An activity log shows what an agent did, but a decision log should also record what was considered and rejected. Without rejected options, reviewers see a simplified path that doesn't reflect the actual decision-making process. This omission hinders trust and recovery, as it makes it difficult to challenge a decision without reconstructing the entire session. Execution history, while valuable, differs from decision context.A proposed decision record separates choice from execution, including a "revisit_when" field for rejected options. This indicates that a rejected option might be suitable under different circumstances. The interface should employ progressive disclosure, providing layers of information from a general overview to specific details. When execution fails, the decision and execution results must be kept separate to avoid implying that the chosen action succeeded.A companion validator can ensure the decision log's completeness, checking for context, evidence, rejected options with reasons and revisit conditions, and execution details. This validation highlights omissions, providing a stable surface for review tools. Before making this pattern default, research involving operators reviewing agent work with different log formats is recommended.This research should measure explanation accuracy, recovery time, and the usefulness of detailed information. Rejected options are crucial for understanding decision boundaries, providing the counterfactual context needed for questioning, repairing, and learning from agent traces. These proposals aim to improve the transparency and auditability of AI agent decisions.