14 — Checklists, glossary and sources

14 — Checklists, glossary and sources#

Handbook index

Customer discovery worksheet#

• Deployment environment, countries/regions and disconnected-operation requirement.
• Named AI tools/client versions, operating systems, API usage and uploads.
• Number of users/services, peak concurrency and burst request volumes.
• Current proxy/SSE/firewall, remote-access path, IPv6 and direct-egress controls.
• Identity source and how a gateway can authenticate each client type.
• Customer PKI and device/application trust-distribution method.
• Required deterministic/semantic categories, languages and false-positive tolerance.
• Whether raw content may reach customer-local or external inference.
• Retention, SIEM, data residency, backup and recovery objectives.
• Operating team, maintenance window and incident response contacts.
• Procurement model, billable capacity definition and support expectations.

Before pilot#

• Required client routes and supported adapters identified.
• Inspection authority and trust distribution approved by customer.
• Customer-local identity, storage and inference dependencies configured.
• Mandatory policies and failure behavior reviewed.
• Protected/uninspected/monitor-only status visible.
• Synthetic upstream proves block and masking behavior.
• Direct bypass paths tested and controlled or documented.
• Metadata exports checked for synthetic secret leakage.
• Backup and recovery owners assigned.

Before production#

• Chapter 12 tests completed for the exact release/profile.
• Capacity measured with failure headroom and realistic payloads.
• Unknown identity, expired policy, model failure and full spool tested.
• Gateway and database failure drills completed.
• Upgrade and isolated restore demonstrated.
• Certificate/key rotation and recovery demonstrated.
• Required support matrix and limitations accepted.
• Alert destinations and escalation owners verified.
• No vendor credentials, customer data or shared private CA in distribution.
• Customer can operate without unapproved vendor calls.

Before Marketplace publication#

• Seller/product setup and commercial model finalized.
• All paid nodes and infrastructure charges explained.
• Source AMI and templates satisfy current AWS review requirements.
• Image and dependency findings remediated.
• Template regions, parameters and IAM scopes tested.
• Clean independent buyer account completes launch and setup.
• License/usage behavior validated for the selected offer.
• Installation, recovery, update and uninstall guidance complete.
• Required listing diagram and usage/support artifacts prepared.
• Limited-release testing completed before public submission.

Before a release upgrade#

• Manifest/signatures verified and compatibility bounds checked.
• Backup and decryption/recovery access confirmed.
• Schema migration/rollback limits understood.
• Canary cohort and success criteria chosen.
• Failure capacity maintained during drain/replacement.
• Audit spool handling verified before deleting/replacing nodes.
• Post-upgrade policy, identity, traffic and SIEM checks passed.

Glossary#

Primary sources#

External AWS requirements were reviewed on 2026-09-19 and must be rechecked before submission. Design recommendations in this handbook are Zotline proposals, not statements that AWS or an existing partner has approved the architecture.

Repository evidence is linked in chapter 02. The original blueprint remains a shorter architecture overview.