Skip to main content
PATCH
Core - Set an annotation cell
Related guide: Experiments guide. See the Experiments guide for a walkthrough with examples.

Authorizations

Authorization
string
header
required

Bearer authentication header of the form Bearer <token>, where <token> is your auth token.

Path Parameters

experiment_id
string
required

ID of the experiment.

run_id
string
required

ID of the run.

annotation_column_id
string
required

ID of the annotation column.

datapoint_id
string
required

ID of the row.

Body

application/json
value
object
required

Annotation value. type is required: range uses {type: range, value: number}; single- and multi-select use {type: categorical, value: string} or {type: categorical, values: [{value: string, sentiment: string}]}; text and correction use {type: free_form_text, text: string}.

task_id
string
required

ID of the task column whose response is being annotated. An annotation column holds one cell per task column.

Response

OK

A cell: the value of one column in one row.

cell_id
string

Cell ID. Token metric cells are derived when read and use {row_id}_{task_column_id}_{metric_key}.

row_id
string

Row the cell belongs to.

column_id
string

Column the cell belongs to.

status
enum<string>

Execution state. CELL_STATUS_NOT_RUN: never executed, which includes dataset cells. CELL_STATUS_QUEUED: waiting to execute. CELL_STATUS_RUNNING: executing. CELL_STATUS_COMPLETED: finished with a value. CELL_STATUS_FAILED: execution failed, see error_message. CELL_STATUS_CANCELED: canceled before it finished.

Available options:
CELL_STATUS_UNSPECIFIED,
CELL_STATUS_NOT_RUN,
CELL_STATUS_QUEUED,
CELL_STATUS_RUNNING,
CELL_STATUS_COMPLETED,
CELL_STATUS_FAILED,
CELL_STATUS_CANCELED
value
object | null

Stored documents are returned as-is; strings are wrapped as {"text"}, arrays as {"messages"} on messages cells and {"items"} elsewhere, other scalars as {"value"}. Null for a cell that has no value yet. Prompt and agent cells hold the response with last_message, model, usage, latency and total_cost; evaluator and annotation cells hold type and value (or text), with an optional explanation.

artifacts
object | null

Null when the cell holds no trace or response reference.

error_message
string

Reason the cell failed. Cleared when the cell is queued again by Execute a run.

type
enum<string>

Kind of cell, set by its column. CELL_TYPE_INPUTS, CELL_TYPE_MESSAGES and CELL_TYPE_TEXT hold data (expected output and job output are text). CELL_TYPE_RESPONSE: prompt column. CELL_TYPE_AGENT_RESPONSE: agent column. CELL_TYPE_OPENRESPONSE: openresponse column. CELL_TYPE_EVALUATOR, CELL_TYPE_METRIC and CELL_TYPE_ANNOTATION: evaluator, metric and annotation columns.

Available options:
CELL_TYPE_UNSPECIFIED,
CELL_TYPE_TEXT,
CELL_TYPE_NUMBER,
CELL_TYPE_INPUTS,
CELL_TYPE_MESSAGES,
CELL_TYPE_RESPONSE,
CELL_TYPE_AGENT_RESPONSE,
CELL_TYPE_EVALUATOR,
CELL_TYPE_METRIC,
CELL_TYPE_ANNOTATION,
CELL_TYPE_OPENRESPONSE
task_column_id
string

Task column an evaluator, metric or annotation cell belongs to. Empty on the placeholder cells of a run without task columns.

created_at
string<date-time> | null

Null on cells written before timestamps were recorded.

updated_at
string<date-time> | null

Time of the last write to the cell. Null on cells written before timestamps were recorded.