Methodology & Sourcing

What we source from

The core of every article is public comments on Japanese-language YouTube videos. That is where Japanese people speak candidly to each other — the local consensus, before it is polished for tourists.

We do not use Google Maps or Tabelog reviews as a source (for policy, aggregation, and risk reasons). We may add official map links for basic venue info, but we never store rating scores.

How we choose a video

We pick videos whose comments are dense with local-knowledge signals ("as a local," "I live here," "born and raised") and concrete place names. We avoid bait-driven or appearance-focused celebrity-YouTuber videos.

How we translate

We excerpt rather than reproduce comments in full. We show the original Japanese (verbatim) alongside the English translation for transparency. We do not spin the meaning or invent statements that were never made.

Software fetches the comments and checks every quote, like count and number against the source before an article is published; AI assists with selecting, translating and summarising, under rules a person sets and reviews. Neither is allowed to invent a comment, a number or a fact.

How we count

The story is the aggregate: how many comments agreed, and how many likes the top voice earned. Every number is the real like count on YouTube — we never fabricate it. Which comments we choose, how we cluster them, and what we summarize — that editorial judgment is our added value.

How many comments we actually read

We do not read every comment. For each video we pull the top comments in YouTube's own popularity order — usually the top 60 to 200 — and work from those. When an article says a video has 11,939 comments, that is the video's total; the consensus we describe is drawn from the top slice, and each article states how many comments it was picked from. Like counts are always the real count on the individual comment.

What we leave out

Some of the most-liked comments under a video are not about the place at all — political jabs, remarks aimed at a nationality, religious cynicism, or accusations against a named business. We leave those out of the consensus, and we say so here rather than pretend the thread had none. If the top-liked comment is one of these, the article's "most-liked" figure refers to the most-liked comment that is actually about the topic. We also drop comments whose author handle we would not want to print, and we anonymise a business when the only thing said about it is negative and unverified.

Facts that do not come from the comments

Comments are opinions with a date on them. Where a reader needs a fact to act — a ticket tier, a price, a rule that changed, an opening status — we sometimes add it in the FAQ or in a place note. Every such fact carries a "Sources" line with the page we checked and the date we checked it. Anything without a Sources line is what the comments said, and nothing more. Prices and rules move; the date tells you how old our check is.

How we call a verdict

Some articles carry a small badge — “locals say go,” “locals say skip,” or “locals split.” This is not our review score. It is an editorial classification of what the comment consensus in that article already says: if the most-liked comments overwhelmingly warn people off, we mark it skip; if they overwhelmingly endorse, go; if they genuinely disagree, split. Articles that explain rules or customs get no badge, because there is nothing to be for or against. The evidence for every badge is quoted in the article itself, with like counts.

What we don’t do

• Fabricate or infer comments (we only translate statements that exist)
• Reproduce comments in full or exhaustively (we excerpt, translate, and send traffic to the source)
• Convert like counts into star ratings (AggregateRating) or otherwise misrepresent numbers
• Scrape Google Maps or Tabelog reviews

Corrections & removal

For factual errors, rights concerns, or removal requests, please use the contact form. We review and respond (correct or remove) promptly.

When we correct an article we show an updated date on it (in the dateline at the top). Where a correction changes a fact, we say in the article what was changed. Re-counting numbers and revisiting a translation are treated the same way.