Pega Export Tutorial¶
This tutorial explains the Value Stream path for Pega-style export files. The exact reader settings depend on the export format, but the workflow is the same: configure the source, validate, probe, run, and inspect reports.
Configure the Source¶
Pega Dataset Export-style workspaces usually define a source in
catalog/pipelines.yaml with:
- Reader kind
pega_ds_exportfor zip-based Pega exports, orparquetwhen exports have already been converted. - A
file_patternthat matches the export files. - A
group_by_filenameregex that groups files into idempotent chunks. - Timestamp parsing transforms for fields such as
OutcomeTimeandDecisionTime. - Calendar derivation for report grains such as Day, Month, Quarter, Year, and Summary.
- Outcome filters and defaults for fields that may be absent in some exports.
See examples/demo/catalog/pipelines.yaml for a runnable Parquet-based
Pega-style configuration example (this is the only example workspace checked
into the repository).
In AI Configuration Studio, use Workspace sample when the Pega archive is
already under the active workspace's data/ folder. Select the archive, click
Use Workspace Sample, review the generated draft, and apply it only after
the Studio validation gate passes.
Validate¶
uv run valuestream validate examples/demo
Probe¶
uv run valuestream probe examples/demo ih --limit 5
Confirm that expected fields such as channel, issue, group, outcome, model control group, propensity, and timestamp columns are present after transforms.
Run¶
uv run valuestream run examples/demo ih
Use --force after changing source transforms, processor dimensions, or outcome
classification:
uv run valuestream run examples/demo ih --force
Review Reports¶
Start the UI:
uv run valuestream serve examples/demo --port 8501 --headless
Use:
- Data Load to confirm discovered chunks and source run status.
- Reports to review engagement, propensity, funnel, experiment, and response time pages.
- Pipelines / Ops to inspect runs and chunks.
- Catalog to review generated sources, processors, metrics, and dashboards.
Generate Synthetic Pega-Shaped Data¶
Use this when you need a realistic test dataset shaped like a Pega export but cannot use production data. It derives the schema from a real export sample and generates configurable volumes:
uv run valuestream generate-pega-dummy \
--source path/to/source.json \
--output-dir /tmp/value-stream-demo-data \
--start-date 2026-01-01 \
--days 7
--source accepts a Pega JSON/NDJSON export or a zip/gzip/tar.gz archive of
JSON records. See the CLI reference
for volume, outcome-rate, and output options.
Common Pega Checks¶
| Check | Why it matters |
|---|---|
| Timestamp parsing succeeds | Calendar grains and freshness depend on valid time columns |
| Outcome values are normalized | Engagement and conversion metrics depend on positive and negative definitions |
| Natural keys are stable | Deduplication depends on consistent identifiers |
| Group-by dimensions are approved | Report filters only work for persisted dimensions |
| Propensity fields are numeric | Model quality metrics require numeric score distributions |