I finally figured out how to filter for thingies within, say, the province of Groningen, and to load everything into JOSM:
(this example extracts everything with operator starting in Biblionet; turns out that’s the public libraries operated by Biblionet Groningen)

[out:xml][timeout:90]; 
//{{geocodeArea:Netherlands}}->.searchArea;  // search within entity called Groningen (city) 
area[admin_level=4]["name"="Groningen"][boundary=administrative]->.searchArea; // specify: within province (admin_level=4) 
( 
  node[operator ~ '^Biblionet'](area.searchArea); 
); 
out meta;

Unclassified and residential highways within a place, say Ezinge, plus recursions up and down (partly from this link ):
(One lesson I learned the hard way: it’s dangerous to merge and/or split ways in that type of sparse editing. Do make sure to “download parent ways / relations” in JOSM, and still be careful! It’s very easy to break relations such as bus lines / routes)

[out:xml][timeout:90];
{{geocodeArea:Ezinge}}->.searchArea;  
(
  way[highway~"unclassified||residential"](area.searchArea);
  >;
);
(._;rel(bn););
(._;rel(bw););
(._;rel(br););
out meta;

Equivalently, but specifying that Ezinge is a ‘woonplaats’ rather than any random entity:

[out:xml][timeout:90];
area[admin_level=10]["name"="Ezinge"][boundary=administrative]->.searchArea; 
(
  way[highway~"unclassified||residential"](area.searchArea);
  >;
);
(._;rel(bn););
(._;rel(bw););
(._;rel(br););
out meta;

Useful (for me) admin levels within (ref):

• NL:
  • 3: country (European part)
  • 4: provincie
  • 8: gemeente
  • 10: woonplaats
• DE:
  • 2: country
  • 4: Bundesland
  • 5: Regierungsbezirk
  • 6: Kreis / Landkreis / kreisfreie Stadt
  • 7: Verbandsgemeinde / Samtgemeinde o.ä.
  • 8: Gemeinde
  • 9: Bezirk / Gemeindeteil mit Selbstverwaltung
  • 10: … ohne Selbstverwaltung
• FR:
  • 3: France métropolitaine
  • 4: regions (7 new ones)
  • 6: départements
  • 8: communes

And a query for sparse downloads of things I’m normally interested in while surveying (won’t download “parent relations” of ways, so need to download those manually before splitting/merging ways):

[out:xml][timeout:25][bbox:{{bbox}}];
(
  node[highway];
  way[highway];
  node[amenity=bench];
  node[amenity=bicycle_parking];
  way[amenity=bicycle_parking);
  node[amenity=waste_basket];
  node[amenity=post_box];
  node[leisure=picnic_table];
  node[tourism=picnic_site];
  node[amenity=recycling];
);
// recurse down
( ._; >; )
out meta;

Human-readable output (not for import into JOSM)

Here’s a bit of Overpass goodness I learned while preparing changeset osm.org/changeset/79665597:

[out:csv(::id,"duration", "interval","name")];
area[admin_level=6]["name"="Pyrénées-Atlantiques"][boundary=administrative]->.searchArea; 
(
  //relation[duration](area.searchArea);
  relation[interval](area.searchArea);
);
out;

Finds all relations with tag ‘interval’ in département Pyrénées-Atlantiques (64); turns out, that’s for bus lines.

The interesting part, though, is the first line: instead of outputting everything and the kitchen sink as XML (for direct import into JOSM), I only output the specified information: relation ID (so I can load the relation into JOSM if needed), and the specified tags duration, interval, and name.

Another use case for this: usage statistics using regex. I found houses with addr:street tag starting in lower-case (“de Kalverweide”), inconsistent with the street name tagged as “De Kalverweide” (upper case). Lower-case street names seemed obviously wrong to me, so I optimistically changed the addresses in changeset 80142765 and alerted the ‘kadaster’ office. They replied quickly (!) and said I was wrong. The lower-case letter was right, and it’ll stay. That surprised me more than a little. Overpass to the rescue!

[out:csv (::id, "addr:street") ][timeout:125];
{{geocodeArea:Nederland}}->.searchArea;
(
  node["addr:street"~"^de .+"](area.searchArea);
  //node["addr:street"~"^De .+"](area.searchArea);
);
out;

The first query finds all nodes (typically: house addresses) with addr:street starting in “de “ in “Nederland”, the second query finds the upper-case variant. ‘^’ means: start of string, ‘.+’ means: one or more occurrence of “any letter” (so hypothetical streetnames ‘de ‘ of ‘De ‘ are not matched).

There are > 250,000 addresses starting in upper case De, much more than the ~35,000 addresses starting in lower case. The ratio is similar for ‘het ‘ vs. ‘Het ‘. Lower-case street names are exotic in this country but not unheard of. I still find them bizarre, but hey, OSM is about facts, not about my feelings about them. I’ve reverted my first changeset (80199834).

(Note: if I remember correctly, Overpass output can be restricted to just the number of occurrences, not the occurrences themselves. For large numbers of occurrences, that’s the way to go, otherwise output becomes unwieldy. I don’t recall how this works. Also, I prefer seeing the street names matched, so I can visually check that the regex does what it should be doing.)