Repository navigation
Fix unrecorded bitrates being read as bad ones - #78
Merged
Merged
Conversation
Asking for voices in Dhaka lands on one dead station. Measured causes: Bangladesh has 26 directory entries against 23 brands Wikidata knows, 12 of those entries carry no tags at all, and both Jago FM rows share one dead stream. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
The directory is community edited, so a station is often filed twice with one stream between the rows. Both Jago FM entries in Bangladesh point at the same stream and it is off the air, so asking for voices in Dhaka spent two of its attempts on the same silence. Measured: Bangladesh 26 entries to 24, popular top 500 to 482. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Every mock station shared one stream URL, so collapsing duplicate streams correctly reduced them to a single entry and the end-to-end expectations no longer held. Real stations have their own streams. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
The directory writes 0 when it has no bitrate on file, so the quality floor dropped the station outright. Of the live stations measured, 43% in Bangladesh, 45% in Egypt and 50% in Nigeria carry no bitrate, and it fell hardest on the stations the voices mode exists to find. Bangladesh offered three voice stations out of six, two of them Quran recitations and the third off the air. Radio Vivid Voice and Spice FM are tagged talk and were dropped only for missing metadata. Unknown now passes and the stream probe decides. A bitrate that is on file and below the floor is still refused. Measured voices pool: BD 3 to 6, EG 2 to 9, NG 5 to 18, KE 4 to 10. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
raj-khan
added a commit
that referenced
this pull request
Sep 19, 2026
* Add backlog for live and local radio coverage Asking for voices in Dhaka lands on one dead station. Measured causes: Bangladesh has 26 directory entries against 23 brands Wikidata knows, 12 of those entries carry no tags at all, and both Jago FM rows share one dead stream. * Collapse stations that share a stream URL The directory is community edited, so a station is often filed twice with one stream between the rows. Both Jago FM entries in Bangladesh point at the same stream and it is off the air, so asking for voices in Dhaka spent two of its attempts on the same silence. Measured: Bangladesh 26 entries to 24, popular top 500 to 482. * Format dedupe files with Prettier * Give each mock station its own stream in the e2e fixtures Every mock station shared one stream URL, so collapsing duplicate streams correctly reduced them to a single entry and the end-to-end expectations no longer held. Real stations have their own streams. * Fix unrecorded bitrates being read as bad ones The directory writes 0 when it has no bitrate on file, so the quality floor dropped the station outright. Of the live stations measured, 43% in Bangladesh, 45% in Egypt and 50% in Nigeria carry no bitrate, and it fell hardest on the stations the voices mode exists to find. Bangladesh offered three voice stations out of six, two of them Quran recitations and the third off the air. Radio Vivid Voice and Spice FM are tagged talk and were dropped only for missing metadata. Unknown now passes and the stream probe decides. A bitrate that is on file and below the floor is still refused. Measured voices pool: BD 3 to 6, EG 2 to 9, NG 5 to 18, KE 4 to 10.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Why
This is the main reason you could not find anyone talking in Dhaka.
Radio Browser writes
0when it has no bitrate on file, which normalizing turns intoundefined.meetsQualityFloorread that as 0 kbps and dropped the station. Unknown was being judged as bad.The gap is not small, and it falls hardest on exactly the stations the voices mode exists to find:
Bangladesh has six speech stations and was offering three: two Quran recitation channels and Jago FM, whose stream is off the air. Radio Vivid Voice and Spice FM are both tagged
talkoutright and were discarded only because nobody had recorded a bitrate. Nigeria lost thirteen of its eighteen the same way.What changed
An unrecorded bitrate now passes the floor, and the stream probe decides. The probe already rejects anything that does not answer, which is the thing a listener actually cares about. A bitrate that is on file and genuinely below the floor is still refused, so the BBC World Service at 56 kbps still gets in and a 32 kbps music stream still does not.
Measured
Voices pool per country, before to after:
Music pools roughly double in the same places (India 37 to 83, Nigeria 25 to 43).
Dhaka now offers six voice stations, and the three added are the ones tagged for speech all along: Radio Vivid Voice, Spice FM and Universal Health Radio.
A deliberate omission
The task asked for known-good bitrates to be preferred over unknown ones in ordering. I did not partition the candidate pool that way. Doing so would bury the unknown stations behind the same few known ones, which in Bangladesh means the probe would keep landing on the two Quran channels and the dead Jago FM: the exact complaint this fixes. Quality preference is applied where it costs no variety, in dedupe's
betterEntry, which keeps the row that has a recorded bitrate when two entries describe one stream.Checks
npm run lint,npm run format:check,npm run typecheck,npm test(435 passed) andnpm run buildpass, and the full Playwright suite (34 tests) passes locally.