No database was a product decision.

Open Work 001 asks seven questions about a proposed AI intervention. Some answers may reveal weak ownership, sensitive data, uncertain risk or a service that should not use AI at all.

That made the first product decision straightforward: Kingsbury should not receive the answers.

The tool has no account, form endpoint, database or analytics call. Selections live in a JavaScript object for the current page. The scoring runs locally. The decision record is assembled locally and downloaded through a temporary browser object. Close or reload the page and the answers are gone.

This was not a privacy line added after the tool was built. It shaped the tool.

What the constraint improved

There is no consent ceremony because there is no transfer to explain. A visitor can start immediately, inspect the method and leave with a useful record without becoming a lead.

The architecture is smaller too. There is no authentication, storage model, retention policy, deletion flow or protected results page. Fewer moving parts means fewer ways for a deliberately small diagnostic to acquire an unjustified operating burden.

Most importantly, the incentive stays clean. The tool exists to help someone make a better decision. It does not need to turn uncertainty into contact data before providing value.

The implementation is inspectable in the public decision engine. Its seven questions, weighted choices, hard-stop conditions and result logic are readable without a build process or account.

What the constraint cost

Local-only state is not universally better. It has consequences:

• answers do not follow a person to another device;
• a closed or refreshed page cannot recover the session;
• Kingsbury cannot observe completion rates or learn which questions create difficulty;
• a team cannot share a live result link;
• there is no central history of past decisions.

Those could become useful features. They would also change the data responsibility and the nature of the relationship. The absence of a database is justified for this version because the diagnostic takes minutes, produces a downloadable record and does not need continuity to fulfil its purpose.

If later evidence shows that saved sessions materially improve the decision, storage can be reconsidered with an explicit retention and access design. It should not arrive because persistence is the default shape of a web product.

The observation

Data minimisation is easiest when it is an architectural decision, not a policy promise.

“Nothing leaves your browser” is a property of Open Work 001 that a reader can test against the code. That is stronger than collecting the answers and asking the reader to trust an explanation about what happens next.

The limitation is real. So is the freedom it creates.