Westwiki - User contributions [en]

站点账号轮换策略说明

2026-10-29T08:37:33Z

GroveBrook: [ProfileMax] 发布测试

== 为什么要轮换站点账号 ==

同一账号在高频发布节奏下容易被站点风控标记,轻则限速、重则封禁。把发布压力均匀摊到站点下的多个账号上,是保持长期可用性的基础手段。

== 轮换策略 ==

* '''最近最久未用''':每次发布优先选择 lastLoginAt 最靠前的可用账号;
* '''失败退避''':登录失败或写入被拒的账号直接进入冷却,下一次投递自动换人;
* '''凭据分层''':active 状态且密码齐全的账号才进入候选池,无凭据账号会被健康检查标 expired。

== 操作约定 ==

批量发布作业按站点串行投递,同一站点同一时刻只占用一个浏览器身份(Profile),发布成功后回写账号最近使用时间,保证轮换计量准确。

轮换器的实现与发布测试共享同一套账号池,压测示范站点为 [https://txwiki.com/wiki/Main_Page Txwiki]。

<blockquote>账号池告警阈值:单站点可用账号少于 2 个时,应导入新凭据而不是压缩发布节奏。</blockquote>

发布质量检查清单

2026-10-29T08:34:24Z

GroveBrook: [ProfileMax] 发布测试

== 发布后质量检查清单 ==

每一篇发布成功的文章都要过以下检查,任一项不过都应进入修复流程。

=== 基础可达 ===

* 发布页 HTTP 状态码 200(重定向后可达);
* 页面标题与文章标题一致。

=== 链接与 rel ===

* 目标链接在页面 HTML 中存在且 href 与投放计划一致;
* rel 属性按投放计划携带(nofollow/sponsored 等),缺失记 missing;
* 无死链(跳转后 404 即 dead)。

=== 排版与内容 ===

* HTML→wikitext 转换无内容丢失(标题/列表/加粗原样保留);
* 段落无明显截断,末尾完整。

示范站点 [https://txwiki.com/wiki/Main_Page Txwiki] 的检查记录可在「发布 → 质量评分」Tab 查询。

站点账号轮换策略说明

2026-10-29T08:32:01Z

GroveBrook: [ProfileMax] 发布测试

== 为什么要轮换站点账号 ==

同一账号在高频发布节奏下容易被站点风控标记,轻则限速、重则封禁。把发布压力均匀摊到站点下的多个账号上,是保持长期可用性的基础手段。

== 轮换策略 ==

* '''最近最久未用''':每次发布优先选择 lastLoginAt 最靠前的可用账号;
* '''失败退避''':登录失败或写入被拒的账号直接进入冷却,下一次投递自动换人;
* '''凭据分层''':active 状态且密码齐全的账号才进入候选池,无凭据账号会被健康检查标 expired。

== 操作约定 ==

批量发布作业按站点串行投递,同一站点同一时刻只占用一个浏览器身份(Profile),发布成功后回写账号最近使用时间,保证轮换计量准确。

<blockquote>账号池告警阈值:单站点可用账号少于 2 个时,应导入新凭据而不是压缩发布节奏。</blockquote>

MediaWiki 批量发布指南(压测版)

2026-10-29T08:28:14Z

GroveBrook: [ProfileMax] 发布测试

== 为什么做批量发布压测 ==

批量发布的价值不只在吞吐量,更在于把「发布成功率」从主观印象变成可验证的指标:站点可用性、凭据有效性、落栈结果、以及发布后的回读校验,全部由系统记录并汇总成质量评分。

本轮压测使用 15 个 MediaWiki 压测站点,每个站点绑定一个专属发布镜像(指纹浏览器 Profile)与站点账号池,发布作业由调度器按站点串行投递,避免同账号并发发布触发风控。

== 账号导入与凭据管理 ==

* 站点账号密码在导入时即以 AES-256-GCM 加密落库,API 永不返回明文,只在投递时解密注入执行载荷;
* 每个站点保留原始创建时间与最近登录时间,账号轮换按「最近最久未用」进行;
* 无有效凭据的账号会被健康检查标记为 expired,不再进入投递候选。

