Appendix B · Research guide v0.1

Interview Consent and Guide

45 minutes · Target n 15-25 · Feeds Chapter 4 (buying committee stalls)

Structured qualitative interviews for The Prove Gap Report (Q3 2026). Recruits opt in via survey Section Z (SZ3) or design-partner outreach. Aevesa-sponsored; research only, not a sales call.

45-minute structure

BlockTimeGoal
A Warm-up5 minRole, agent landscape, PMM priors
B Stack and Prove Gap10 minLayers deployed, missing proof
C Sign-off and stall12 minHidden buyers, timeline, rejected evidence
D Evidence bar10 minAccepted vs rejected artifacts
E HITL and architecture5 minApproval binding, diagram gaps
F Close3 minWish-list, quote permission

Block A - Warm-up (5 min)

1. In one sentence, what is your organization trying to accomplish with AI agents in the next 12 months?
2. What is your role in production decisions for agent workloads?Who else must say yes before write or tool access in production?
3. How many agent workloads on inventory, and how many have tool or write access?
4. If you completed the survey: your PMM Level was [X]. Does that feel accurate?

Block B - Stack and Prove Gap (10 min)

5. Walk me through your agent stack as you would draw it for a risk committee.Is there a distinct Prove layer, or is it folded into logging?
6. What do you use for Route and Classify today?Production or pilot only?
7. Where does runtime enforcement happen before a high-risk tool executes?
8. When you hear "Prove Gap," what gap do you actually have?Policy on paper, enforcement, or evidence auditors trust?

Probes: Prompts are not policy. Logs are not proof. Policy theater vs post-mortem governance vs disconnected HITL.

Block C - Sign-off and stall (12 min)

9. Tell me about the most recent agent workload that stalled or failed production sign-off.Stage, duration?
10. When did Legal or audit first ask you to prove what the agent did?
11. Who were the hidden buyers engineering did not expect?
12. What would have unblocked sign-off: budget, evidence, or architecture?

Block D - Evidence bar (10 min)

13. What artifact closed or almost closed your last production sign-off?
14. What evidence was rejected as insufficient?
15. Could audit re-verify the same action 90 days later with today's artifacts?
16. Have you run offline verification without vendor UI login?What would acceptable proof look like?

EU/UK probe: Art. 12 logging and Art. 14 oversight shaping approval flows. DENY paths documented?

Block E - HITL and architecture (5 min)

17. Where does human approval live: Slack, email, ticketing, or crypto-bound receipt?
18. If you added one box to your architecture diagram this quarter, what would it be?

Block F - Close (3 min)

19. What do you wish you had before your last exec or risk committee meeting?Champion Kit artifact validation.
20. What do peers get wrong about agent governance in 2026?
21. May we use an anonymized quote (role + industry only)?
22. Anything we should have asked?

Post-interview capture (24h)

IDField
INT_IDInterview ID (INT-2026-###)
INT_PMM_ESTInterviewer PMM estimate 0-5
INT_STALL_STAGEPrimary stall: Security · Legal · Audit · Procurement · Budget
INT_EVIDENCE_ACCEPTEDArtifacts that worked
INT_EVIDENCE_REJECTEDArtifacts rejected
INT_PROVE_GAP_TYPEPolicy · Enforcement · Evidence · Mixed
INT_QUOTE_CANDIDATEBest anonymous quote (≤280 chars)
INT_CHAPTER4_TAGStall map · Security · Legal · Audit · Procurement · Champion

Interviewer protocol

Do

Do not

Saturation target (n=15): ≥10 stall stories, ≥8 evidence-rejection anecdotes, ≥5 HITL-not-bound examples.