Nuclear Corrective Action Desk

Independent prototype, by The AI Pipe

Twelve findings, eleven answers. The twelfth is that two of them are one.

An audit closed on Friday. The agent reads each finding in the auditor's own words, classifies it, looks up the clause it answers to, and computes every date the request runs on. Then it does the part nobody has time for: it tests the extent of condition across the programme registers, on the attributes the records actually carry, and finds that two findings raised in different elements by different auditors stand on the same unkept column. It never determines whether a condition is significant. It cites the criterion that says who does.

Measured here, end to end on the whole desk: run one and watch. That covers classification, the clause lookup, the clocks, the request pre-fill, the extent of condition and the tracker. It does not perform the audit, does not inspect an item, does not determine significance, does not determine a cause, does not disposition anything, does not notify any customer, does not verify closure and reports nothing to any regulator.

What this shows

The problem
Findings arrive from an internal audit, a customer surveillance and a supplier audit. Somebody has to classify each one the same way every time, tie it to the clause it answers to, start the clocks, write a request that a department can actually answer, and notice the thing no single finding says: that two of them have the same mechanism underneath.
The result
The run below is real and is recorded: twelve findings classified, the site clause, the public criterion, the customer clause number and every date computed from the procedure, and the extent of condition tested across four registers with every line carrying the reference it came from.
What is simulated, and what is not covered
The supplier, the twelve findings, the four registers, the department response and the four site documents are invented. Two documents are not: 10 CFR 50 Appendix B and the NRC Inspection Manual definitions of root cause, extent of condition and extent of cause are quoted verbatim with their source. CSA N299.2-16 is copyrighted and no line of it appears anywhere here: the clause number each element answers to is read out of a public audit checklist that a Canadian operator publishes for its own suppliers, and the number is named without the text. There is no SharePoint connection, no Power BI dataset, no Copilot and no Power Platform app: what the page shows is the structured output such a list would receive. Performing the audit, inspecting an item, determining significance, determining a cause, dispositioning anything, notifying a customer and verifying closure are all outside this desk.
What a pilot would need from your side
Your blank corrective action request form, your corrective action and audit procedures in whatever form they exist today, the column headings of whatever holds your CAR log rather than its contents, a short conversation with whoever runs that log and one person who can say whether a classification is right, and counts rather than findings: how many arrive from each source, how many are reclassified after the first read. Suppliers, findings and departments stay invented until your legal and security leads have agreed, in writing, which fields a pilot may see, who holds them, where they sit, how long they are kept and how they are deleted. No finding record is needed to start, and removing names is not anonymisation.
Who built it
Frederic de Lavenne de Choulot, The AI Pipe. Built as a demonstration, not sold as a quality or regulatory product, and not affiliated with IS International Services.

Findings on the desk

Programme registers

Procedures and public text

Fictional desk. Do not paste real audit, supplier or quality records.
Enter to send, shift and enter for a new line. Live tool use over twelve findings, four registers, four site documents and two public texts.