The shown lead drifts for a source slower than real time, and the drift is taken for a new stream #179

Closed
opened 2026-11-16 16:21:31 -05:00 by cmoriarty · 0 comments
Owner

What happens

In the live tests of 2026-11-16 (build ff90c98), the game page's "s ahead / behind" for shd247.world links (candyshop.st playlists) was wrong by 20 to 40 s:

  • Greece vs. Serbia: shown "5.2 s ahead", trusted, at 20:22 UTC. Two minutes later the join onto it was confirmed by the sound and found the stream 16.7 s short of what the channel had already sent, so the picture went back 43.4 s.
  • Germany vs. Netherlands: the same site's link read +15.8, −22.7, +16.8 and −28.7 s within six minutes.
  • The log said "its clock moved −34.3 s, taken for a new stream behind it" (20:18:56) and "−41.3 s" (20:21:31), and dropped the link's pairs each time. They were found again within two seconds at the same relation (−5060.547 s both times).

Joins go by the timestamp relation and landed where they should every time (checked against the broadcast's clock). Only the figure shown and the dropped pairs are wrong.

Why

  • A link's clock is the largest segment end timestamp − time first listed of the last 10 minutes (offsets.clock_of, CLOCK_SECONDS). That holds for a stream whose delay is steady.
  • shd247.world's sources list media slower than real time. The game page said "Pace 0.90×" and "Source slower than real time". So their delay grows by about 6 s a minute, and the 10-minute maximum keeps a reading from minutes ago: the link looks up to a minute further ahead than it is.
  • _listed measures each reading against that maximum. Once the drift passes 30 s (RESTART_SECONDS), it is taken for another stream behind the link and the link's pairs are dropped.

Done when

  • A link's clock follows a steady drift: earlier readings are carried forward by the link's own measured pace before the largest is taken, so a 0.9× source reads its delay as it is now.
  • Only a sudden jump between readings counts as another stream behind the link; a steady drift drops no pairs.
  • Tests: a 0.9× source keeps its pairs and shows its lead within a second or two of the truth; a sudden jump still drops them.
## What happens In the live tests of 2026-11-16 (build ff90c98), the game page's "s ahead / behind" for shd247.world links (candyshop.st playlists) was wrong by 20 to 40 s: - Greece vs. Serbia: shown "5.2 s ahead", trusted, at 20:22 UTC. Two minutes later the join onto it was confirmed by the sound and found the stream 16.7 s short of what the channel had already sent, so the picture went back 43.4 s. - Germany vs. Netherlands: the same site's link read +15.8, −22.7, +16.8 and −28.7 s within six minutes. - The log said "its clock moved −34.3 s, taken for a new stream behind it" (20:18:56) and "−41.3 s" (20:21:31), and dropped the link's pairs each time. They were found again within two seconds at the same relation (−5060.547 s both times). Joins go by the timestamp relation and landed where they should every time (checked against the broadcast's clock). Only the figure shown and the dropped pairs are wrong. ## Why - A link's clock is the largest `segment end timestamp − time first listed` of the last 10 minutes (`offsets.clock_of`, `CLOCK_SECONDS`). That holds for a stream whose delay is steady. - shd247.world's sources list media slower than real time. The game page said "Pace 0.90×" and "Source slower than real time". So their delay grows by about 6 s a minute, and the 10-minute maximum keeps a reading from minutes ago: the link looks up to a minute further ahead than it is. - `_listed` measures each reading against that maximum. Once the drift passes 30 s (`RESTART_SECONDS`), it is taken for another stream behind the link and the link's pairs are dropped. ## Done when - A link's clock follows a steady drift: earlier readings are carried forward by the link's own measured pace before the largest is taken, so a 0.9× source reads its delay as it is now. - Only a sudden jump between readings counts as another stream behind the link; a steady drift drops no pairs. - Tests: a 0.9× source keeps its pairs and shows its lead within a second or two of the truth; a sudden jump still drops them.
Sign in to join this conversation.
No labels
No project
No assignees
1 participant
Notifications
Due date
The due date is invalid or out of range. Please use the format "yyyy-mm-dd".

No due date set.

Dependencies

No dependencies set

Reference
cmoriarty/sports-breaker#179
No description provided.