CLI Changelog

5.1.2 (2026-10-12)

Bug Fixes

Publish now reports a partition column missing from an input file by name, instead of stopping on a DuckDB Binder Error.

5.1.1 (2026-10-09)

Bug Fixes

Publish now stops before upload when a partition column has NULL values or more than 100 distinct values, naming the column to fix.
Publish now runs each input aggregation against your local files, stopping on rows the platform would reject and warning when a query returns no rows.

5.1.0 (2026-10-08)

Features

Publishing a model now uploads each file entry of an input’s sources separately, so an input with more than one data source no longer has to merge them into one file before publishing.
pd4castr test and a local run now read the source named by the source-binding axis’s default, so an input with several sources runs locally without a merged file.
pd4castr fetch now fetches that default source when it’s a fetcher, and names every other source it skips.
When a publish fails partway through uploading input files, the error now names the inputs already uploaded and says that re-running pd4castr publish uploads them again.

5.0.0 (2026-10-06)

Features

Project configs now declare an input’s or dataset’s storage location as storageLocation, and the CLI renames source to storageLocation in an existing config the next time it publishes, tests, or fetches with it.

Bug Fixes

Publishing now retries a static input or dataset upload when the network connection drops, instead of failing the publish on the first dropped connection.

4.14.2 (2026-09-24)

Bug Fixes

pd4castr test now reports the row number of the first output row with a missing or unparseable forecast_datetime or a missing output column, so a bad row deep in a large output can be found directly.

4.14.0 (2026-09-21)

Features

pd4castr publish and pd4castr test now run the model’s runDatetimeQuery against the local test inputs and print the instant it resolves to, so a query that returns no rows, the wrong column, or a value that is not a full timestamp fails on your terminal instead of as a failed run on the platform.

4.12.3 (2026-09-08)

Bug Fixes

publish now reports if a dataset is refused by the platform (i.e. missing a specified axis value)

4.12.2 (2026-09-08)

Bug Fixes

Publishing a model now allows up to thirty minutes for each dataset and static input file to upload, instead of failing on anything that takes more than ten seconds.

4.6.1 (2026-07-14)

Bug Fixes

Publishing an automatic-run model now reports that the run will start automatically once input files are processed, instead of incorrectly claiming the run did not trigger.

4.6.0 (2026-07-10)

Features

Publish now validates input-aggregation queries, blocking those whose SQL references an unknown column/table or omits a required output column (datetime, category_name, input_name, value).

4.5.0 (2026-07-09)

Features

Publish now validates sensitivity queries by dry-running them against local test inputs, blocking publishes whose SQL references an unknown column, table, or missing file.

4.4.0 (2026-07-06)

Features

Warns when publishing removes an access grant, listing which organisations and which grants were removed.

4.3.1 (2026-07-06)

Bug Fixes

publish --skip-trigger no longer prints a misleading “Triggering model run…“

4.3.0 (2026-07-01)

Features

Adds an optional key to input aggregations in the project config, with automatic backfill from the published model on next publish.

4.2.0 (2026-06-19)

Features

The existing model config’s description field has been migrated to a new summary field (max 500 characters), displayed in model info dialog in the pd4castr UI.
The model config’s description field has been repurposed to support long-form markdown which will be rendered in the model details page. Alternatively a descriptionFile field can be set which will inline a markdown file at publish time.
Model run view params can now be configured with an optional description (max 500 characters) via the model config file, shown as an info tooltip beside the param’s field label in the pd4castr UI.

4.1.0 (2026-06-11)

Features

Publishing a model now warns when a declared view would render an empty table, either because the query returns no rows or because it joins a dataset/input that the query drops as always-empty. Publishing still succeeds; the warning prints with the view and the table to check.

4.0.0 (2026-06-09)

Features

--skip-checks flag is removed. Model I/O tests and more robust pre-publish validation now occurs for all publishes.
--input-dir flag is removed. pd4castr fetch results now live at test_input/<key>.<targetFileFormat>.
Adds datasets[] and views[] to the model config: you can now declare parameterised SQL Model Run Views and their Datasets that will be made available in the pd4castr UI.
inputs[].file is now required on every static input. Existing configs will continue to work and will be migrated on next publish.
Renamed inputs[].inputSource to inputs[].source in the model config. Existing configs will continue to work and will be migrated on next publish.

Bug Fixes

Fixes pd4castr publish failing the pre-publish checks on Linux.

3.0.0 (2026-05-21)

Features

Internal network requests use the new API gateway. This is not a true breaking change, but 3.0.0 will be the minimum version required in the future when the previous paths are deprecated.

Bug Fixes

pd4castr publish now runs the same model output validation as pd4castr test, catching empty output, an invalid forecast_datetime, or a missing output variable before publishing.

2.3.0 (2026-05-08)

Features

Add support for the new annual time horizon.

2.2.0 (2026-05-05)

Features

pd4castr test validates the model’s output, now failing and providing feedback when forecast_datetime is an invalid format or if an output variable is missing.

2.1.0 (2026-04-29)

Features

login command now caches authentication when using --client-credentials, resulting in less network calls

2.0.0 (2026-04-21)

Features

Add a new top-level description field to .pd4castrrc.json; the value is displayed in the model information UI.