For example, orthographic conversion modes between zab_Simp and zab_Phon are detected as translation pair modes:

$ curl 'https://beta.apertium.org:/apy/listPairs'
... {"sourceLanguage": "zab_Phon", "targetLanguage": "zab_Simp"}, {"sourceLanguage": "zab_Simp", "targetLanguage": "zab_Phon"}, ...

However, conversion between them doesn't work via APy:

$ curl 'https://beta.apertium.org:/apy/listP/apy/translate?langpair=zab_Simp|zab_Phon&q=dizh'
{"status": "error", "code": 503, "message": "Service Unavailable", "explanation": "internal error"}

This does work in the command-line version of the mode:

$ echo "dizh" | apertium -f none -d . zab_Simp-zab_Phon
dìiʼzh

Some modes between different locales may legitimately be more than just orthography; i.e., they should be thought of as full translation pairs (e.g., between American and British English). So I'm wondering whether, instead of fixing this, it would make sense to name the orthography-only modes -ortho or something and have a separate detection process for them, and serve them separately (and eventually have a different interface for them in html-tools).