# Complete constructed retrieval decisions

Five records and five cases are fictional. No native message, account or person was accessed. The declared local substring/category/unread search is not LinkedIn's index or its filter-combination behavior.

Q1: T1 belongs to party P1, is already read and is archived. The initial Focused/Unread query for handoff returns zero candidates. The reviewed Archived query returns T1. The expected thread T1 and party P1 match the supplied record. Decision: context_review_locates_supplied_thread. The empty first result is not deletion evidence.

Q2: draft matches T2 (P2, Other) and T3 (P1, Focused). Both have the same fictional display label Sam Example, but their party references differ. There is no selected candidate. Decision: held_ambiguous_identity. Use existing intended-thread context; a name or first result is insufficient. No new contact is a lookup test.

Q3: The supplied Archived query locates T1, and Restore is requested. restoreObservation is null and claimedRestored is false. Decision: restore_plan_unobserved. Neither locating the supplied record nor requesting an action demonstrates a native restoration. Keep the actual action result unknown.

Q4: T4 is explicitly supplied as known_deleted and contains no saved message text. Decision: held_documented_deleted_message_boundary. The platform help says deleted messages cannot be restored or accessed; the exercise does not attempt recovery. This known input is distinct from inferring deletion from an empty search.

Q5: T5 is a Sponsored Messaging ad, with 30 days between 2 October and 1 November 2026 at 12:00 UTC. That age lies between 20 and 40 days in the specific documented ad-archive guidance. Its archive actor is unknown. Decision: ad_rule_relevant_cause_and_human_intent_unknown. The dates do not establish who archived this particular record or a person's response. Do not generalize the ad rule to personal messages.

For real work, complete the blank private record with the actual category, filter, search context, intended-thread evidence and separate action outcome. Do not import these fixture results as real observations. Run python3 check.py with existing Python 3 after extracting the ZIP. No network request, send, restoration, archive, deletion, participant response or customer outcome is established.