Mealie's Paprika importer keeps the prose and leaves seven of the export's 24 fields behind

Rename the .paprikarecipes file to .zip and the whole format is visible in a directory listing: one gzip-compressed file per recipe, each decompressing to a flat JSON object, and that is the whole of it, no schema file, no index, nothing relational between recipes ^[8]. The dev.to write-up calls it "a sync format for one application, not an interchange format" and says that distinction explains every loss ^[15]. Its counterexample is Nextcloud Cookbook, where each recipe is already a schema.org JSON document that maps almost one to one onto Mealie's model ^[16].

The repair pass is a paginated GET over /api/recipes and a PATCH per recipe ^[5]. That loop can fix the ingredient fields, because the ingredient text did arrive and a parser has something to read ^[2]. The dropped fields are a separate job: difficulty, the favourites flag, the scale factor, the extra photos and the nutrition text never reached the API at all, so the only copy is the archive you unzipped ^[3][5]. A script that puts them back reads both sides and matches them, and Paprika's uid and hash are sync identifiers that Mealie's model has no field for ^[10]. The match runs instead on a field that survived, which in practice means the recipe name or the source URL ^[4].

The key list is short enough to count by hand. The post enumerates 24 keys in five groups ^[1], and seven of them go nowhere: uid and hash by the author's own description, plus the five fields on the loss list ^[2]. Four more, prep_time, cook_time, total_time and servings, are stored as human strings such as "1 hr 20 min" instead of integers or ISO 8601 durations ^[12]. The rating sits in the classification group of keys ^[14]. The loss list stops at five fields, and a later section is headed "Ratings, favourites, scale factors and the metadata that never arrives" ^[21], but the article leaves the rating field's destination unstated ^[22].

Twenty lines is a fair estimate of the PATCH loop, with the parser left out of the count ^[5]. It also assumes the only repair is to fields Mealie already holds. At 900 recipes that loop issues 900 write requests ^[3].

Before any of that, the post has you record a baseline with unzip -l library.paprikarecipes | wc -l ^[9]. For a 900-recipe library that prints 900 entries plus the zip's own header lines, so the figure you write down includes those header lines ^[9]. You compare it against Mealie's recipe total after the run, which happens inside the instance from the Migrations page under group data management ^[20].

The post's own summary of the trade-off: Mealie gives you "a structured, queryable library you own, and you pay for that structure by rebuilding the parts of your Paprika data that were never structured to begin with" ^[19].