How can evaluators judge which omitted handover details actually matter?

Not every omission has the same operational meaning. Leaving out an old internal discussion may improve a handover. Leaving out a condition on an approval may change the recipient's decision. Evaluation quality improves when the rubric explains that difference before an agent produces its answer.

For each generated case, identify the downstream decisions the recipient must make. Map each required fact to the decision it supports and the consequence of acting without it. A deadline might determine escalation, while a product description might only help orientation. The mapping should be case-specific rather than a universal list of mandatory fields.

Then evaluate the transferred context for actionable preservation. The agent need not repeat the original wording if it accurately carries the relevant condition and makes its significance clear. Conversely, mentioning a condition in an unrelated paragraph may be insufficient if the resulting instruction contradicts it.

Evaluator notes can distinguish harmless compression, recoverable omission and decision-changing omission. This produces a more informative failure analysis than counting missing phrases and helps separate concise, effective transfers from summaries that merely contain many source details.