• The focus is still on "web video."
• The primary delivery model is unicast IP over the Internet. We can make assumptions based on the behavior of IP networks rather than radio broadcast or optical media, for example.
• Video-on-demand, video conferencing, screen-casting, multistreaming all primary use cases.
• Want a useful, consistently deployed feature set, not tons of profiles and extensions.
• Strong support for parallel encode/decode within each frame.

Specific deficiencies in VP8 to address:

• VP8 has *almost* enough framing information to parse it as an elementary stream. It'd help some applications (including support) if it could be parsed without an extra container.
• Lack of 4:4:4 or equivalent for screen-casting
• Lack of alpha support
• Inflexible/expensive segmentation

• Bytestream vs bitstream? Literals vs bool-coded partitions? Which parts of the stream should a semi-smart sever be able to cheaply parse/rewrite?
• Ability to insert "informative" blocks into the bitstream? things a dumb decoder or MITM could always ignore but interested parties could parse and act on?
•