Insights & Architecture
Deep dives on risk-informed design, code-to-diagram automation, continuous compliance, and trust engineering.
A code scanner finds a hundred places where something called email appears. How many data items is that? Joining on the name alone dropped our grouping precision from 86% to 51%. The fix was to stop trusting names and join only on evidence the parser can show you.
Sorting data item mentions into data items is an agent task: look at a candidate group, read around it, decide. A frontier coding model does it well and costs too much to run on every scan. We built the same loop around Jev, gave it parse-tree tools instead of a file viewer, and it landed above the…
Every decision about how a system handles personal data now gets made in a conversation with a coding agent, and the transcript is sitting on disk. We wrote a tool that reads those transcripts. The next step is an interviewer that forms a hypothesis from them and asks you to confirm it, instead of…
You cannot improve a scanner you cannot score, and nobody had labeled where the personal data lives in a real codebase. So we had three sizes of model label it, argue about it, and let the biggest one arbitrate. The labels are not perfect. They are a compass needle, and the first thing they found w…
A dataflow diagram is not the output of a scan. It is a query over a knowledge base, and the gaps in that knowledge base tell you exactly which questions to ask a person. We built an interviewer on that idea. It drew its first system-context diagram after one question. It also makes dumb decisions,…
