实验室 — Lavi Sahu

07实验室 — 建站日志

我搭这个站点的方式，就是我希望 AI 供应商搭规划系统的方式：亮出 diff，而不是递上推销词。

实验室 — 开源

八个仓库，不是样机

公开、能跑的代码，不是作品集幻灯片：供应链规划工具、一套 AI 配置系统、一个离线语音听写应用，还有搭起这个页面本身的那套设计系统。点开卡片，看看它究竟能干什么。

Resilience Radar — Python (stdlib only) — self-contained HTML dashboard

Supply chain resilience control tower — risk radar, what-if simulation (time-to-recover / time-to-survive), and mitigation ranking. No dependencies to install; the dashboard is one self-contained HTML file.

git clone https://github.com/LaviSahu/resilience-radar.git

What the code actually does

# risk_score = likelihood(1-5) * impact(1-5), range 1-25.
TIER_THRESHOLDS: list[tuple[int, RiskTier]] = [
    (20, RiskTier.CRITICAL),
    (12, RiskTier.HIGH),
    (6, RiskTier.MEDIUM),
    (1, RiskTier.LOW),
]

def likelihood_score(event: Event, classification: Classification) -> int:
    """1-5, from event confidence (0-1) plus a source-reliability adjustment."""
    base = event.confidence * 5
    adjustment = SOURCE_RELIABILITY_ADJUSTMENT.get(classification.source_category, 0)
    return _clamp(base + adjustment, 1, 5)

src/resilience_radar/scoring.py — classic likelihood × impact risk scoring, plus a per-source reliability adjustment (geopolitical/regulatory signals score noisier than, say, a supplier or logistics event).

实验室 — 建站日志

验证台账 — 上面这些东西，每一块实际上是怎么查的

人眼看过

Every animation on the rebuilt pages

The merge-conflict resolutions of 2026-09-19

读了，没跑过

The twin's system prompt

Each repo description (2 of 8 carry verbatim code excerpts)

没人检查过

Everything the twin tells you

Whether anyone has ever used these repos

第三列才是诚实的那列。它只在缺口消失时才会清空 — 不会更早。

实验室 — 建站日志

别只听我说 — 亲手扳动杠杆。

三件能上手玩的工具 — 情景模拟器、库存计算器、牛鞭效应图。没有幻灯片，没有空话：你扳动杠杆的每一刻，每个数字都在你的浏览器里实时算出来。

实验室 — 建站日志

一个真的 AI 功能，不是样机

这不是内嵌一个聊天机器人。这是一个 Cloudflare Pages Function 直接调用 Workers AI — 域名锁定到本站，按 IP 限流，系统提示词要求它宁可说「我不知道」，也不编造客户名字或数字。问问它 S&OP 或库存规划。

问问我的 AI 分身

未经核实

一个演示：一个语言模型，带上一段用我的口吻写的短提示词。它可能错得信心十足 — 别照它说的话行事。问它 S&OP、IBP、o9、库存。

问任何关于规划的事。分身会用我的口吻回答。

Twin 自评 — 语气：接近。事实：没有任何人核实过，包括我。模型：@cf/meta/llama-3.3-70b-instruct-fp8-fast，上限 512 tokens，提示词要求它说「我不知道」而不是编造客户名字或数字。这是一道护栏，不是一份保证。

AI 生成 · 不能替代真正的 Lavi

实验室 — 建站日志已合并，未上线。

建站日志 — 2026-09-19

实验室 — 建站日志refkit

Refkit — 28 ports, 1 in production

Every card below is a real VengeanceUI port — working code, not a screenshot. Exactly one of them, VeCreepy, is actually load-bearing right now: it's the Connect link in this site's own footer. The other twenty-seven are verified to render, not yet asked to carry a real page's weight. Same rule as the build log above — a component only counts as shipped once it's doing a job, not once it looks good in a gallery.

Lab — essays

Five frameworks, worked through properly

Not case studies from my own client work — original, research-driven essays on real named frameworks and formulas (bullwhip ratio, TTR/TTS resilience, the Kraljic matrix, Theory of Constraints, digital twins), each grounded in a cited real source, not invented. Pick a card for the summary, or read the full essay.

For the fuller story behind the bullwhip math — two 2000s-era factory fires and what they cost — see Buffer Shock, on the Blog →

# risk_score = likelihood(1-5) * impact(1-5), range 1-25. TIER_THRESHOLDS: list[tuple[int, RiskTier]] = [ (20, RiskTier.CRITICAL), (12, RiskTier.HIGH), (6, RiskTier.MEDIUM), (1, RiskTier.LOW), ] def likelihood_score(event: Event, classification: Classification) -> int: """1-5, from event confidence (0-1) plus a source-reliability adjustment.""" base = event.confidence * 5 adjustment = SOURCE_RELIABILITY_ADJUSTMENT.get(classification.source_category, 0) return _clamp(base + adjustment, 1, 5)

Refkit — 28 ports, 1 in production

Five frameworks, worked through properly