Skip to main content
Legacy table scoring is deprecated for new workflows. Use Scorecards for new evaluation workflows and migration planning.
Scoring turns a sheet into a summary signal you can compare across runs, prompt versions, and workflow changes. Use score configuration when you need a single sheet-level score. Use scorecards when you need reusable criteria, row-level evidence, and calculation history. Click Score in the Table toolbar to open the score panel after computed columns produce the outputs or checks you care about.
Score button in the Table toolbar

​
Score panel

The score panel shows the current result, column and sub-score breakdowns, configuration, and recalculation status.
Score panel showing average score, column breakdown, and score configuration

​
Configure scoring

In Scoring configuration, choose a scoring mode and the columns that count toward the score.
Score configuration panel with scoring mode, score columns, Boolean token settings, assertion aggregation, and Recalculate
For non-custom, non-aggregate modes, choose Score columns. Changes save automatically; use Recalculate after changing score settings.

​
Scoring modes

​
Boolean scoring

Boolean mode converts selected column values into pass/fail results. Configure true tokens, false tokens, and assertion aggregation (Mean, All, or Any). Use Boolean scoring for assertion columns, quality checks, moderation checks, format checks, and other pass/fail evaluations.

​
Numeric scoring

Numeric mode averages selected numeric outputs. Use it when columns return scores, distances, similarity values, ratings, or normalized metrics. For comparable version history, make sure higher values consistently mean better quality.

​
Auto detect scoring

Auto detect chooses boolean or numeric handling based on the selected score column outputs. Use it when the selected columns are already clean booleans or numbers and you do not need custom token rules.

​
Custom code scoring

Custom code mode scores the whole sheet with Python or JavaScript. The scorer receives sheet data and must return a deterministic object with a numeric score. Required key:
Optional keys:
Supported score_matrix shapes:
  • 2D matrix: list[list[cell]].
  • Single-table 3D matrix: list[list[list[cell]]] where the top-level length is 1.
Matrix cells can be numbers, strings, null, or objects like { "value": 0.92, "positive_metric": true }. Use custom code when score logic depends on multiple columns, row-level weighting, custom sub-scores, or a custom matrix display.

​
Winner and aggregate scoring

Winner / aggregate mode summarizes one column into a qualitative result. Available questions: Choose the source Column. For lowest or highest value, optionally choose Show winner as to display a label from the same row instead of only the metric value. Use aggregate scoring for model bakeoffs, latency comparisons, cost comparisons, routing decisions, or any sheet where the result is a winner rather than an average.

​
Read the score

Use the score summary to compare the average score or aggregate result, inspect column and sub-score breakdowns, and see skipped values or recalculation errors.

​
Recalculate after changes

Recalculate the score after:
  • Changing the scoring mode.
  • Changing score columns.
  • Updating true or false tokens.
  • Editing custom scorer code.
  • Changing aggregate settings.
  • Rerunning computed cells that feed the score.

​
API references

Get sheet score

Read the current score result.

Configure score

Configure scoring for a sheet.

Recalculate score

Queue a score recalculation.

Score history

Read score history for a sheet.

Scorecards guide

Learn when to use scorecards and how to migrate.

Get scorecard

Read the active scorecard and calculation state.