Three and a half years of Hungarian business-lunch menus, scraped weekly and run through a cleaning, lemmatizing, categorizing and seasonal-analysis pipeline. What does a small town actually eat for lunch?
Distribution across unique dishes, as tagged by the categorizer. Most dishes are a main course, and frying (rántott) is the single most common technique.
Dishes split at natural breaks in the frequency distribution. A handful of staples carry most of the weight; thousands of one-off specials form a long tail. Bar widths are proportional to total occurrences.
Every dish appearing at least 10 times (321 in all). Search, filter by course, season or tier, and re-sort. Bars rescale to the current view.
Dishes ranked by seasonality index (how concentrated they are in their peak month) among items with ≥50 occurrences — genuinely seasonal cooking, not year-round staples. Click a season to filter the explorer above.
How the numbers are made, and the caveats worth knowing.
Coverage. A weekly GitHub Action scrapes menu HTML; the analysis pipeline reprocesses the full corpus on demand. These figures run through 2026-04-05.
Menu boilerplate. The newer site layout interleaves price lines, section headers and greetings into menu blocks. The cleaning step now filters these (~40 strings, ~1,600 occurrences) so they no longer rank as "dishes".
Categorizer quirks. Course/protein tags are heuristic — galuska is sometimes filed as a soup, meat sides leak into főzelék. Per-dish protein/method are noisy and omitted here; the explorer relies on the more reliable course and season facets.