open source carve-outs in writing and points to GitHub's Balanced Employee IP Agreement as a model. Confidentiality and security are critical: maintainers must not disclose private company information or bypass security controls.

The author acknowledges limitations: this approach is weakest for client-billed time, junior engineers without leverage, or regulated environments. It is strongest for senior maintainers fixing public dependencies their employers use. The cleanest version is treating open source maintenance as part of engineering work, maintaining projects you already work on, and improving shared tools your job touches. The goal is not to "steal" from work but to balance what work takes from open source with what you can give back. Undisclosed does not mean reckless, and maintainers should use their judgment when policies or contracts change the risk. A sustainable performance review is healthier than burning out for a company that could replace you tomorrow.