== 发布流程与回读校验 ==

发布作业分两种模式:'''publish''' 直接落正式页面,'''draft''' 落到账号个人空间的沙盒页,用于在不动正式内容的前提下验证登录与写入链路。落栈成功后系统记录 revisionId,并用 HTTP 回读发布页:确认页面可访问(HTTP 200)、目标链接存在且 rel 属性符合预期。

<blockquote>发布结果(status/message/revisionId/articleUrl)与目标链接检查记录都会在「发布任务 / 质量评分」Tab 中按站点汇总,压测结论以这两处数据为准。</blockquote>

== 压测站点附录 ==

本轮参与压测的站点包括 youngwiki、eastwiki、hopewiki 等 15 个 MediaWiki 环境,完整站点清单见「发布 → 站点」Tab;示范站点为 [https://txwiki.com/wiki/Main_Page 压测站点 Txwiki]。

Vibe Protein Design: How Natural Language Is Changing Gene-Design Software

2026-10-28T19:49:19Z

GroveBrook: **Vibe protein design** refers to an emerging practice in which a person describes, in everyday natural language, the kind of protein they would like an artificial intelligence (AI) system to produce, and the system responds with candidate molecular

'''Vibe protein design''' refers to an emerging practice in which a person describes, in everyday natural language, the kind of protein they would like an artificial intelligence (AI) system to produce, and the system responds with candidate molecular designs. The label borrows from software culture, where "vibe coding" describes building programs by stating what you want rather than writing every line by hand. Applied to biology, the same idea is beginning to reshape gene-design software: the interface is shifting from technical parameters to conversation, and protein design — long one of the most technical corners of biotechnology — is where that shift is most visible.

== What It Is ==

The short version is striking: you can apparently now tell AI tools what sort of protein you want them to make. Instead of filling in forms, adjusting numeric sliders, or scripting a design pipeline, a user writes something closer to a plain-language brief — the protein's intended job, the conditions it should work under, the properties it should have — and the tool proposes molecular candidates that match the description.

The word "vibe" captures the spirit of this interaction. In software, vibe coding means describing an application's desired behavior in casual language and letting an AI handle the implementation. Vibe protein design applies the same logic to molecules: the human supplies the outcome they care about, and the AI infers the molecular details that might achieve it. The description does the work that blueprints used to.

Part of what makes this notable is accessibility. Protein design has traditionally required specialized training and purpose-built software, which kept participation limited to a relatively small community. If a natural-language interface can carry part of that technical weight, the earliest stage of design — turning an idea into a concrete proposal — becomes open to a much wider range of people. That possibility, more than any single feature, explains the attention the approach has received.

The qualifier "apparently" matters. These capabilities are new, and the tools behind them are young; how well they perform outside carefully prepared demonstrations is still being explored. What can be said with confidence is directional: gene-design software is moving toward natural language as a primary interface, and the full consequences of that move are only beginning to be tested.

== Core Concepts ==

A few biological and computational ideas explain why this shift is possible.

'''Proteins are defined by sequence and shape.''' A protein is a chain of smaller chemical units, amino acids, strung together in a specific order. That order — the sequence — determines how the chain folds into a three-dimensional shape, and the shape largely determines what the protein does: what it binds, what it breaks down, what signals it carries. Change the sequence and you can change the function. This link between sequence, structure, and function is the foundation of all protein design.

'''De novo design aims for molecules that do not exist in nature.''' Rather than borrowing a protein found in an organism and modifying it, designers try to create entirely new sequences with desired behaviors. Historically this has been slow, expert-intensive work, guided by physical models,

== References ==
* [https://arstechnica.com/science/2026/10/vibe-genes-naturalish-language-comes-to-protein-genome-design-software/ Vibe genes? Natural(ish) language comes to protein, genome design software]

[[Category:vibe]]
[[Category:natural]]
[[Category:you]]
[[Category:ai]]
[[Category:to]]