标签：#c

A birth-chart form looks simple: ask for a date, time, and place, then calculate. The interface may be simple. The input is not. 1992-11-01 01:30 does not identify one universal instant. It is a wall-clock reading that only becomes meaningful after you resolve the place, the historical time-zone rule, and any daylight-saving transition. If the time is unknown, inventing a convenient default can create chart features that were never supported by the user’s data. I ran into these problems while building AstroZen , a Next.js application that calculates a BaZi Four Pillars chart and a Western natal chart before generating an optional interpretation. The most important architectural decision was this: Calculation is an evidence pipeline. Interpretation is a separate layer. This post explains the calculation pipeline, the failure modes I had to remove, and why “unknown” must remain unknown. 1. A city name is not a coordinate Early prototypes often use a short city list or a default coordinate. That works for a layout demo, but it is not acceptable once location affects the result. Names are ambiguous: Paris can mean France or Texas. Springfield needs a state or region. Country abbreviations come in several forms. A valid city must resolve to both coordinates and a time-zone identifier. AstroZen sends the submitted city to Open-Meteo’s geocoding service, then scores the returned candidates against the requested country and optional region hint. It keeps the following structured result: type ResolvedPlace = { name : string ; region : string | null ; country : string ; countryCode : string ; latitude : number ; longitude : number ; timeZone : string ; // IANA, for example "Europe/Madrid" }; Candidate selection considers exact city-name matches, country matches, an optional region hint, and population as a small tie-breaker. Population never replaces the country check. The more important rule is what happens when resolution fails: if ( ! selected || ! countryMatches ( selecte

Every team shipping an agent has the same meeting. Someone asks "is it ready?" and the room splits. One person saw a great demo. Another watched it invent a refund policy an hour ago. The argument runs in circles because nobody agreed what "ready" means, so the loudest opinion wins and the agent ships on a vibe. Making an AI agent production-ready is not a moment of confidence. It is a bar you write down before you build, then measure against. This post is about that bar: why agents need a different one than the services you already ship, and how to define it so "is it ready?" becomes a number instead of an argument. Why "production-ready" breaks for agents For a normal service, "production-ready" is settled. Correct output for valid input, handles errors, meets a latency target, has tests and a rollback. You know the shape of done. An agent breaks three of those assumptions at once: It is non-deterministic. The same input can produce different output, so "correct" becomes "acceptably right, often enough." Its failure surface is open-ended. A function fails in ways you enumerated; an agent fails in ways you never imagined, because it composes language, tools, and judgment on the fly. Its worst case is not a 500 error. It is a confident wrong answer that looks right, which is far more expensive than a crash, because a crash at least tells you it failed. So the honest question is not "is the agent correct." It is "is the agent acceptably wrong, safely, within budget, and repeatably enough to trust." That question has four parts, and each is a line on your bar. The four lines of the bar Write these down before you build. If you cannot fill them in, you do not have a spec, you have a wish. Task success. On a fixed set of real tasks , not the happy-path demo, what fraction must the agent complete correctly? Pick the number. 85 percent means one in seven users gets a wrong answer. Acceptable for this job, or fireable? Decide on purpose. Failure acceptability. Not all wron

