Skip to content

site: re-record the landing clips to end on the result - #346

Merged
vishr merged 1 commit into
mainfrom
site/footage-scroll-2
Oct 18, 2026
Merged

vishr merged 1 commit into
mainfrom
site/footage-scroll-2

Conversation

@vishr

@vishr vishr commented Oct 18, 2026 •

Copy link
Copy Markdown
Member

Summary

  • New landing clips recorded on the demo running v2026.10.15: the dashboard request now asks for headline stats and trends, p95 latency by operation, slowest dependencies and recent slow traces, and the resulting dashboards have nine panels, none empty. Clips are 20–21 s.
  • Recordings settle the conversation, return to the question and scroll through the evidence to the end of the answer; the dashboard recording scrolls through every panel and back to the top. Waits compress to 0.6 s through freezedetect while typing, scrolls and dwells stay real time. Typing runs at a quicker cadence so the longer request does not dominate the clip.
  • A take fails when the run shows a failed step, when a chat run has fewer than four evidence views or any empty view, or when the saved dashboard has fewer than six panels or any empty non-text panel.
  • Clips are content-addressed on assets.fanout.run (manifest site/src/data/landing-media.json), with new posters and landing copy matching the request.

The dark chat clip comes from an earlier take that shows "2 steps failed" for no-data guidance; a follow-up stops counting guidance as failed steps, after which that clip can be re-recorded.

Verification

  • node --test site/scripts/*.test.mjs (52), UI lint.
  • All eight clip URLs return 200 from assets.fanout.run.
  • Frame sheets reviewed for every clip.
  • just check in the pre-push hook.

The chat clip now waits for the conversation to settle, scrolls from the
question through the evidence to the end of the answer and holds there;
the dashboard clip trims the panel-loading interval, scrolls through
every panel and returns to the headline stats. The encoder also detects
near-static stretches with ffmpeg freezedetect and shortens them like
model and tool waits, never inside the typing, scroll or final hold.
Final holds are 3-4 s and clips run 21-24 s.

The landing page uses the new recordings from the OpenTelemetry Demo on
v2026.10.13: chat ends on checkout's Kafka "orders publish" finding, and
both clips show the question-anchored chat.

Claude-Session: https://claude.ai/code/session_018HfyEXHyGD1utmKF4CQtyQ
@vishr vishr changed the title site: re-record the landing clips to end on the result, without waits site: re-record the landing clips to end on the result Oct 18, 2026
@vishr
vishr force-pushed the site/footage-scroll-2 branch from 63f349d to 7dc5d5b Compare October 18, 2026 22:34
@vishr
vishr merged commit c121f69 into main Oct 18, 2026
16 checks passed
@vishr
vishr deleted the site/footage-scroll-2 branch October 18, 2026 22:56
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant