# replay — Stream dataset exports over WebSocket

`s6 replay` replays a `StructuredDataset` directory and exposes its
`context["export"]` payloads over a lightweight WebSocket service. Unlike
`s6 track`, it does not run inference or mutate the dataset frames; it simply
relays what is already recorded.

## Usage

```bash
# Replay a dataset forever on the default service endpoint
s6 replay ./datasets/session_001 --repeat

# Bind to a different host/port and slow playback down
s6 replay ./datasets/session_001 --host 127.0.0.1 --port 9000 --hz 10

# Disable replay throttling backpressure
s6 replay ./datasets/session_001 --max-queue 0
```

## Behavior

- Validates that the dataset directory exists and contains `data.jsonl`.
- Starts a background producer process that reads the dataset through
  `StructuredDataset` and enqueues frame records.
- Serves WebSocket clients on `/ws/telemetry`.
- Also exposes `/queue_size` for a simple backlog health check.
- Sends JSON payloads with:
  - `export`: the recorded `context["export"]` object, or `{}` when absent
  - `stats.queue_size`: the current queue backlog when available
- `--repeat` loops the dataset indefinitely.
- `--hz` controls playback rate in frames per second; values less than or equal
  to `0` disable the sleep throttle.
- `--max-queue` applies backpressure to the producer; `0` disables that limit.

## Notes

- This command is useful when a downstream consumer only needs recorded export
  telemetry and not the full pipeline runtime.
- The default bind address is `0.0.0.0:8001`.