Problem
Three exact regressions in same-ground run adhoc-2026-08-31_01-21-47 are checker false negatives. Their preview artifacts satisfy the semantic task contract but use valid SDK persistence forms that the checkers do not inspect:
skill-flow-datafabric-smoke-query: the complete nine-leaf FilterBuilder tree is at inputs.detail.filter; the checker reads only runtime CEQL and configuration.essentialConfiguration.savedFilterTrees, so it misses the Contains operator.
skill-flow-datafabric-trigger-lifecycle: the query entity is at inputs.detail.entityName; the checker accepts only inputs.detail.pathParameters.entityName.
skill-flow-ipe-ceql-where: the generic list-all-records operation with objectName: groups and /groups is semantically the requested List Groups operation; the checker requires group in the node type itself.
Expected
Normalize each semantic value across the valid live and SDK representations. Add positive coverage for both authoring-loop forms and a negative control for every widened condition. Do not weaken the business assertions.
Evidence
Source run: /home/tmatup/root/workspaces/runs/adhoc-2026-08-31_01-21-47.
Problem
Three exact regressions in same-ground run
adhoc-2026-08-31_01-21-47are checker false negatives. Their preview artifacts satisfy the semantic task contract but use valid SDK persistence forms that the checkers do not inspect:skill-flow-datafabric-smoke-query: the complete nine-leaf FilterBuilder tree is atinputs.detail.filter; the checker reads only runtime CEQL andconfiguration.essentialConfiguration.savedFilterTrees, so it misses theContainsoperator.skill-flow-datafabric-trigger-lifecycle: the query entity is atinputs.detail.entityName; the checker accepts onlyinputs.detail.pathParameters.entityName.skill-flow-ipe-ceql-where: the genericlist-all-recordsoperation withobjectName: groupsand/groupsis semantically the requested List Groups operation; the checker requiresgroupin the node type itself.Expected
Normalize each semantic value across the valid live and SDK representations. Add positive coverage for both authoring-loop forms and a negative control for every widened condition. Do not weaken the business assertions.
Evidence
Source run:
/home/tmatup/root/workspaces/runs/adhoc-2026-08-31_01-21-47.