Japan Trip Planner

Everyone plans a trip the same way: a folder of saved videos, a document somebody started two years ago, and a group chat. All three are write-only. Nothing in them is searchable, nothing is on a map, and nothing tells you whether the place is even open in the month you are going.

This merges them. A Python pipeline pulls rows out of the TikTok Vault catalogue and out of a shared document, geocodes what it can, deduplicates against what is already there, and emits one typed list. A place is allowed to carry several sources, which turns out to be the useful signal rather than noise: the same onsen turning up in a saved video and in a document written two years earlier is the strongest recommendation in the dataset.

Every place is then rated for the specific week of the trip, not for Japan in general. Half of what makes a place worth going is seasonal, and a planner that tells you somewhere is wonderful without telling you it is shut in February is worse than no planner, because it costs you a day.

The page itself is password gated at the edge, in front of the static asset, so the HTML of a private page is never sent to an unauthenticated request. It fails closed: with no password configured nothing is reachable at all, which is the right way round for a gate, since a missing secret must never mean an open door.