torchcell.sga.resolve_orientation

resolve_orientation(grid: DataFrame, layout: DataFrame, n_rows: int = 14, n_cols: int = 22, inner_row0: int = 2, inner_col0: int = 2, blank_name: str = 'Blank_media', empty_thresh: float = 1.0) → tuple[DataFrame, str, float]

Return (merged, best_op, agreement).

grid is the image-order quantification (row 1..n_rows, col 1..n_cols, size, …). layout uses plate coordinates (from read_echo_picklist), where the inner block starts at (inner_row0, inner_col0). We try all 4 orientations, map grid -> plate coords, join to layout, and pick the op whose blank/plated pattern best matches empty/present colonies.