Walter Shewhart invented the control chart in 1924. It was — and remains — one of the most important ideas in quality engineering. But it's been a hundred years, and the math hasn't moved. Your X-bar chart still uses ±3σ limits calculated from an initial baseline. Your capability index still assumes the process is stable and normally distributed. Your rules for detecting special causes still rely on Western Electric patterns from the 1950s. None of this is wrong. But there's more information available than these methods extract. Bayesian SPC gets it. ## What Bayesian SPC Actually Does Traditional SPC asks: "Is this point outside the control limits?" That's a binary question with a binary answer. Bayesian SPC asks: "Given everything I've observed, what do I believe about this process right now?" That's a fundamentally different question — and the answer is a probability distribution, not a yes/no. Here's what that unlocks: ### Adaptive Control Limits Fixed control limits assume the process doesn't change. In reality, processes drift, improve after interventions, and shift when raw materials change. Bayesian adaptive limits update their belief about the process parameters as new data arrives. When a genuine shift occurs, the limits adapt — rather than forcing you to manually recalculate after a "Phase II" study. ### Online Changepoint Detection (BOCPD) Bayesian Online Changepoint Detection doesn't just tell you a point is out of control. It estimates *when* the process changed and *how confident* you should be that a change occurred. Instead of applying run rules and debating whether three points in Zone B constitute a signal, you get a posterior probability that a changepoint occurred at each observation. ### Bayesian Cpk Traditional Cpk gives you a point estimate. Bayesian Cpk gives you a posterior distribution. The difference matters enormously when your sample size is small — which it always is in the real world. A traditional Cpk of 1.33 from 30 samples might have a 95% credible interval of [0.9, 1.8]. That's the difference between "capable" and "who knows." ### Anytime-Valid Sequential Testing Classical hypothesis tests require you to fix your sample size in advance. If you peek at the data — which everyone does — your p-values are inflated and your false positive rate explodes. E-processes and anytime-valid confidence sequences let you monitor continuously without penalty. Test after every batch. Stop when you have enough evidence. No guilt, no adjustment needed. ### Decision-Theoretic Alarms Classical SPC treats all false alarms equally and all missed detections equally. In practice, a false alarm on a high-volume automotive line costs $50,000 in downtime. A missed detection on a medical device costs lives. Bayesian decision-theoretic alarms let you specify asymmetric loss functions — the cost of stopping versus the cost of missing — and optimize the alarm threshold accordingly. ## Why This Matters for Manufacturing If you're running SPC in a food plant, semiconductor fab, or automotive tier-1, you already know the frustration: your charts trigger on noise, miss subtle drifts, and give you point estimates when you need uncertainty bounds. Bayesian SPC doesn't replace Shewhart. It extends him. You still get control charts. You still get capability indices. But now they carry the uncertainty information that lets you make better decisions. ## What's Available in Svend Svend includes 35+ Bayesian SPC capabilities: - **Bayesian Cpk** with posterior distributions and credible intervals - **Adaptive control limits** that update as the process evolves - **BOCPD** for online changepoint detection with posterior probabilities - **Anytime-valid sequential testing** using e-processes - **Bayesian DOE** for design of experiments with prior information - **Bayesian Gage R&R** that propagates measurement uncertainty - **Decision-theoretic alarms** with configurable loss functions - **Bayesian acceptance sampling** based on posterior defect rates These sit alongside all the classical SPC tools you'd expect — I-MR, X̄-R, X̄-S, p, np, c, u, CUSUM, EWMA — plus 200+ additional statistical analyses. Minitab doesn't offer any of this. Neither does JMP. ## Getting Started Every Svend account — including the free tier — can run Bayesian Cpk calculations. The [free Bayesian Cpk calculator](/tools/bayesian-cpk-calculator/) doesn't even require an account. For the full Bayesian SPC suite, any paid plan includes everything. No add-on modules, no per-seat licensing for advanced features. A hundred years is a good run for any methodology. It's time for the update.