c24. Synthetic Live Session Replay

Build a qmap experiment path for replaying a synthetic agent session in real time so we can work out playback, reveal, timing, and layout behavior before real session feeds exist.

Use one canonical 10-second scenario: the agent initializes, runs a research phase that touches 24 HTTP URLs, then shifts into an edit phase that uses 4 MCP servers and touches 6 files. This should feel live, not like a static fixture swap. Actors and flows should accumulate over time, phase changes should be legible, and the end state should remain readable.

Treat this as reusable infrastructure, not a one-off animation. Separate scenario authoring from timeline compilation and from playback state. The playback source should emit normal qmap session-shaped data so real semantic sessions or reduced runtime sessions can later plug into the same UI path without a rewrite.

Keep scope tight. Start with one deterministic scenario, one replay transport, and a small playback UI surface (play, pause, restart, scrub, optional 1x/2x). Do not solve every future session type in this cycle. End with a recommendation on whether the replay machinery should stay experiment-local or graduate into a narrow reusable abstraction.