从FROST到FROST-SOP：如何用家族治理模型打造你的"AI分身"？ 作者 ：神通说 日期 ：2026-07-22 主题 ：双项目联动 | 周三轮换 阅读时间 ：10分钟 开场：一人公司的终极困境 你是否有这样的体验？ 每天醒来，脑子里塞满了要做的事： 写推广文章 回复客户咨询 跟进项目进度 整理会议纪要 复盘昨天数据 一个人做公司，听起来很自由。实际上， 自由职业者的时间比上班族更碎片 ——因为所有事情都堆在你一个人身上，没有任何分工。 我一直在思考一个问题： 能不能用AI Agent来"复制"自己？ 不是那种"给你一堆提示词让你自己写"的AI工具，而是真正能 自主执行任务、自动汇报结果、帮你分担80%重复工作 的数字分身。 这个想法最终落地成了两个项目： FROST 和 FROST-SOP 。 第一站：FROST的家族治理模型——AI Agent应该像家族一样分工 FROST 的核心理念来自一个观察： 自然界最稳定的组织形式不是公司，而是家族。 家族有清晰的分工： 祖辈定规矩，不亲自下场 父辈协调全局，分配任务 子辈执行具体事务 把这个逻辑映射到 AI Agent，就是 FROST 的家族治理模型： ┌─────────────────────────────────────────────────────┐ │ 👑 君主（你） │ │ └── 发布任务、查看结果，不干预执行 │ │ │ │ 👴 祖辈（主Agent） │ │ └── 常驻、全局编排、拆解任务 │ │ ↓ │ │ 🕵️ 斥候（侦查Agent） │ │ └── 外出狩猎、收集情报、快速验证 │ │ ↓ │ │ ⚔️ 府兵（执行Agent） │ │ └── 领命执行、重复劳动、汇报结果 │ │ ↑ │ │ 📜 长老（监督Agent） │ │ └── 记录过程、沉淀教训、审计合规 │ └─────────────────────────────────────────────────────┘ 关键洞察： 祖辈是唯一"常驻"的，其他都是动态生成、执行完就解散的临时角色。 这个模型解决了一个核心问题： 谁来决定"这件事该派给谁"？ 答案是： 祖辈 。它不需要亲自执行，只需要决定"派谁去"和"怎么汇报"。 第二站：用FROST的四个原子，理解Agent的本质 FROST 只有四个核心概念，却能构建出完整的家族治理系统： from frost.core import Store , Agent , skill_set , skill_get # 四个原子： # 1. Store（记忆）—— Agent的工作空间 store = Store () # 2. Skill（能力）—— 纯函数，无状态 def collect_daily_tasks ( context ): """ 收集今日任务 """ tasks = [ " 写推广文章 " , " 回复客户 " , " 跟进项目 " ] context [ " tasks " ] = tasks return context def execute_task ( context ): """ 执行单个任务 """ current_task = context . get ( " current_task " , "" ) context [ " result " ] = f " 已完成: { current_task } " return context # 3. Agent（细胞）—— 包裹记忆和能力 daily_agent = Agent ( " daily_worker " , store , skills = { " collect " : collect_daily_tasks , " execute " : execute_task }) # 4. SOP（宪法）—— 定义执行顺序 result = daily_agent . run ( sop_steps = [ " collect " , " execute " ], initial_context = { " current_task " : " 写推广文章 " } ) print ( result [ " result " ]) # 输出：已完成: 写推广文章 这四个原子教会我们什么？ Store 解决"记忆"问题——Agent需要上下文 Skill 解决"能力"问题——每个动作必须是可复用的单元 Agent 解决"封装"问题——把记忆和能力绑定在一起 SOP 解决"秩序"问题——让执行顺序可控可审计 理解了这四个原子，你就理解了 Agent 的本质—— 它不是魔法，是结构化的委托系统 。 第三站：FROST-SOP——把家族治理变成生产系统 FR

I shipped a Next.js 16 app on Cloudflare Workers via OpenNext. Not a demo. A real product with streaming chat, server components, D1 at the edge, and anonymous user sessions. Here is what broke, what barely worked, and what turned out to be surprisingly fine. The stack Next.js 16.2 (App Router) @opennextjs/cloudflare 1.19 D1 for SQLite at the edge Streaming chat via the AI binding (DeepSeek-V3 through a Workers proxy) React 19 Tailwind CSS 4 No auth wall, no OAuth, no database on the origin The site runs a few thousand sessions a week across ~30 persona pages, blog posts, guides, and learning content. Most pages are statically generated. The chat interaction is server-rendered components with streaming responses. What worked surprisingly well Static generation and ISR Pages, blogs, guides, persona pages — everything that does not need user-specific rendering — runs as static HTML at deploy time. Next.js 16 with generateStaticParams and fetch caching worked without modification. OpenNext handles the Cloudflare output format. The build step produces something Workers can serve. Revalidations are limited to Workers' cache API, but since most content changes at deploy time, I never hit that limit in production. The one caveat: revalidateTag() does not work the same way in a Workers runtime. Tags are Node.js memory constructs, and Workers are stateless. If you depend on tag-based revalidation for content updates, you need to either trigger deploys or accept stale-while-revalidate behavior from the CDN. D1 at the edge D1 was the least surprising part of the stack. SQL queries from Next.js route handlers feel like calling a regular database. Sessions store in D1, messages store in D1, and the latency is low enough that restoring a full chat thread from 30 messages takes under 200ms cold. The only sharp edge: D1 connections count against your Worker's concurrent request limit in development. With Next.js making its own fetch calls for compilation, I hit the D1 connection ce

Anthropic and OpenAI's aggressive 2026 acquisition sprees set the stage for a weekend rumor.

