Prove Gap Report Survey Instrument (Appendix A)

Section 0 - Screening

Terminate if S0A = none, or unqualified role/size combo.

ID	Prompt	Response
`S0A`	Organization’s use of AI agents	Chat only · Planning · Pilots w/ tools · Production w/ tools · None (terminate)
`S0B`	Your role	Security · Compliance/Legal · Platform · Audit · Procurement · Executive · Other
`S0C`	Organization size	<500 · 500-4,999 · 5,000-19,999 · 20,000+
`S0D`	Regulated industry?	Yes · No

ID	Prompt	Response
`S1A`	Primary industry	Financial · Healthcare · Tech · Manufacturing · Retail · Public · Services · Other
`S1B`	HQ region	NA · EU/UK · APAC · MEA · LATAM
`S1C`	Deployment contexts (multi)	Customer-facing · Employee · IT/ops · Dev/coding · Back-office · R&D
`S1D`	Influence on production decisions	Final approver · Strong influencer · Contributor · Observer
`S1E`	Agent workload count (optional)	Integer or Unknown

Feeds Chapter 1; validates 14% sign-off hypothesis via S2C.

ID	Prompt	Response
`S2A`	Furthest stage any workload reached	Ideation · Policy only · Non-prod pilot · Limited prod · Broad prod w/ write
`S2B`	Formal agent inventory?	No · Informal · Yes, quarterly+
`S2C`	Full security/compliance sign-off (Likert 1-5)	Validates ~14% approval stat
`S2D`	Blocked or rolled back after pilot?	No · One · Multiple · N/A

Multi-select per layer. Derived: STACK_LAYER_COUNT.

ID	Layer	Options (multi-select)
`S3A`	Document	Policy · Risk register · Architecture diagrams · Vendor DD · None
`S3B`	Route	LLM gateway · API mgmt · MCP routing · Direct API only · None
`S3C`	Classify	AI-SPM · Guardrails · DLP · Classification tags · None
`S3D`	Enforce	Model guardrails · Pre-exec intercept · Policy engine · HITL gate · None
`S3E`	Prove	Vendor logs · SIEM · Decision records · Tamper-evident receipts · Prove product · None
`S3F`	Verify	Security offline test · Audit offline test · Third-party portal · Runbook · None

Likert 1-5. Derived: CONFIDENCE_GAP = S4A minus mean(S4B,S4C,S4D). Validates 82% confidence hypothesis via S4A.

ID	Statement
`S4A`	Executive leadership is confident our agent policies protect against harmful actions.
`S4B`	We can prove what each production agent did, under which policy version, at decision time.
`S4C`	Runtime enforcement exists on high-risk tool paths before side effects.
`S4D`	An external reviewer could verify our agent evidence without vendor logins.
`S4E`	Our observability stack satisfies internal audit for agent workloads.

Same questions as the PMM widget. Calculates PMM_LEVEL 0-5 and PROVE_GAP flag.

ID	Widget	Prompt
`S5A`	q1	Autonomous agents at your organization (0-3)
`S5B`	q2	Governance documentation maturity (0-2)
`S5C`	q3	Route layer: traffic direction (0-2)
`S5D`	q4	Classify layer: inventory and protection (0-2)
`S5E`	q5	Runtime enforcement on high-risk paths (0-3)
`S5F`	q6	Proof artifacts for decisions (0-3)
`S5G`	q7	Offline verification without vendor login (0-2)
`S5H`	q8	Ongoing assurance / drift monitoring (0-2)

PMM gates: Level 4 requires S5F≥2 and S5G≥1; Level 5 requires S5F≥3, S5G≥2, S5H≥1. See outline Chapter 3 and Appendix C rubric.

Feeds Chapter 4. S6B ranks top 3 from S6A selections.

ID	Prompt
`S6A`	Factors that stalled sign-off (multi): Security · Legal · Compliance · Audit · Procurement · Budget · Integration · Alignment · No stall · Other
`S6B`	Rank top 3 stall factors
`S6C`	Longest single review cycle: <4wk · 1-3mo · 3-6mo · 6mo+ · N/A
`S6D`	Hidden buyers drive more delays than engineering (Likert 1-5)

ID	Prompt
`S7A`	Evidence requested for sign-off (multi): Policy PDFs · Architecture · Pentest · SIEM · SOC2 · Screenshots · Crypto receipts · Offline demo · Third-party verifier · None
`S7B`	Evidence rejected as insufficient
`S7C`	Could audit re-verify a decision 90 days later?

ID	Prompt
`S8A`	Pressures apply (multi): EU AI Act · US sector · UK/other · Board AI risk · Insurance · Customer clauses · None
`S8B`	Art. 12-style logging influences architecture (Likert 1-5)
`S8C`	Art. 14-style oversight influences approval flows (Likert 1-5)
`S8D`	Legal asked for proof of what agent did (not just policy)

ID	Prompt
`S9A`	Tool categories in stack (multi): Gateway · Guardrails · AI-SPM · GRC · Observability · SIEM · IAM · Prove layer · DIY · None
`S9B`	What is missing from architecture diagram? (open, optional)
`S9C`	Primary governance integration: Gateway webhook · Sidecar/PEP · Post-hoc export · Manual only · Not defined

ID	Prompt
`S10A`	Agent security/compliance incident in last 12 months
`S10B`	Incident type if applicable (multi, optional)

ID	Prompt
`S11A`	Where human approval lives: None · Email/chat · Ticketing · Slack w/ partial log · Crypto-bound receipt
`S11B`	Can identify approver, timestamp, policy version months later (Likert)
`S11C`	DENY decisions recorded with same rigor as approvals

ID	Prompt
`S12A`	Architecture diagram includes distinct Prove/Evidence layer?
`S12B`	Route + Classify alone will satisfy production sign-off (Likert)
`S12C`	Time to add offline-verifiable proof to one prod path

ID	Prompt
`SZ1`	Contact with aggregated benchmark results?
`SZ2`	Work email (if opt-in)
`SZ3`	45-minute research interview? (optional)
`SZ4`	Anonymous quote attribution? (optional)