Point this at your own AIOStreams addon. Every stream it offers you is opened, read at the byte level, and returned only if its container really holds an Arabic subtitle track.
Release names, tracker tags and NFO files are ignored completely. A stream that cannot be read is left out rather than shown with a question mark.
Copy it from your AIOStreams page — it ends in /manifest.json.
What counts as proof
Accepted
A subtitle track whose container language field is Arabic — ara, arz, ary, ar-EG and the rest of the family.
Accepted
A subtitle track with no declared language whose muxer-written track name is Arabic. Marked as such in the result.
Rejected
Anything read from a filename, a release group, a tracker label or an external subtitle listing.
Left out
Files that could not be read — a dead link, an unreadable container, a host that refuses range requests.
How it stays fast
It reads kilobytes, not gigabytes
The track table of a Matroska file sits near the front, and the file itself says where. A typical verdict costs one range request of about 48 KB against a 20 GB remux.
A verdict is remembered per file, not per title
Files are identified by content, so the same episode reached through a season pack, another tracker or another title is already answered.
Results arrive during the first open
Verified streams are returned as soon as enough of them exist. Everything still being read finishes in the background, so there is nothing to reopen.