Spaces:
Running
Running
Commit History
-
feat(mcp): inspect and repair flat category data from AgentSpot
Four read/write tools so an agent can be told "the chart named X is too flat" and actually fix it: list_board_charts what's on a board and the columns behind it inspect_board_data which charts look synthetic, with the real values adjust_chart_data rebalance one onto a realistic spread (preview by default) undo_chart_adjustment restore the snapshot taken before an adjustment Resolution is deterministic rather than inferred. The model TML already states everything needed: `column_id` carries TABLE::PHYSICAL_COLUMN, `column_type` says measure or attribute, and `model_tables[].joins[].on` carries the real join predicate -- so no fuzzy matching and no LLM parse, and it resolved 131/131 vizzes across six real boards with no SQL errors. Reshaping keeps the existing rank order, so the leading category keeps leading and simply stops being tied. Additive measures keep their grand total, leaving every KPI tile on the board untouched; rates keep their mean and stay inside…6a8d963
