Conversation worksheet: one case, one next action
Protocol 0.1-rc.4-candidate.3 · MCP 0.2.0rc3 · Skill/contract 0.2.0-rc.3
Use paper or your normal notes. The advisor records evidence pointers; the participant need not complete technical fields. This is a conversation aid, not a replacement for the full evidence contract or an automatic pass. Reuse existing records when appropriate.
Before the six questions
Workflow / one completed case: ______ Owner: ______ Date / run ID: ______
Where does it start and end? ______ Proposed artifact revision: ______
Risk: safety/rights effects? ___ irreversible external action? ___ sensitive data? ___ untrusted input to actions? ___
Record yes, no or unknown with a reason—not a guess. The first two “yes” answers require high-depth FULL, the others at least moderate FULL; any unknown prevents QUICK-6. Combine these with the six risk dimensions in the full profile. Path selected: ______
Ask, point to evidence, decide the next step
| Ask aloud | Advisor's short record |
|---|---|
| 1. Show one recent case of today's work. Who checked, fixed or chased it? | Steps, times, handoffs, rework, volume, observation window and evidence pointer: ______ |
| 2. Whose difficulty should improve, and how will we know? Who else is affected? | Need and success condition: ______. Access/use, privacy/security, unequal effects, human control: ______. Applicable concern → requirement/check; justify exceptions. |
| 3. When something goes wrong, who notices and how do they recover? | Normal → edge → recovery → endpoint; saved work and authority limits: ______ |
| 4. Show the requirement, the artifact and the check together. | Exact revisions and executed result: ______. Changed requirement/context or artifact? ______ |
| 5. What work remains for people? | Review ___ correction ___ escalation ___ rework ___ residual manual ___ minutes/case. Basis: ______ |
| 6. What bounded engineering step could this evidence support? | Human quality review ___; unresolved findings ___; owner ___; resource cap ___; revisit trigger ___ |
For QUICK-6, stop at the first missing or failed gate. Mark later gates NOT_EVALUATED. A known critical finding remains visible. Continue after remediation in a new linked FULL record—not by overwriting this one. “Unknown” is useful evidence about what to learn next.
Leave with an understandable result
Recommendation: consider a bounded engineering step / revise / gather evidence.
Because: ______ Next action: ______ Owner: ______ Revisit when: ______
Preparation elapsed/person-minutes: ______ Session elapsed/person-minutes: ______
Adjudication minutes: ______ Capture/reporting minutes (already inside session): ______
Estimated gross benefit: ______ Oversight burden: ______ Net effect: ______
Missing baseline means ROI is indeterminate. These are not measured operating-performance results. Separate protocol review expense from operating costs.
Ask the participant: “In your own words, what does this result allow us to do—and what does it not allow?” Record their answer before explaining again. Do not coach it into a favorable usability result.
The advisor retains exact versions, artifact/requirement fingerprints and evidence locations in the detailed private record. A fingerprint is not proof of adequate testing. A recorded human review is not independently authenticated by the software. Owner funding authorization and deployment evaluation are separate decisions.
Permission / anonymization / retention agreement: ______ Previous run if any: ______