LLM-81: Add Excel export for summary comparisons and CLI command

[benglewis]

User description

Motivation

• Provide a convenient way to export per-dataset comparison workbooks (one sheet per dataset) from summary CSVs for analysis and reporting.
• Make metric column names consistent across the codebase to avoid string duplication and reduce chance of typos.

Description

• Added an export_excel utility (llm_behavior_eval/export_excel.py) that merges two summary_brief.csv files, normalizes metrics, writes an .xlsx workbook with one sheet per dataset, and inserts comparison charts; it uses xlsxwriter and validates input overlap and numeric values.
• Introduced metric name constants in llm_behavior_eval/evaluation_utils/metrics.py and updated BaseEvaluator to use these constants instead of hardcoded column header strings.
• Exposed the Excel export as a CLI command export-excel in llm_behavior_eval/evaluate.py via the export_excel_command wrapper and registered it on the typer app.
• Documented the feature in README.md including usage example and install note to enable the optional excel dependency.
• Added an optional dependency group excel with xlsxwriter>=3.2.0 and included excel in the dev dependency group in pyproject.toml.
• Added tests: a new tests/test_export_excel.py suite validating sheet sanitization, file output, and overlap checking, and a CLI test in tests/test_evaluate_cli.py that checks the new command forwards options correctly.

Testing

• Ran the test suite with pytest including the new tests/test_export_excel.py and modified tests/test_evaluate_cli.py tests, and they passed.
• Verified the new CLI unit test test_export_excel_command_passes_new_option_names passed and correctly exercises the export_excel_command argument mapping.

---

Codex Task

---

Generated description

Below is a concise technical summary of the changes proposed in this PR:
Add Excel export as a CLI workflow via the new export_excel helper and register the export-excel command so teams can generate comparison workbooks directly from the summary_brief.csv outputs, complete with optional labels and dataset filtering along with docs and dependency updates to install xlsxwriter. Introduce metric column constants in evaluation_utils.metrics and consume them in BaseEvaluator so CSV generation stays consistent while also powering the Excel exporter.

Topic	Details
Excel export flow	Implement Excel export workbook generation, CLI wiring, docs, dependency extras, and tests to let users compare pretrained/unlearned summaries via export_excel and export-excel commands with customizable labels and sheet selection. Modified files (7) README.md llm_behavior_eval/evaluate.py llm_behavior_eval/export_excel.py pyproject.toml tests/test_evaluate_cli.py tests/test_export_excel.py uv.lock Latest Contributors(2) User Commit Date orr@hirundo.io LLM-75-Fix-Result-Dire... March 01, 2026 mishana4life@gmail.com LLM-64-Change-results-... February 16, 2026
Summary metrics	Standardize summary metric headers by using evaluation_utils.metrics constants inside BaseEvaluator so generated CSVs remain aligned with the export tooling. Modified files (2) llm_behavior_eval/evaluation_utils/base_evaluator.py llm_behavior_eval/evaluation_utils/metrics.py Latest Contributors(2) User Commit Date orr@hirundo.io LLM-76-Add-option-to-r... March 02, 2026 blewis@hirundo.io LLM-68-Remove-empty-co... February 19, 2026

This pull request is reviewed by Baz. Review like a pro on (Baz).

[baz-reviewer]

Spec Reviewer Report    📬

Checkout in Baz

2 / 3 requirements met for ticket:

Please add the ability to create an Excel spreadsheet (.xlsx) as an output

1 unmet requirement

#	Requirement	Explanation
1	Dev group installs Excel optional extra - Not Met	pyproject.toml was not modified in the diff to declare an 'excel' optional extra or to include it in the dev dependency group; therefore packaging metadata changes required by this requirement are missing. evidence pyproject.toml:50-53 optional 'excel' extra with xlsxwriter pyproject.toml:64-72 dev dependency group installs llm-behavior-eval[excel,…] attempted:true

2 met requirements

#	Requirement	Explanation
1	Export Excel comparison subcommand	New CLI command forwards dataset filters and model names to export logic that merges summary CSVs, produces one sheet per dataset, and adds a comparison bar chart for the two models. evidence llm_behavior_eval/evaluate.py:584-644 new export-excel CLI command llm_behavior_eval/export_excel.py:117-189 merges data, filters datasets, writes sheets, adds chart tests/test_evaluate_cli.py:463-492 confirms CLI options forward to export function
2	Per-dataset Excel sheet with comparison chart	The new export path groups merged metrics by dataset, writes each dataset’s category table with both model columns, and inserts a horizontal bar chart comparing the two models per sheet. evidence • llm_behavior_eval/export_excel.py:145-189 dataset loop writes tables and adds bar charts per sheet • tests/test_export_excel.py:41-87 ensures the Excel file contains the requested dataset sheet and numeric values for both models

Note: Some optional integrations are missing, so it might not be possible to check some of the requirements.
For best results, make sure the following are integrated: Figma

---

Used resources:
Hash: 30befef | Ticket: link

To rerun the Spec Reviewer, comment "baz rerun spec review".

[shmuelyo]

lgtm