Upload a spreadsheet, answer a few plain-language questions, and get a statistically trusted Shewhart Control Chart — with an interpretation that tells you whether the change you're making is a real improvement or just your system's normal variation.
Always available at charts.hthgse.app
What is a Shewhart Control Chart?
Developed by Walter Shewhart in the 1920s to solve the problem of prediction, a Shewhart Control Chart plots your data in time order (or by subgroup) and draws the boundaries of the variation your system would produce on its own. Its visual nature makes it readable by everyone — no deep statistical knowledge required.
If only common causes are present, the system is stable — reacting to single points wastes effort, and only redesigning the system will improve it. A special cause is different: something specific happened, and understanding it is the first step to improvement.
In Arizona, a team working on FAFSA completion used these charts to discover three schools quietly outperforming the rest of their system — a bright spot nearly lost to inattention. Positive deviance, once seen, can be studied and replicated.
A chart shows you where meaningful change happened — not why. That's the next move: go and see. Visit, listen, and learn in context. Bright spots found on a chart become practices worth studying — and spreading.
Why this tool
From our experience supporting improvement teams, the biggest barrier to learning from variation is that creating a Shewhart Control Chart usually means learning a new statistical package first. This tool removes that barrier — the analysis becomes a conversation with your own data.
Why learn from variation?
All data wobbles. Some of that wobble is the system's natural rhythm; some of it is a real change worth investigating. When we can't tell which is which, we make two costly mistakes — chasing wrong explanations, or missing the moment something truly shifts. A Shewhart Control Chart draws that line for you.
Reacting to points that are simply part of the system's natural, predictable ups and downs.
Treating a meaningful signal as if it were just another up-or-down fluctuation.
Ready when you are
Bring a spreadsheet and a question. The generator handles the statistics.
The chart family
You don't need to know which one you need. The generator looks at your data, asks about your question, recommends the fit — and always confirms with you before drawing.
For attribute data — you're counting how many, how often, or what share.
For continuous data — things you measure rather than count.
When most periods would show zero, chart the space between events instead.
For getting oriented — and for questions that aren't about time order.
Privacy
School data deserves more caution than most software gives it. This tool is built to need as little of your data as possible — and to keep none of it.
Your upload is parsed in memory and never written to persistent storage — no database, no bucket, no saved file. When your session ends, your data is gone. What we keep is anonymous usage counts — turns, timings, chart types — never your messages, your file, or its contents.
Every upload comes with a plain reminder: no student names, IDs, or personal identifiers — aggregated counts and rates are all a control chart needs. The assistant reads your column structure before analysis, and when something looks student-level, it says so and helps you move to aggregated data instead.
To do its work, a small sample — your column names and up to 20 rows — is sent to Anthropic's Claude API for reasoning (Anthropic doesn't train on API data), and your full table goes to our own chart service, which computes your chart in memory and keeps nothing.
No account. No login. Nothing stored.