0.18.0

Warning

Breaking change in Grade C discovery dataset uploads: each dataset's indicators now go on a sheet of their own, named with the dataset's ID and holding an Indicator column and a Units of measure column. The Excluded regional values field is removed. A workbook that still has Excluded regional values, Indicators coverage on the Datasets sheet, or Relevant indicators not present is rejected in full, and the error names the column. A dataset ID must also be a valid sheet name: at most 31 characters, none of : \ / ? * [ ], not starting or ending with ', not History, Datasets or Instructions, and unique across the workbook. A CSV upload stays flat, with the indicators ;-separated in one indicators_coverage column. See statgpt/cli/README.md. #765 (#764)

Warning

Breaking change in the Availability Query tool (agent and MCP): partial_query is now a list of typed filters instead of a map, so the schema is valid for OpenAI strict function calling. Callers must send [{"dimension_id": "SERIES", "codes": ["51401"]}] instead of {"SERIES": ["51401"]}, and an empty list ([]) instead of {} for availability across the whole dataset. #769 (#770)

Features

• Grade C discovery datasets read their indicators from a separate sheet per dataset. A new admin limit caps how many indicators a dataset can list. #765 (#764, #776)
• QuantHub and StatGPT SDMX proxy datasets take their update date from the dataUpdatedAt annotation first, then from lastUpdatedAt, then from the citation's last_updated. Dataflows are requested with forceDataflowDynamicAnnotations=true so the provider returns the annotation. A dataset with its own updated_at sources in its config is not affected. #773 (#774)
• Stages relayed from downstream deployments keep their nesting instead of all showing at the top level. They also keep the status they were closed with, so a step that failed downstream no longer shows as completed. #749 (#751)

Fixes

• The Availability Query tool works with models that use OpenAI strict function calling. #769 (#770)
• Series and dataset attributes in SDMX-JSON 1.0 data responses from SDMX 2.1 sources, such as QuantHub or IMF, are read correctly. Before, series attributes were looked up in the dataset's attribute list, dataset attributes were dropped, and an attribute without a value shifted the ones after it onto the wrong attributes. #778 (#779)

Other

• bump opentelemetry from 1.39.1 to 1.45.1 (instrumentations from 0.60b1 to 0.66b1), black from 26.3.1 to 26.10.1 and virtualenv from 21.7.13 to 21.14.13 to address security advisories (#777)
• bump setuptools from 83.0.0 to 84.2.1, fastmcp from 3.4.4 to 3.4.11 and filelock from 3.32.7 to 4.2.4 to address security advisories (#783)

Deployment Changes

• For new updated_at strategy to work on QunatHub sdmx registry QuantHub application config should have DISABLE_DATAFLOW_ANNOTATION_INJECTION": {"value": "false", "type": "BOOLEAN"}