Start here: one workflow, six questions

Protocol 0.1-rc.4-candidate.2 · Engineering commitment only

You do not need to read the full protocol to begin.

This is a guided conversation before investing engineering time. It helps you see what is demonstrated, what is still assumed, and what to do next. It does not certify an AI system or approve a launch.

Bring three things

  1. A recent example of how the work happens today.
  2. The proposed screen, workflow, prototype, or code revision.
  3. Someone who knows the work and someone accountable for the next decision. One person may fill both roles in a small team.

An advisor can record the answers while the participant talks. No technical setup is required for a paper review. The MCP and Skill help record the same evidence; they are optional interfaces.

Begin with this question

“Walk me through one recent case—from the request arriving to the work being finished. Where did someone have to check, fix, or chase it?”

If nobody can show a current case or quantify the work, stop here. The next action is to observe the current workflow. Do not invent a baseline to justify a prototype. For a genuinely new service, investigate the closest real workaround or leave the baseline and ROI unresolved.

Choose the path together

Low risk, with records already available: use QUICK-6.
Higher or unknown risk, missing evidence, or a longer conversation: use FULL.

The 15-minute target is untested and does not include undisclosed preparation. Record preparation and session time separately. Needing more time or an accessible format is not a failure.

What you leave with

Example: a booking prototype looks finished, but nobody can show what happens after a payment failure. The useful result is not “72% ready.” It is “Define and validate recovery of the booking details; the product owner owns that revision.”

Examples are constructed, not pilot results. Read more scenarios, inspect the criteria, or see what changed.