Live target-size compression cases — 2026-10-23 Execution: an authenticated public video-size-reducer.com session in Chrome on macOS, with browser automation assisting the product testing. These are first-party measurements, not an independent review. Source videos are public licensed material or original fictional test material. No private footage was used. Method 1. Select the source in the public browser uploader. Use target-size mode, decimal MB, H.264 MP4 and Fast. Gameplay and screen: Original resolution; prepared five-minute Sintel excerpt: 720p selected. The actual encoded dimensions and duration are in measurements.json. 2. Submit one compression task for each source. Download the resulting MP4 using the website's authorized download control. 3. Measure the actual downloaded byte count, SHA-256 and media streams with local FFprobe. Run a full decode with FFmpeg -xerror -i RESULT -f null -. Source preparation and verification use FFmpeg/FFprobe 8.1.2; the server encoder version was not independently recorded. 4. Extract matching source/result frames at game 30s, screen 24s, Sintel 240s: ffmpeg -ss SECOND -i FILE -frames:v 1 -q:v 3 FRAME.jpg. No enhancement, cropping or upscaling is applied. JPEG frames cannot prove motion or audio quality. Open either image to inspect its native dimensions. 5. Publish only completed, downloaded, byte-checked and fully decoded results. No target results are inferred from bitrate calculations or the older local Discord tests. Sources and reuse Gameplay: SuperTuxKart authors; recording uploaded by PantheraLeo1359531. Wikimedia 1080p VP9 derivative, 52,046,748 bytes, 70.834 seconds. Source and attribution: https://commons.wikimedia.org/wiki/File:Benchmark_SuperTux_Kart_Base_Profile_2025.07.04_-_10.19.10.02.webm License: CC BY-SA 4.0, https://creativecommons.org/licenses/by-sa/4.0/ The displayed gameplay frames and recompressed derivatives remain CC BY-SA 4.0. Screen: original 60-second recording of our fictional test webpage, prepared for the September 16 case study. No real personal records. Original MP4: 3,316,953 bytes. To obtain a source larger than 10MB without padding or increasing bitrate, repeat it four times by stream copy: ffmpeg -stream_loop 3 -i screen-source.mp4 -map 0:v -c copy screen-loop-240s.mp4 Prepared source: 13,265,235 bytes, 240 seconds, 1920x1080, no audio. This is a prepared fixture, not a newly recorded four-minute session. Original source: https://assets.video-size-reducer.com/case-studies/2026-09-16/screen-source.mp4 Original recording method: /case-studies/2026-09-16/methodology.txt Long: Sintel (2010), © Blender Foundation | www.sintel.org. A continuous 299.419352-second H.264 MP4 excerpt, 203,756,551 bytes, 1920x818. Prepared without re-encoding from the September 16 six-minute source (231,348,205 bytes), using a 43-second seek and 298-second stream-copy interval; keyframe preroll and timestamp normalization make the final container duration 299.419352 seconds. This shorter excerpt fits the existing account credit reserve. It is not the full film or a camera original. Preparation command: ffmpeg -ss 43 -i long-source.mp4 -t 298 -map 0:v:0 -map 0:a:0 -c copy -avoid_negative_ts make_zero long-299s.mp4 Preparation parent source: https://assets.video-size-reducer.com/case-studies/2026-09-16/long-source.mp4 Original film attribution: https://commons.wikimedia.org/wiki/File:Sintel_movie_4K.webm License: CC BY 3.0, https://creativecommons.org/licenses/by/3.0/ Interpretation MB = 1,000,000 bytes; MiB = 1,048,576 bytes. Targets are ceilings, not exact resulting sizes. Verify actual result bytes against the destination limit. A successful byte ceiling and full decode do not prove visual quality, message delivery or platform acceptance. No Discord post or email transmission was performed. These cases do not validate an 8MB target or predict results for other sources, settings, hardware or load. No independent server telemetry or precise upload/queue/encoding/end-to-end timings were captured in this batch. Such unavailable values are null, not zero. Credits used are read from each completed task’s settings panel; they are not inferred from the pre-upload estimate or account-balance changes. They do not imply a fixed number of free videos. The old September 23 MiB cases and September 30 local FFmpeg comparisons retain their separate methods and dates. Local private task receipts and downloaded outputs are retained in the operator's ignored evidence folder. Public screenshots contain only matching source/result frames, not account details or authenticated download URLs. SHA-256 values identify the exact files measured.