4. 돌아와서, 기록을 잇다

After the trip, we had more than an itinerary and photos: a shared plan, guidance sent to the family, changes made on the ground, receipts, and photo metadata. This final article brings those records together to review what Luan and the home server actually did, what was verified, what remained incomplete, and what could be improved next time.

We began with a travel webpage, then connected ntfy on the home server, revised Calendar, the travel document, and message text during the trip, and compared those changes with receipts and photo metadata afterward. The outcomes were a plan we could return to, scheduled guidance, and records we could reconcile. Each stage supplied evidence for the next. This work carried into Day, but the verified results do not establish a finished, fully automatic service covering every device and every stage.

The first planned notification was sent manually about four minutes late. After execution moved to a user systemd timer on the home server, the next scheduled run returned exit code 0 and four HTTP 200 responses. On October 4 at 10:00, four messages for the 11:00 dolphin show were accepted, with recipient routes and message text checked against the logs. Two next-day notices and one letter sent on a separate October 3 request were also recorded as successful. These are verified examples, not an overall delivery rate or a measure of satisfaction. Server acceptance, device display, and human acknowledgment are separate outcomes; the child’s reaction to the letter was not recorded.

Moving the bus departure to 14:00 on October 4 required coordinated changes to Calendar, the travel document, and message text. The old 14:30 notification was removed, and the already-expired 13:00 advance notice was not sent late. We also checked 29 guidance start times against Calendar; that number describes consistency checks, not successful sends. Unknown costs stayed marked as unknown. Keeping several representations consistent required work, suggesting a clearer source of truth and revision linkage next time. The family made the on-the-ground decisions; Luan organized information around them.

After returning, we placed receipts, the existing EXIF list, and confirmed changes beside the itinerary. One Beppu day could be ordered as African Safari in the morning, Umi Jigoku, Kamado Jigoku, and Shiraike Jigoku in the afternoon, and Chinoike Jigoku in the late afternoon. Receipts from the safari, Umi Jigoku ticket office, and Chinoike Jigoku shop supported the photo sequence. This established relationships between records, not exact arrival times or a complete traveled route. The full set of original photos was not rechecked, and missing expenses or photo gaps were not reconstructed. The illustrations retain fictional examples or broad time bands and sequence, without exposing personal photos, exact times, or coordinates.

Four gaps stood out. A scheduled job still needed execution checks; one verified run did not establish a trip-wide absence of delays. Immediate notification and later retrieval were different: Cache: no on Day’s publishing request prevented later lookup. Removing it and deploying the change verified 24-hour caching of user-sent messages, with 40 server tests and sender-label regression tests passing, but persistent visibility in the iOS app remained unverified. Server acceptance also did not establish that someone read or acknowledged an important message, and connectivity remained a dependency. Finally, retrospective evidence was scattered across several sources, leaving some reasons and gaps unresolved.

These are proposed improvements, not claims of completed implementation. Before departure, test scheduled execution, locked-screen display, and later retrieval on real devices, including authentication failures, wrong recipients, duplicate runs, restart, and lost connectivity. Link each schedule revision to its document, message, and job, checking old-job cancellation together with the new guidance. Distinguish ordinary notices from messages requiring a person’s acknowledgment, and agree on a meeting point if connectivity is lost. During the trip, attach brief reasons and source references to changes; afterward, review unresolved items first and check shared material for unnecessary personal photos or location data.

The concrete outcome was an experience connecting planning, guidance, changes, and records. The webpage provided a reference; ntfy carried notices; on-the-ground decisions were reflected in Calendar and the document; receipts and photo metadata helped reconstruct the sequence afterward. Each operational question required different evidence: execution time, schedule freshness, server acceptance, later retrieval, and human acknowledgment. Preserving both the verified results and the unfinished checks makes this experience useful for the next trip and for the work continuing in Day.