Appendix A — Execution record reference

The appendices collect execution records, equations, library roles, and practice questions with an answer key. The Glossary and References follow them.

A release decision needs records of what the system was asked to do, what information it used, and what it actually did. This table connects each technical concept to a concrete record and the question that record can answer.

Table A.1: Execution records connect system behavior to the evidence needed to investigate it.

Technical concept	Concrete record	Question answered
Prompt specification	Versioned messages, examples, and output schema	What was the model asked to do?
Context budget	Final model input, token allocation, omissions and truncation record	What information entered or was omitted from the call?
Evaluation example	Input, expected behavior, slice labels, rubric	What does success mean for this case?
Retrieval trace	Source versions, query, filters, and the candidates, scores and selected chunks saved at each captured stage	Which evidence was found, selected and supplied?
Workflow state	Typed state snapshot and node transition	Which step changed each state field, and what ran next?
Tool trace	Tool identity, validated arguments, access decision, approval when required, attempted action and result	What action actually ran?
Agent trajectory	Ordered decisions, calls, observations, and stop reason	How did the system reach the final state?
Memory trace	Retrieved memory IDs, versions, and promotion sources	Which past information affected the run?
Outcome record	Validated answer or environment state plus metrics	Did the product meet the task and policy?