Raycaster / evalsBack to SpreadsheetBench

SpreadsheetBench

50952

1/1Pass

SpreadsheetBench task 50952. Inspect the exact spreadsheet prompt, compare published model runs, open the agent response trace, and review grades for workbook editing.

Spreadsheet editingDual harness
ssb-50952
SpreadsheetBench
2 models
Dual (parsed + original)
PromptWorkbook objective and answer cells
ResponseOpen each trace to inspect tool use and edits
GradeCompare score and pass/fail by model

Task prompt

What the agent was asked to do

You are solving a spreadsheet benchmark task in a real workbook. Objective: Produce the correct final workbook state for the expected answer region. What matters: - Only the values in the expected answer region will be graded. - The workbook is the answer. Instructions: 1. Read the workbook and inspect the relevant data region first. 2. Infer the required result for the provided workbook instance. 3. Write the final value(s) directly into the expected answer region. 4. Do not rely on prose, formulas in your chat response, pseudocode, or VBA as the answer unless the benchmark explicitly requires those to be written into cells. 5. If the natural-language task asks for a general method, formula, or macro, convert that into the concrete result needed for this workbook instance. 6. Keep your final text response short and only summarize the workbook cells you changed. Relevant data region(s): Expected answer region(s): G2:G5 Expected answer sheet(s): Task: How can I retrieve the corresponding ID from a table in Excel based on multiple search parameters? The parameters are found within the comments column (B:B), and have the corresponding ID to be shown in the G:G column as shown, which are not fixed in length and can appear anywhere within the comments. These comments may be separated by various characters such as spaces or dashes, and may include extra information irrelevant to the search. Additionally, the search parameters themselves may contain dashes, and not every comment will necessarily have a corresponding ID. I have attempted to split the comments into separate columns, but this method fails as the comments vary in format, the parameters aren't consistently presented, and I'm dealing with a large dataset. I'm working with a couple thousands of data.

Published trajectories

Agent runs on this task

Curated dual-harness runs. Best scored run per model.

ModelHarnessScoreResultLinks
GPT-5.4showcasedual1/1Pass
Gemini 3.1 Produal1/1Pass