VSL comparison · exploratory
Exploratory only — not eligible for public cold claims.
Declared builds: not staged
Page-host and preparation mode are recorded with every run.
Prefer the separate page host: adaudit-speed-test-55l.pages.dev
“Cold/warm” are first/later session-exposure phases, not verified cache temperatures. Fresh documents do not reset TLS/cache/coalescing or prior keepalive/beacon work. Media URLs, caches and player storage are not changed per run. Each report records the settings/embed mode, telemetry policy and build observed running in the page; declared builds are not assumed for runs without evidence. Production-parity fixtures use backend settings for both players and disable extra tester prewarming. Telemetry delivery and attribution remain unverified.
JS not started
Idle. No external assets until a run begins.
Exploratory · modes and first-frame methods separated
Completed runs only: median [min–max], per-metric measured n. Cold/warm are first/later session-exposure phases. rVFC and playing estimates never share a median. Missing data is unknown, never zero. Attempted/valid/failed/unmeasurable counts and partial reports remain visible.
| Arm / exposure / n / failures | First frame ms | Start / peak quality | Time to 720p ms | Stalls count / ms | Switches / Volume fallbacks |
|---|
Current / latest run
- Player arm
- Choose a run
- Running build (observed in page; declared hashes marked)
- Unknown
- Insert → first video frame callback
- Unmeasured
- Insert → first observed media request / response byte
- Unknown
- Where the time went (all insert-relative; observations, not causal proof)
- Unknown
- Insert → poster paint
- Unknown
- Start / peak quality / reach 720p
- Unknown
- Actual advancing playback
- Unmeasured
- Startup waiting / post-frame stalls
- Unmeasured
- Completed bytes before first frame
- Unknown
- Completed bytes so far
- Unknown
- Network information
- Not exposed
- Measurement / engine / Volume fallbacks
- Unknown
- Page host / post-insert resource hosts / overlap / prewarm coverage
- Unknown
Load, prewarm and idle are excluded. rVFC is callback observation, not exact optical display; playing fallback is an estimate. Media-extension matching is best-effort: extensionless/native-HLS requests and cross-origin video/timings may be invisible. Bytes are exposed lower bounds, not full traffic. No-overlap observations cannot prove neutrality.
Prewarm attempts (not proof equally warm)
Sanitized pre-insert activity (navigation-relative ms)
Quality switches
Waiting events
Sanitized post-insert resources
Fetch/XHR evidence (exposed headers only; no CDN HIT inference)
Request failures (safe raw evidence; categories are not confirmed causes)
Saved reports
Saved reports: 0 · export format aav-vsl-reports-v4