A game is Ready after a 10-second check: watch its links and grade how they keep pace #94

Closed
opened 2026-11-07 20:04:41 -05:00 by cmoriarty · 3 comments
Owner

What happens

A link is Ready after verify(), which:

  • reads the manifest;
  • downloads one segment and times it;
  • reads the manifest again one target duration later.

That is about 10 seconds of evidence. After that, every 150 s, the re-check reads the manifest and 64 KB of a segment. Neither check asks whether the source keeps up with real time over minutes. A stream nobody could watch passes both, and the viewer is the first to find out.

Ole Miss vs. Georgia, 2026-11-07, about 7:50 pm ET. I polled the playing link's JSON playlist once a second for 63 s, from outside the app:

Measure Seen
Video published 13 new segments: 50 s of video in 63 s, 0.79× real time
Gaps between new segments 1.6–6.3 s (target duration 4 s)
Download per segment 0.1–1.3 s: the CDN was fine, the source falls behind
  • The channel fell behind: its splice offset grew from +0.1 s to +58 s in 13 minutes, made up with filler.
  • It was the only stream: all 10 sites the channel landed on that evening carried this one stream (#91).
  • The game showed Ready the whole time, and every standby verified.

The idea

Watch the links the app would play, continuously and cheaply, and let readiness say how they have played.

  1. What is watched: the kept link and the standbys of every followed live game (the watch list), not the whole slate.
  2. How:
    • Poll each playlist about every 2 s, as a player does, and note when each new segment is listed and how long it is.
    • Every 30–60 s, download the newest segment the way a channel does (upstream.fetch_concurrency): time it, check its bytes (media_shape) and count its frames (cadence).
    • The 150 s re-check folds into this.
  3. The score, over a sliding 3 minutes:
    • pace: seconds of video listed per second of wall clock (tonight 0.79);
    • longest gap between new segments (tonight 6.3 s);
    • frames: the share of sampled segments short of their frame rate;
    • download headroom: segment length over download time;
    • errors and refusals.
  4. What Ready means:
    • A link that verifies is Ready at once, as now, so the guide doesn't wait.
    • Its grade starts as Unproven and becomes Steady or Choppy after 2 minutes of watching.
    • The card, the Watchlist page and the guide show the grade; #97 says how.
    • A link whose playlist stops moving loses Ready, as a failed re-check does today.
  5. Cost:
    • A few KB per poll, plus one segment of about 1.5 MB a minute: roughly 25–50 KB/s per link.
    • With 3 links for each of 10 followed live games, about 1–1.5 MB/s.
    • No site reads: only CDN traffic, like a player's.

To settle in the design:

  • the pace and gap thresholds between Steady and Choppy (for example: Steady at pace ≥ 0.97, with no gap longer than the buffer covers);
  • the poll interval for short-window playlists (tonight's lists 3 segments and expires 12 s after it is published).

Done when

  • In the unit lane: a scripted source at 0.8× real time is graded Choppy, and a steady source with one 6 s burst is graded Steady.
  • On a live game: a followed game's card shows its grade before anyone tunes in.
## What happens A link is Ready after `verify()`, which: - reads the manifest; - downloads one segment and times it; - reads the manifest again one target duration later. That is about 10 seconds of evidence. After that, every 150 s, the re-check reads the manifest and 64 KB of a segment. Neither check asks whether the source keeps up with real time over minutes. A stream nobody could watch passes both, and the viewer is the first to find out. **Ole Miss vs. Georgia, 2026-11-07, about 7:50 pm ET.** I polled the playing link's JSON playlist once a second for 63 s, from outside the app: | Measure | Seen | |---|---| | Video published | 13 new segments: 50 s of video in 63 s, **0.79× real time** | | Gaps between new segments | 1.6–6.3 s (target duration 4 s) | | Download per segment | 0.1–1.3 s: the CDN was fine, the source falls behind | - **The channel fell behind:** its splice offset grew from +0.1 s to +58 s in 13 minutes, made up with filler. - **It was the only stream:** all 10 sites the channel landed on that evening carried this one stream (#91). - **The game showed Ready the whole time,** and every standby verified. ## The idea Watch the links the app would play, continuously and cheaply, and let readiness say how they have played. 1. **What is watched:** the kept link and the standbys of every followed live game (the watch list), not the whole slate. 2. **How:** - Poll each playlist about every 2 s, as a player does, and note when each new segment is listed and how long it is. - Every 30–60 s, download the newest segment the way a channel does (`upstream.fetch_concurrency`): time it, check its bytes (`media_shape`) and count its frames (`cadence`). - The 150 s re-check folds into this. 3. **The score, over a sliding 3 minutes:** - **pace:** seconds of video listed per second of wall clock (tonight 0.79); - **longest gap** between new segments (tonight 6.3 s); - **frames:** the share of sampled segments short of their frame rate; - **download headroom:** segment length over download time; - **errors and refusals.** 4. **What Ready means:** - A link that verifies is Ready at once, as now, so the guide doesn't wait. - Its grade starts as Unproven and becomes Steady or Choppy after 2 minutes of watching. - The card, the Watchlist page and the guide show the grade; #97 says how. - A link whose playlist stops moving loses Ready, as a failed re-check does today. 5. **Cost:** - A few KB per poll, plus one segment of about 1.5 MB a minute: roughly 25–50 KB/s per link. - With 3 links for each of 10 followed live games, about 1–1.5 MB/s. - No site reads: only CDN traffic, like a player's. **To settle in the design:** - the pace and gap thresholds between Steady and Choppy (for example: Steady at pace ≥ 0.97, with no gap longer than the buffer covers); - the poll interval for short-window playlists (tonight's lists 3 segments and expires 12 s after it is published). ## Done when - **In the unit lane:** a scripted source at 0.8× real time is graded Choppy, and a steady source with one 6 s burst is graded Steady. - **On a live game:** a followed game's card shows its grade before anyone tunes in.
Author
Owner

Built in PR #99 (merged as 732ab0a), deployed 2026-11-08 04:19Z (container 52cd495f611a); the OpenSpec change is archived in 038f7a9. This stays open until it is seen working on a live followed game: a grade before the tune, a Warm background channel, and a Choppy stream that stays put.

Built in PR #99 (merged as 732ab0a), deployed 2026-11-08 04:19Z (container 52cd495f611a); the OpenSpec change is archived in 038f7a9. This stays open until it is seen working on a live followed game: a grade before the tune, a Warm background channel, and a Choppy stream that stays put.
Author
Owner

Live test, UTEP vs. Hawai'i, 2026-11-07 23:22–23:55 ET (first followed game after the deploy). Seen working: the kept link and both standbys were watched and graded on the readiness row and the channel: Unproven at first, then Steady after 2 minutes (givemestreams.cfd, pace 0.99, about 11 s of buffer), then Choppy (sportsframe.site, pace 0.83, about 34 s) and back to Steady (pace 1.0) as the source recovered. The game's grade followed its best kept stream.

Live test, UTEP vs. Hawai'i, 2026-11-07 23:22–23:55 ET (first followed game after the deploy). Seen working: the kept link and both standbys were watched and graded on the readiness row and the channel: Unproven at first, then Steady after 2 minutes (givemestreams.cfd, pace 0.99, about 11 s of buffer), then Choppy (sportsframe.site, pace 0.83, about 34 s) and back to Steady (pace 1.0) as the source recovered. The game's grade followed its best kept stream.
Author
Owner

Closed at the owner's word (2026-11-08): the fix is merged to develop in PR #99 (merge 732ab0a) and deployed. A new issue will be opened if a problem shows up.

Closed at the owner's word (2026-11-08): the fix is merged to develop in PR #99 (merge 732ab0a) and deployed. A new issue will be opened if a problem shows up.
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#94
No description provided.