fix(safe-synthesizer): fail fast with a clear error on non-tabular input

[marcusds]

Summary

A Safe Synthesizer job whose input fileset has nested list/dict columns (e.g. function-calling JSONL like xlam-function-calling-60k) crashed deep inside the library's preflight with an opaque, unactionable error:

preflight.check_crash: TypeError: unhashable type: 'list'
nemo_safe_synthesizer.errors.ParameterError: Pre-flight check failed with 1 error(s)

Root cause: the library's ConstantColumnCheck runs data[col].dropna().nunique(), which can't hash list-valued cells. Safe Synthesizer is a tabular synthesizer, so nested columns are never valid input — but nothing told the user that.

Change

Validate the loaded DataFrame in run_config() (the common chokepoint for both platform and local runs) before handing it to SafeSynthesizer, and raise a ParameterError that names the offending columns and explains the requirement:

Input data is not flat tabular: column(s) tools, metadata contain nested
list/dict values. Safe Synthesizer expects one scalar value per cell —
flatten these columns or serialize them to strings (e.g. JSON) before running.

Only object-dtype columns are scanned (numeric/bool are skipped), so the check is cheap.

Notes

• Raises the library's own ParameterError so it surfaces the same way as other bad-input errors.
• Does not attempt to synthesize nested data — that's out of scope for a tabular synthesizer; this just replaces a cryptic crash with a clear, actionable message.
• The underlying ConstantColumnCheck crash is a latent bug in nemo_safe_synthesizer 0.1.7 (it should guard unhashable columns like its sibling checks do) — worth a separate upstream report.
• Unrelated to the runtime-setup dependency fix in fix(safe-synthesizer): repair runtime setup dependency resolution #856.
• 3 pre-existing ty diagnostics in this file (lines 220/416/440) are untouched by this change and present on main.

Test

test_validate_flat_tabular_data_rejects_nested_columns (raises naming the columns) and test_validate_flat_tabular_data_allows_flat_columns (flat data passes). Full test_local_run.py suite green.

🤖 Generated with Claude Code

Summary by CodeRabbit

• Bug Fixes
  • Added early validation for tabular input data.
  • Inputs containing nested lists, dictionaries, or sets now receive a clear error identifying the affected columns and recommending flattening or serialization.
  • Flat, scalar-based tabular data continues to process normally.

[github-actions]

Suite	Lines Covered	Line Rate	Branch Rate
Unit Tests	27140/34858	77.9%	62.1%
Integration Tests	15925/33570	47.4%	19.9%

[coderabbitai]

📝 Walkthrough

Walkthrough

Safe Synthesizer validates loaded DataFrames for nested values in object columns before synthesis. Invalid columns raise ParameterError; flat and nested inputs are covered by unit tests.

Changes

Flat tabular validation

Layer / File(s)	Summary
Validation and test coverage plugins/nemo-safe-synthesizer/src/nemo_safe_synthesizer_plugin/tasks/safe_synthesizer/__main__.py, plugins/nemo-safe-synthesizer/tests/unit/test_local_run.py	_validate_flat_tabular_data detects list, dictionary, and set values, run_config invokes it before synthesis, and tests cover rejection and acceptance cases.

Suggested reviewers: matthewgrossman, mckornfield

🚥 Pre-merge checks | ✅ 5

✅ Passed checks (5 passed)

Check name	Status	Explanation
Description Check	✅ Passed	Check skipped - CodeRabbit’s high-level summary is enabled.
Title check	✅ Passed	Accurately summarizes the main change: early validation with a clearer error for nested Safe Synthesizer input.
Docstring Coverage	✅ Passed	No functions found in the changed files to evaluate docstring coverage. Skipping docstring coverage check.
Linked Issues check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check	✅ Passed	Check skipped because no linked issues were found for this pull request.

✨ Finishing Touches

📝 Generate docstrings

• Create stacked PR
• Commit on current branch

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Commit unit tests in branch safe-synthesizer-tabular-input-validation/mschwab

---

_{Comment @coderabbitai help to get the list of available commands.}

[coderabbitai]

🧹 Nitpick comments (1)

plugins/nemo-safe-synthesizer/tests/unit/test_local_run.py (1)

71-91: 📐 Maintainability & Code Quality | 🔵 Trivial | ⚡ Quick win

Cover the run_config() integration path.

These tests call the private validator directly, so they would still pass if the new call at __main__.py Line 321 were removed. Add a regression test that passes nested data through run_config() and verifies ParameterError is raised before Safe Synthesizer runs.

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@plugins/nemo-safe-synthesizer/tests/unit/test_local_run.py` around lines 71 -
91, Extend the tests around _validate_flat_tabular_data with a run_config()
integration regression test that supplies nested tabular data, asserts
task_main.ParameterError, and verifies Safe Synthesizer is not invoked. Exercise
the public run_config() path rather than calling the private validator directly,
using the existing test helpers and mocking conventions.

🤖 Prompt for all review comments with AI agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

Nitpick comments:
In `@plugins/nemo-safe-synthesizer/tests/unit/test_local_run.py`:
- Around line 71-91: Extend the tests around _validate_flat_tabular_data with a
run_config() integration regression test that supplies nested tabular data,
asserts task_main.ParameterError, and verifies Safe Synthesizer is not invoked.
Exercise the public run_config() path rather than calling the private validator
directly, using the existing test helpers and mocking conventions.

---

ℹ️ Review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Enterprise

Run ID: 0228e1fb-88a0-4c7c-8e5e-30ec190b9792

📥 Commits

Reviewing files that changed from the base of the PR and between 26319c9 and 0cb6bf7.

📒 Files selected for processing (2)

• plugins/nemo-safe-synthesizer/src/nemo_safe_synthesizer_plugin/tasks/safe_synthesizer/__main__.py
• plugins/nemo-safe-synthesizer/tests/unit/test_local_run.py