A repeatable way to get a class of customer use cases into production — architecture, rollout, common blockers, support model — documented after the third similar launch.

The day-to-day workflow followed by frontline operators, analysts, reviewers, or support staff who will actually use the deployed system.

A customer-facing technical partner who has earned enough trust to challenge the customer's assumptions, explain tradeoffs honestly, and guide the deployment toward production rather than toward a good demo.

A deliberately adversarial or high-risk test case used to probe model, agent, security, or workflow weaknesses before production.

A concrete deliverable — diagram, runbook, eval set, integration spec, launch checklist, code sample — that helps a deployment move forward or be handed off.

A launch-blocking issue in the deployment path — often not an engineering problem but unclear business ownership, missing approval, or an unresolved data issue.

A blocker specific to trusting model output: hallucinations, missing citations, brittle prompts, failed evals, or unclear accountability for wrong answers.

A threshold used to decide when model or agent output is good enough to proceed, requires human review, or should fall back to a simpler path.

A reusable architecture for a class of deployments: system boundaries, integrations, security patterns, and common variants. Built from a real deployment, not invented from scratch.

A metric tied to the performance of a specific workflow — cycle time, backlog, error rate, throughput, escalation rate. Should move when the deployment works.