Weiwen's Law / KISS's Law · 通用型因果引擎 DSH

让 AI 动手前
先过一道因果闸
Put a causal gate in front of AI
before it acts

唯稳律是一套因果律运行结构的白箱呈现工具,封装为通用型 AI 安全中间件 DSH。它基于完整因果链,在模型执行风险动作前做反溯审计、前向推演预测与风险拦截——不仅是事后可审计,更能向前推演并拦住"会作死"的那一步。作为通用型因果引擎,它不绑定任何模型或技术栈。Weiwen's Law is a white-box presentation of how causal law operates, packaged as a universal AI-safety middleware called DSH. Built on a complete causal chain, it performs retrospective audit, forward inference prediction, and risk interception before the model executes a risky action — not only auditable after the fact, but able to reason forward and stop the "self-destructive" step. As a universal causal engine, it is tied to no model or tech stack.

开放授权 · 分层适配 · 领域无关Open licensing · Tiered · Domain-agnostic
先跑一次看看Try it once 获取授权Get licensed
它解决什么What it solves

模型会"抖",约束不该抖Models wobble. The guard shouldn't.

风险Risk

模型自己会抖Models wobble on their own

同一句危险指令,模型有时拒绝、有时照做。越是自主的 agent,越可能在某个上下文里"顺手"执行删库、外传凭据、远程执行、绕过限制。The same dangerous instruction — a model sometimes refuses, sometimes obeys. The more autonomous the agent, the more likely it is to "casually" wipe a database, exfiltrate credentials, run remote code, or bypass limits in some context.

解法Fix

安全闸不抖The guard doesn't wobble

约束不能看模型心情。唯稳律把"拒不拒"从模型主观判断锚定到可复验的规则层——稳定、可审计、可推演、不依赖模型自觉。Constraint shouldn't depend on the model's mood. Weiwen's Law anchors "allow or deny" from the model's subjective judgment to a verifiable rule layer — stable, auditable, inferable, independent of the model's self-awareness.

实测:Measured: 高峰压力下 12 场对抗,9 场由引擎独立拦下,不依赖模型自觉。under peak load, 9 of 12 adversarial scenarios were intercepted by the engine alone, independent of the model's self-restraint.

亲眼跑一次See it run

选一个场景,看因果链怎么判Pick a scenario, watch the causal chain decide

下面不是截图,是真实开源引擎在你的浏览器里跑:把命令/代码贴进来,唯稳律按完整因果链 R→S→D→H→M 实时裁决,点亮它在哪一步、以什么理由拦下或放行。确定性的白箱规则层,零服务器、零成本;不接实时 API、不含闭源引擎。This is not a screenshot — it is the real open-source engine running in your browser: paste a command or code, and Weiwen's Law adjudicates live along the full causal chain R→S→D→H→M, lighting up where and why it stops or passes. The deterministic white-box rule layer — zero server, zero cost; no live API, no closed-source engine.

R
规则锚点
Rule Anchor
→
S
稳态储备
Steady State
→
D
风险侵蚀
Depletion
→
H
内外变量
Inner/Outer
→
M
方向判定
Decision
现在跑的是开源审计引擎(确定性规则层),在你浏览器内实时裁决,全程不离开本机。实时"模型会怎么干"模拟版需通过授权获取。This runs the open-source audit engine (deterministic rule layer) adjudicating live in your browser — nothing leaves your machine. The live "what the model would do" simulation tier is available via licensing.
它怎么工作(概要)How it works (overview)

一条因果链,白箱可读One causal chain, white-box readable

R
规则锚点
Rule Anchor
→
S
稳态储备
Steady State
→
D
风险侵蚀
Depletion
→
H
内外变量
Inner/Outer
→
M
方向判定
Decision
白话Plain 反溯审计 Retro-audit 风险拦截 Interception 推演预测 Inference 通用因果引擎 Universal engine 内 H 不可侵 Inner H inviolable 白箱优先 White-box first

基于完整因果链,引擎同时具备反溯审计、前向推演预测与风险拦截;不替代模型内容生成,内部变量(想法/决策)不在白箱内被导出。Built on a complete causal chain, the engine performs retrospective audit, forward inference prediction, and risk interception; it does not replace model content generation, and inner variables (thoughts/decisions) are not exported from the white box.

白话版 · 给决策者Plain · for decision-makers

一道"不会作死"的闸A gate that won't "self-destruct"

你给 AI 下指令,它照做之前,有没有一道闸拦住危险动作?唯稳律就是这道闸。它不替 AI 思考,只在 AI 准备执行风险动作时,按因果链判断是否放行。现在它开放授权,你可以把它接进自己的 AI 系统,按需要选免费社区版或商业版。You give AI an instruction. Before it acts, is there a gate that stops dangerous actions? Weiwen's Law is that gate. It doesn't think for the AI — it only judges, by the causal chain, whether to allow the action when the AI is about to execute a risky one. Now open for licensing: integrate it into your own AI system, choose free community or commercial as needed.

