Measuring: a stream whose timestamps don't keep pace with its sound is found again every few minutes and takes a third of the comparisons #186

Closed
opened 2026-11-17 01:14:42 -05:00 by cmoriarty · 1 comment
Owner

What happens

Ravens vs. Chargers, 2026-11-17 01:17 to 04:00 UTC (build 9542642). The link is topsurge.st, stream hls.lshls.st/secure_hls.php?path=*/index.m3u8&v=…; its v= value changed every 25 minutes.

  • No relation held. Each match found the same sound at a new relation, further back by about the time since the last one.
    • From edgestream.pro, then playing, between 01:19:29 and 01:36:37: −18,148.3, −18,264.1, −18,276.6, −18,463.0, −18,671.4, −18,915.3, −19,153.1, −19,203.2 and −19,243.2 s.
    • That is 1,095 s further back in 17 minutes. It slid between matches with no drop logged between them too: −19,203.2 then −19,243.2 s, at 01:35:52 and 01:36:37.
    • The other links measured against the same playing stream (wildgames.bz, my1click.top, ccrackstreams.top) were trusted and held steady.
  • It was "taken for a new stream behind it" 11 times. The jumps were −114.9 to −282.8 s, then −615.1 s and −6,143.5 s. The five-minute summaries counted 22 clock restarts.
  • It was never trusted. The game page showed it as found, at −36 to −67 s, all evening.
  • It took 211 of the night's 596 comparisons (35%), and 133 of those matched. Comparisons run one at a time, so the other links' measuring waited behind it.

Nothing went wrong on screen:

  • no swap went to it;
  • a relation heard only once is used for a join only if the join's own sound check confirms it (aim.NOT_FOUND).

By contrast, forprivateuse.sbs's 6 drops that night were real restarts. Between them its relation held to the millisecond (−8,163.740 and −8,163.742 s at 01:51:12 and 01:51:59), and it was found again and trusted within minutes.

Why

As far as the log shows, this stream's timestamps hardly move while its sound plays on. So the timestamp relation for the same sound slides by the time that passes.

Each match makes a new "found" relation, and the next sample disagrees with it. _listed takes the slide for a new stream behind the link and drops its pairs.

The link is never ruled out, so it keeps its full share of comparisons.

Done when

  • A link whose matches disagree by about the time between them is marked as not measurable:
    • its state is none, with a reason such as "its timestamps do not follow its sound";
    • no seconds figure is shown for it;
    • its comparisons slow to an occasional re-test.
  • A stream that really restarts (the forprivateuse.sbs pattern) is still found again and trusted.
  • Tests:
    • A fake stream whose timestamps stand still while its sound advances is marked within a few comparisons and stops taking more.
    • A stream with real restarts still reaches trusted.
## What happens Ravens vs. Chargers, 2026-11-17 01:17 to 04:00 UTC (build 9542642). The link is topsurge.st, stream `hls.lshls.st/secure_hls.php?path=*/index.m3u8&v=…`; its `v=` value changed every 25 minutes. - **No relation held.** Each match found the same sound at a new relation, further back by about the time since the last one. - From edgestream.pro, then playing, between 01:19:29 and 01:36:37: −18,148.3, −18,264.1, −18,276.6, −18,463.0, −18,671.4, −18,915.3, −19,153.1, −19,203.2 and −19,243.2 s. - That is 1,095 s further back in 17 minutes. It slid between matches with no drop logged between them too: −19,203.2 then −19,243.2 s, at 01:35:52 and 01:36:37. - The other links measured against the same playing stream (wildgames.bz, my1click.top, ccrackstreams.top) were trusted and held steady. - **It was "taken for a new stream behind it" 11 times.** The jumps were −114.9 to −282.8 s, then −615.1 s and −6,143.5 s. The five-minute summaries counted 22 clock restarts. - **It was never trusted.** The game page showed it as found, at −36 to −67 s, all evening. - **It took 211 of the night's 596 comparisons (35%), and 133 of those matched.** Comparisons run one at a time, so the other links' measuring waited behind it. Nothing went wrong on screen: - no swap went to it; - a relation heard only once is used for a join only if the join's own sound check confirms it (`aim.NOT_FOUND`). By contrast, forprivateuse.sbs's 6 drops that night were real restarts. Between them its relation held to the millisecond (−8,163.740 and −8,163.742 s at 01:51:12 and 01:51:59), and it was found again and trusted within minutes. ## Why As far as the log shows, this stream's timestamps hardly move while its sound plays on. So the timestamp relation for the same sound slides by the time that passes. Each match makes a new "found" relation, and the next sample disagrees with it. `_listed` takes the slide for a new stream behind the link and drops its pairs. The link is never ruled out, so it keeps its full share of comparisons. ## Done when - A link whose matches disagree by about the time between them is marked as not measurable: - its state is `none`, with a reason such as "its timestamps do not follow its sound"; - no seconds figure is shown for it; - its comparisons slow to an occasional re-test. - A stream that really restarts (the forprivateuse.sbs pattern) is still found again and trusted. - Tests: - A fake stream whose timestamps stand still while its sound advances is marked within a few comparisons and stops taking more. - A stream with real restarts still reaches trusted.
Author
Owner

A correction to the description above, from the full log. The timestamps do move; the relation is what keeps moving.

  • It was mostly "found again", not dropped. The relation was replaced at a new place ("found again") 109 times that night. The drops ("taken for a new stream behind it") were the minority, and the log names only the first in each five-minute span: 22 were counted, 11 named.
  • After 02:00 the movement was steady: about 5 s at each sample 46 s apart, +3.9 to +6.3 s each time. The stream's sound runs about 10% slower than its timestamps say, so no relation holds for more than a moment.
  • The figure shown was not to be trusted either. In a test of this case, main showed a link 46.8 s behind while it was 4 s behind.

The fix is in #188. A pair whose relation moves at three matches in a row reads "unsteady":

  • it is compared every three minutes;
  • it shows no figure;
  • a join onto it is always checked by its sound.
A correction to the description above, from the full log. The timestamps do move; the relation is what keeps moving. - **It was mostly "found again", not dropped.** The relation was replaced at a new place ("found again") 109 times that night. The drops ("taken for a new stream behind it") were the minority, and the log names only the first in each five-minute span: 22 were counted, 11 named. - **After 02:00 the movement was steady:** about 5 s at each sample 46 s apart, +3.9 to +6.3 s each time. The stream's sound runs about 10% slower than its timestamps say, so no relation holds for more than a moment. - **The figure shown was not to be trusted either.** In a test of this case, main showed a link 46.8 s behind while it was 4 s behind. The fix is in #188. A pair whose relation moves at three matches in a row reads "unsteady": - it is compared every three minutes; - it shows no figure; - a join onto it is always checked by its sound.
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#186
No description provided.