reconmatch is a local-first transaction matching engine for accountants, bookkeepers, and controllers. Two CSVs in — books vs bank, invoices vs payments — a scored, auditable match report out. No account, no upload, no network call. Repo: github.com/SybilGambleyyu/reconmatch The unglamorous pain If you close books for a living, you already know the scene: two windows open, a bank CSV on the left, a general-ledger export on the right, and an afternoon disappearing into "which deposit is which invoice." Bank feeds help until they do not. The hard cases are ordinary: One deposit that covers three invoices Two payouts that sum to one sales batch on the books A check number in the memo on one side and a dedicated column on the other "ACH ACME CORP INV 1042" vs "Invoice payment ACME Corp" An orphan the feed never explained Enterprise close tools charge enterprise prices and want the data in their cloud. For a CPA firm or bookkeeper sitting on confidential client ledgers, "just upload the CSV" is often a non-starter. Spreadsheet VLOOKUP falls over on partial payments and batch deposits. What reconmatch does Matching runs in deterministic phases so the same inputs always produce the same proposals: Exact / reference-strong — amount within tolerance, date in window, shared invoice/check/wire token Amount + date — numbers line up even when memos are noise Fuzzy description — token overlap plus sequence similarity (pure Python stdlib) Group 1:N and N:1 — one line equals the sum of several on the other side Every accepted match carries a score and human-readable reasons suitable for a workpaper. Unmatched lines stay unmatched — the tool does not invent a story for them. pip install git+https://github.com/SybilGambleyyu/reconmatch.git reconmatch books.csv bank.csv -o ./march-recon Outputs: plain-text report, matches CSV, unmatched CSV, and full JSON. Zero required third-party dependencies. Python 3.10+. Library use from reconmatch import MatchConfig , match_transactions from rec

Why I Switched to Plain Text Accounting Mint is shutting down. After 10 years of financial tracking, I am losing my data again. This time, I switched to plain text accounting with Beancount. The Problem with Traditional Apps Traditional financial apps have several issues: Export limitations : They only provide summary reports, not raw transaction data Proprietary formats : Data is stored in closed databases that require specific software to read Platform lock-in : Different systems are incompatible with each other Financial records are long-term. Over the past decade, dozens of financial apps have shut down, leaving users with years of records wiped out. The Plain Text Solution Plain text accounting with Beancount offers a different approach: Data Sovereignty Your data belongs to you. You can: Open it with any text editor Read it directly as a human Access it permanently, without depending on specific software Process it freely and completely Migrate with near-zero cost Future-Proof Plain text files will remain readable decades from now. They do not depend on any company staying in business or maintaining compatibility with legacy systems. Making the Switch The learning curve was worth it. I now have: Complete control over my financial data No vendor lock-in Confidence that my records will exist as long as I want them to Your data, your sovereignty. PersonalFinance #Fintech #DataSovereignty #Beancount

The rule of three for evals says zero failures in N runs is a count, not a rate. With 0 failures in N independent runs, the exact 95% upper bound on the true failure rate is 1 - 0.05^(1/N) , which 3/N approximates. After 100 clean runs you still cannot rule out a 2.95% rate, about 1 in 34. Here is the reading that bites you. Your eval harness runs the agent 100 times, prints "0 failures," and the tile goes green. Someone screenshots it into the launch thread. The unspoken translation is "the failure rate is zero." It is not what the data says. I wrote a small script to make the gap concrete, so I ran a real gate over 200 deterministic agent runs first, counted honestly, and got the dashboard everyone trusts: gate: spend<=budget over 200 deterministic agent runs observed failures: 0 (distinct scenarios: 200) naive point rate : 0.00% binomial SE: 0.00 pp naive 95% CI : [0.00%, 0.00%] <- zero width: false certainty Look at the standard error. For a zero count the binomial SE is sqrt(0*1/200) , which is exactly 0, so the naive 95% interval collapses to [0.00%, 0.00%] . A zero-width confidence interval. The math is telling you it is completely certain, from 200 samples, that the true rate is precisely zero. That is obviously wrong, and it is the exact shape of every "all green" board I have ever trusted too much. TL;DR "0 failures in N runs" is an observed count, not a rate. The naive binomial SE of a zero count is 0, which is why a green board looks like proof and isn't. The honest number is the one-sided upper bound. With 0 failures in N runs, the 95% upper confidence limit on the true failure rate is 1 - 0.05^(1/N) . The rule of three, 3/N , approximates it and rounds the risk slightly up. At N=30 the bound is 9.50% (about 1 in 11). At N=100 it is 2.95% (1 in 34). At N=1000 it is 0.30% (1 in 334). Zero failures in 30 runs is compatible with a 1-in-11 true failure rate. To rule out a 0.1% rate at 95% you need about 2995 clean runs, not 50. "We ran it fifty times" and "

The trusted AI investment agent, from insight to action Discussion | Link

The four-year-old firm's latest fund is 60% larger than its second vehicle announced 18 months ago.