Skip to main content
Tables are the workspace for dataset, evaluation, backtesting, report, and batch workflows in PromptLayer. They keep inputs, computed outputs, scoring, history, and analytics together so a workflow can be built, run, and reviewed in one place. A Table has four parts:
ObjectWhat it does
TableThe top-level workspace for a related evaluation or batch workflow.
SheetA tab with its own rows, columns, score configuration, history, and analytics.
ColumnA field in the sheet. Text columns store data; computed columns run prompts, code, checks, extraction, composition, or other work.
CellA row-column value. Computed cells track status, including queued, running, completed, failed, and stale.
A new Table with options to create from a prompt, import request history, or start blank

​
Start a Table

A new Table opens with one sheet, one text column, and one row. The empty state gives three starting points:
  1. Create from a prompt: set up prompt inputs, an output column, and evaluation checks.
  2. Import request history: select logged requests and add them as rows.
  3. Start blank: build the sheet manually from rows and columns.
Use Create from a prompt when the workflow starts from a Prompt Registry template. Use Import request history when production or test traffic is already logged. Use Start blank when you want to model the sheet yourself.

​
Main workflow

Most Tables follow a simple loop:
  1. Add rows from CSV, request history, or manual entry.
  2. Add text columns for inputs, labels, expected answers, or metadata.
  3. Add computed columns for prompts, code, assertions, extraction, comparisons, composition, or helper functions.
  4. Map computed columns to their sources.
  5. Run the sheet, a column, a row, or selected cells.
  6. Review cell status, configure scoring, and use history or analytics to compare changes.
Table toolbar showing import, export, score, analytics, history, upload, download, and tour controls

​
Product sections

Sheets

Import data, manage sheet tabs, add rows, upload CSVs, and import request history.

Columns

Add, configure, filter, sort, pin, duplicate, run, and map source columns.

Column Types

Reference text columns, computed columns, evaluation columns, helper columns, and composition.

Cells and Runs

Understand cell statuses, stale cells, reruns, selected runs, and cancellation.

Scoring

Configure score columns, Boolean scoring, numeric scoring, custom code, and winner aggregation.

History and Analytics

Review saved versions, score history, diffs, request analytics, and request-level debugging.

​
When to use Tables

Use Tables when you need a repeatable workflow over rows of examples: prompt regression tests, request-log replay, dataset creation, model comparisons, multi-step evaluations, human review queues, or batch jobs. Use legacy Evaluations and Datasets only for existing legacy workflows. For new work, start in Tables.

​
API references

Use the Tables API when you want to automate the same actions from code.

Create a Table

Create a Table programmatically.

Create a sheet

Add a new sheet to an existing Table.

Create a column

Add text or computed columns to a sheet.

Run an operation

Queue recalculation work for rows, columns, cells, or stale work.