Shared columns#

Columns that no single operation owns: they sit on every row, or on every table of one kind. Each is documented once, here.

On every measurement row#

Every row of deliverables/measurements.csv says which colony it describes and where that colony is. The pipeline adds the ten bounding-box columns to every row itself, whether or not MeasureBounds is in your pipeline; they are documented on that page (BBOX). The image bookkeeping columns are on Metadata.

OBJECT#

Metric family: Object#

Column

Description

Type

Object_Label

Integer label uniquely identifying each detected object (colony) within its source image. Acts as the per-image primary key shared by every measurement table, letting all per-colony measurements be joined together and traced back to a single colony on the agar plate.

Identity / design

GRID#

Written on rows from a GridImage only.

Metric family: Grid#

Column

Description

Type

Grid_RowNum

The row idx of the object

Identity / design

Grid_ColNum

The column idx of the object

Identity / design

Grid_RowMajorIdx

The row-major index of the object. Row major is the standard in most programming and data science array libraries. Used for indexing into 2D arrays.

Identity / design

Grid_ColMajorIdx

The col-major index of the object in an array. Lab automation logic uses column-major (column-wise) indexing for well plate operations because 96-well plates are physically arranged with 8 rows (labeled A-H) and 12 columns (numbered 1-12), and this layout maps directly to how multichannel pipettes operate.

Identity / design

With a metadata file#

METADATA_MATCH#

Written only when a run is given --metadata. True on a row that comes from the metadata file alone, with no detected colony.

Metric family: QC#

Column

Description

Type

QC_MetadataOnly

True when the row came from a –metadata CSV key that matched no measured object; every measurement/info column on the row is null.

Quality

On every quality check table#

QUALITY_CHECK#

<name> is the check’s name: ICC writes QC_ICC_Metric and ExpectedVsDetectedCount writes QC_Count_Metric.

Metric family: QC#

Column

Description

Type

QC_<name>_Flag

True when the metric crosses fail_threshold in the bad direction; eligible for curation.

Quality

QC_<name>_Metric

Headline metric in the check’s own units; the bad direction is set by the check’s _HIGHER_IS_BAD flag. Drives Status.

Quality

QC_<name>_Status

Categorical: pass | warn | fail.

Quality

On every growth model table#

MODEL_METRICS#

<metric> is the fitted column without its metric-family prefix: fitting Size_Area writes ModelMetrics_Area_MAE.

Metric family: ModelMetrics#

Column

Description

Type

ModelMetrics_<metric>_MAE

The mean absolute error

Quality

ModelMetrics_<metric>_MSE

The mean squared error

Quality

ModelMetrics_<metric>_RMSE

The root mean squared error

Quality

ModelMetrics_<metric>_R2

The coefficient of determination

Quality

ModelMetrics_<metric>_NumSamples

The number of samples used for model fitting

Quality

ModelMetrics_<metric>_OptimizerLoss

The loss of model fitting

Quality

ModelMetrics_<metric>_OptimizerStatus

The output of the optimizer status

Quality

After curation#

CURATION#

Written once you curate colonies in the results viewer, to deliverables/qc/curation_labels.parquet and to one deliverables/errors/<category>.parquet per category.

Metric family: Curation#

Column

Description

Type

Curation_Category

Error-category token assigned to a removed/triaged object.

Quality