故事 · Goldratt 与《关键链》
Origin · Goldratt and Critical Chain
1997 年,约束理论之父 Eliyahu Goldratt 在小说《关键链(Critical Chain)》里把矛头指向了项目管理。
他发现一个荒诞循环:人估工期时,为了不被追责,总把估算塞到 90% 把握(实际只需一半时间)。
可这些安全垫并没有让项目按时 —— 学生综合症让人拖到截止前才动手,帕金森定律让工作膨胀填满所有时间,
任务延误会向后传递、提前却从不向前。Goldratt 的解法:把每个任务砍到 50% 激进估时,
再把所有被砍掉的安全时间汇成一个集中缓冲挂在关键链末端。波动彼此抵消,这个共享水库远小于各自为政的安全垫总和。
In 1997 Eliyahu Goldratt, father of the Theory of Constraints, turned his attention to project management in the novel Critical Chain.
He spotted an absurd loop: to avoid blame, people inflate every estimate until they're 90% confident they can hit it — even though the real work needs about half that time.
Yet none of that padding kept projects on schedule. Student syndrome made people wait until the deadline before starting; Parkinson's law let the work swell to fill the time allowed;
late tasks pushed the chain back, but early finishes were never passed forward. Goldratt's fix: cut every task to its 50%-confidence estimate,
then pool all the safety you stripped out into a single buffer at the end of the critical chain. Because the variations partially cancel, that shared reservoir is far smaller than the sum of the individual pads.
1 传统每任务塞安全垫 vs CCPM 末端集中缓冲 Traditional Per-Task Pads vs CCPM End-Loaded Buffer
n=52 平方根法则:为什么集中缓冲能更小 The Square-Root Rule: Why a Pooled Buffer Is Smaller
分散安全垫是简单相加(Σ)随任务数线性增长;集中缓冲按 √(Σ 安全时间²) 增长 —— 任务越多,两条线差距越大。这就是"风险池化":独立波动相互抵消。 Scattered pads simply add up (Σ) and grow linearly with task count. The pooled buffer grows as √(Σ safety²) — the more tasks you have, the wider the gap between the two lines. That's risk pooling in action: independent variations partially cancel each other out.
3 现实里的 CCPM 缓冲 CCPM Buffers in the Real World
工程项目:基建、新品研发的多任务链路,用项目缓冲消化天气、供应、返工的随机延误,而非每环节各塞冗余。
Engineering projects: construction and new-product programs use one project buffer to absorb random weather, supply, and rework delays — instead of padding every individual step.
缓冲消耗看板:实时盯缓冲被吃掉多少(绿 / 黄 / 红区)来决定是否干预,比逐任务追进度更聚焦。
Buffer fever chart: track how much of the buffer has been consumed in real time (green / yellow / red zones) to decide when to act — far more focused than chasing individual task progress.
汇入缓冲:非关键链汇入关键链的接点,放小的"汇入缓冲"保护关键链不被支线延误拖累。
Feeding buffers: where a non-critical path joins the critical chain, a small feeding buffer protects the chain from being dragged late by side branches.
多项目环境:用能力约束缓冲错峰排程,避免多项目抢同一资源导致全员同时卡死。
Multi-project setting: a capacity-constraint buffer staggers the schedule so multiple projects don't fight for the same bottleneck resource and stall together.
一句话In One Line
CCPM 的核心是两步:激进估时 —— 把每个任务砍到 50% 把握,逼掉学生综合症与帕金森定律;
集中缓冲 —— 把抽出的安全时间汇成一个共享水库挂在末端。
关键洞察是风险池化:n 个独立任务的波动不会同时往坏处走,按平方根法则相加,
√(Σ 安全²) 远小于直接 Σ。于是更短的承诺工期反而更能兑现 —— 你不再为每个任务买保险,而是为整条链买一份团体险。
这正是把约束理论(关键链 = 资源约束下的最长路径)搬进项目管理的成果。
CCPM rests on two moves: 50-50 estimating — cut every task to its 50%-confidence number to squeeze out student syndrome and Parkinson's law; and
a pooled buffer — collect the stripped-out safety into one shared reservoir at the end of the chain.
The key insight is risk pooling: variations across n independent tasks don't all go bad at once. By the square-root rule,
√(Σ safety²) is far smaller than a plain Σ. So a shorter committed date actually becomes more reliable — you stop buying insurance for each task and buy one group policy for the whole chain.
This is what it looks like to bring the Theory of Constraints (critical chain = the longest path under resource constraints) into project management.
常见误用Common Mistakes
砍了任务估时却不给项目缓冲。激进估时必须配末端缓冲,否则只是在压榨工期、人人加班赶死线。
Cutting task estimates without adding a project buffer. 50-50 estimates only work when paired with an end-loaded buffer — otherwise you're just squeezing the schedule and forcing everyone into death-march overtime.
把缓冲当成可以随意挪用的余量。缓冲是用来吸收随机波动的,要监控消耗速度(绿 / 黄 / 红),而非默许提前花掉。
Treating the buffer as free slack to spend as needed. The buffer exists to absorb random variation — track its consumption rate (green / yellow / red), don't let teams burn it early "just in case".
给每个任务都设固定截止日。任务无个人死线,接力棒一交就开干(接力赛规则),避免提前不传递、延误却累积。
Giving every task its own fixed due date. Tasks don't get personal deadlines — start the moment the baton arrives (relay-race rules) so early finishes get passed forward instead of being hoarded.