Robot Data and Event Streams

Replayable robot data

Robot data rarely arrives in one neat loop. Camera frames, commands, plans, observations, and status updates arrive at different times. Golden data examples show how to keep those streams replayable and exportable while preserving event time and source metadata.

Build a bufferpixi run demo-robotics-data-eventstream

Expected result: timestamped events are appended, queried by window, and printed with source metadata intact.

Join streamspixi run demo-robotics-data-join

Expected result: observation and command events align by event time instead of callback arrival order.

RecordEvery value keeps event time

Logs and datasets can explain when the robot saw, planned, commanded, or reported a value.

ReplayQueries are deterministic

Windowed reads depend on recorded timestamps, not live scheduling jitter.

ExportLineage survives conversion

Dataset rows keep source and conversion metadata so downstream errors remain traceable.

Output To Recognize

Deterministic merged order:
Processing-time profile:
Exact join:
Latest-before join (max_delta=50):
Window join (+/-20):

Those labels are the point of the examples: a later user should be able to tell whether timing alignment, windowing, and lineage survived without reading the implementation first.

What Each Piece Is For

Concept	Use it for	Check
Event	One timestamped value with source metadata.	Value, event time, source, and lineage are explicit.
EventBuffer	A bounded event history.	Window queries are deterministic.
Event-time join	Aligning two streams.	The join uses recorded timestamps, not callback timing.
Window	Aggregating over a temporal slice.	The slice is explicit and replayable.
Lineage	Tracking conversion and source details.	Exported data still explains where it came from.

Why This Belongs In Golden

Core Retriever owns runtime event mechanics. Golden shows the applied robot-data profile: how a camera observation, command, state estimate, or dataset record should carry enough timing and provenance to be debugged later.

Use this path when you need reproducible examples for record/replay, dataset export, or cross-runtime comparison.