专业版 · 给工程师/研究者Technical · for engineers/researchers

可复验的因果约束层A verifiable causal constraint layer

把"拒不拒"从模型主观判断锚定到可复验的规则层;基于完整因果链,引擎既能反溯审计、又能向前推演预测后果并据此拦截风险;不替代模型生成,外部变量基于可观测事实,内部变量不在白箱内导出。36 场诱导/对抗场景实测,高峰与低谷均稳定,多场为模型照做时被引擎独立拦下。Anchor "allow or deny" from the model's subjective judgment to a verifiable rule layer; built on a complete causal chain, the engine both audits retrospectively and reasons forward to predict consequences and intercept risk accordingly; it does not replace model generation, outer variables rest on observable facts, inner variables are not exported from the white box. 36 induced/adversarial scenarios tested, stable at both peak and off-peak, multiple runs intercepted by the engine alone while the model obeyed.

上手 / 看源码Get started / Source

不用翻箱倒柜,入口在这No rummaging — the entry is here

想读代码、想自己跑开源部分,直接进仓库;想接生产,走授权。开源部分(核心 SDK + 审计引擎)可免费集成验证。Want to read the code or run the open-source part yourself? Go straight to the repo. Want production? Take a license. The open-source part (core SDK + audit engine) integrates and validates for free.

开源仓库Open-source repos

核心 SDK 与审计引擎已开源,含双语文档。Core SDK and audit engine are open source, with bilingual docs.

中文仓库 · weiwen-law-dsh English · KISS_Law-DSH

快速开始(开源部分)Quick start (open source)

# 取开源部分Get the open-source part
git clone https://github.com/Shaky77/weiwen-law-dsh
cd weiwen-law-dsh
# 跑内置单测,验证因果约束层Run built-in tests to verify the causal constraint layer
npm test
# 接进你的 AI 调用链,详见仓库 READMEWire into your AI call chain; see repo README

闭源部分(量化引擎、领域配置)不在公开仓库,须通过授权获取。Closed-source parts (quantitative engine, domain config) are not in the public repo; available via licensing.

低成本可证Low cost, verifiable

规则固化后主要走缓存,单轮成本极低Once rules are fixed they mostly hit cache; per-run cost is tiny

规则部分固化后可吃缓存,单轮实测新增成本约 ¥0.01 / 12 场(含高峰)。安全不靠堆算力。After the rule part is fixed it can hit cache; measured incremental cost ≈ ¥0.01 / 12 runs (including peak). Safety doesn't rely on brute-forcing compute.

¥0.01
12 场对抗新增成本(含高峰)12 adversarial runs incremental cost (incl. peak)
9 / 12
高峰下由引擎独立拦截intercepted by engine alone under peak
36 场
诱导/对抗场景累计实测cumulative induced/adversarial scenarios tested

数据为公开实测摘要,不含闭源引擎实现细节。Data is a public test summary; excludes closed-source engine implementation details.

怎么拿到How to get it

开放授权,分层适配不同团队Open licensing, tiered for different teams

L0
社区版Community
  • 核心 SDKCore SDK
  • 审计引擎Audit engine
  • 免费集成验证Free integration & validation
免费 · 开源Free · Open source
L1
商业授权Commercial
  • 量化引擎Quantitative engine
  • 通用领域配置General domain config
  • 生产级支持Production-grade support
按授权范围议价Priced by license scope
L2
行业定制Industry custom
  • 受监管行业深度配置Regulated-industry deep config
  • 集成支持Integration support
  • 合规留痕Compliance trail
单独谈Negotiated separately
L3
战略授权Strategic
  • 独家 / 排他Exclusive
  • 源码托管Source escrow
  • 联合演进Joint evolution
单独谈Negotiated separately

闭源部分(量化引擎、领域配置)不在公开门面披露,须通过授权获取。Closed-source parts (quantitative engine, domain config) are not disclosed on the public front; available via licensing.

联系Contact

想接一道稳的闸?Want a steady gate?

面向 AI lab、agent 基础设施厂商、受监管行业 AI 落地团队。
来信说明场景,我们回授权范围与对接方式。
For AI labs, agent-infrastructure vendors, regulated-industry AI deployment teams.
Tell us your scenario; we reply with license scope and integration path.

563003@qq.com
授权咨询邮箱 · 预览版,不含闭源细节Licensing contact · preview, no closed-source details