A working agreement, not a permission slip

What is the smallest honest pilot?

Build the smallest end-to-end test that can resolve a real uncertainty—and is allowed to stop.

Open Work 001 decides whether AI should be involved. Use this only after the answer is proceed or prototype first.

Keep protected work protected. Do not enter anything covered by a contract, NDA or duty of confidence—even if it has been generalised. Use placeholders or the fictional example instead. This tool shapes future work; it does not ask for evidence from past contracts.

About eight minutes
No email gate
Nothing is stored or sent
Public weakest-boundary logic

What makes a pilot honest?

It tests one uncertainty through real work, compares against a baseline, names who is accountable and defines what would make the work end. Missing governance cannot be offset by a persuasive use case.

No disclosure is required. Use fictional details or describe only future work you are free to discuss. This is not a request for evidence of work already delivered.

01Purpose

Name the real work and the uncertainty. AI is not an outcome.

What repeated work are you changing? What should become better without mentioning AI? What one uncertainty must the pilot resolve?

02Shape

Small is useful only when the test still reaches the real workflow.

What is the smallest end-to-end change? What is explicitly outside this pilot?

03Evidence

Decide what would count before the demonstration can charm you.

What current result or process is the baseline? What evidence will support the final decision?

04Governance

These are hard boundaries. Missing one means the pilot is not ready to run.

What cannot happen without a human decision? Who can continue, change or stop the pilot? How long may the pilot run? What result or event means it should stop?