Explore a trajectory Dataset¶
pChronicle reads trajectory Datasets from local directories or S3, discovers
supported source formats, and exposes normalized runs, steps, and
tool_calls tables.
It is the data layer for captured runtime history and external trajectory files—not the runtime or scheduler. External files are pinned and normalized directly rather than converted into canonical runtime events.
Start with a known Dataset¶
From a Persisting source checkout:
ls shows the discovered Sources. The overview reports the shape of the
Dataset without requiring you to write SQL first.
Ask a specific question¶
pchronicle query examples/data/atif \
'SELECT source, COUNT(*) AS steps
FROM dataset.steps
GROUP BY source
ORDER BY source'
Queries are read-only. Format-specific data is normalized at the Dataset boundary so the same question can span ATIF, ACTF, OpenAI Messages, canonical events, and Storyline Sources where their semantics align.
What you completed¶
You discovered the logical Sources in one Dataset, created a Catalog Snapshot, ran a stable built-in summary, and queried a normalized relation. No Dataset content was modified.
Continue by task: