APPENDIX A: GLOSSARY OF TERMS AND ACRONYMS

TERMS BEGINNING WITH O Observability: the ability to understand what is happening inside a system by looking at the signals it gives off. Observability layer: the fifth ETCLOVG layer, covering the ability to see what happened during a run. Observability team: the team that owns tracing systems and dashboards. Observation: the result of an action, fed back to the model. Offline evaluation: testing a system on saved examples before it is used by real users. Offloading: moving bulky data to storage and keeping only a pointer in the window. On-call: the person currently responsible for responding. On-demand tool discovery: loading tool descriptions only when a task needs them, instead of listing every tool every time. Online evaluation: measuring quality on live traffic while the system is actually in use. Open standard: a publicly documented set of rules that anyone may use, so different products can work together. Open-source: software whose code is public. Open-source software: software whose source code is published for anyone to read and usually use. Open-weight model: a model whose trained numbers are released so that anyone can download and run it. Orchestration: coordinating several agents or steps so they work together toward a goal. Orchestration pattern: a named, reusable way of arranging how agents cooperate. Origin authentication: proving where a piece of software or a message truly came from. Outcome: the final real-world result of the agent's work. Output schema: the agreed structure of the information a tool sends back. Output tokens: tokens the model generates. Overfitting: tuning so narrowly to test cases that performance on new cases suffers. Overlap: shared text between neighboring chunks. Over-reliance: trusting a system more than its record deserves. Over-tuning: improving a system for a test rather than for real work.