fix: render degenerate judge histograms instead of raising (#905)

[chethanuk]

Summary

DatasetProfilerResults.to_report() raises on four judge-profile shapes that model_validate() accepts, so a schema-valid analysis result cannot be rendered. All four are one crash family on one render path (create_rich_histogram_table and its two callers), so they're fixed together.

Related Issue

Fixes #905
Closes #903

Changes

• create_rich_histogram_table (config/utils/visualization.py): max(data.values()) raised ValueError on an empty mapping, one line before the existing max_count <= 0 guard that already handled "nothing to draw". Changed to max(data.values(), default=0), and an empty mapping now renders a no data placeholder row instead of an empty table.
• create_judge_score_summary_table (config/analysis/utils/reporting.py): took a histogram by attribute (histogram.categories), so a MissingValue — which the field is typed to allow — raised AttributeError a frame earlier than the fix above. Parameter widened to CategoricalHistogramData | MissingValue; a MissingValue now builds an empty mapping instead.
• JudgeScoreProfilerResults.create_report_section (config/analysis/column_profilers.py): indexed self.score_distributions.histograms[score_name] unguarded, while score_distributions is itself typed ... | MissingValue — raising AttributeError when missing and KeyError when a summary names a score the histogram map omits. Both now fall back to MissingValue.CALCULATION_FAILED, which the guard above renders.
• Not touched: CategoricalHistogramData's list constraints. Adding min_length=1 would make CategoricalDistribution.from_series raise on an empty Series, which _load_stage_analysis catches and turns into None — trading a visible crash for a silently discarded analysis.

Testing

• Two parametrized tests added, driven through the public to_report() entry point both reporters used. On unmodified main with the tests applied: 5 failed, 58 passed — ValueError at visualization.py:265, AttributeError at reporting.py:174, AttributeError and KeyError at column_profilers.py:146, matching both issues verbatim.
• With the fix, same command: 63 passed.
• uv run --group dev pytest packages/data-designer-config/tests → 652 passed (main: 644; the +8 is exactly the new cases).
• uv run --group dev pytest packages/data-designer-engine/tests → 2257 passed.
• make check-all-fix clean, tree unmodified.

Checklist

• Follows commit message conventions
• Commits are signed off (DCO)
• Unit regression coverage added

[github-actions]

Linked Issue Check

Issue #905 has not been triaged yet. A maintainer needs to review
the issue and add the triaged label for this check to pass.

You can continue working on the PR in the meantime. The check will
re-run automatically once the issue is triaged.

[github-actions]

All contributors have signed the DCO ✍️ ✅
_{Posted by the DCO Assistant Lite bot.}

[greptile-apps]

Confidence Score: 5/5

The PR appears safe to merge; the missing and empty histogram variants accepted by the schema now render without changing populated-histogram behavior.

Summary

• Safely computes histogram scaling for empty mappings and displays a no-data placeholder.
• Converts missing judge histogram values into empty renderable histogram data.
• Falls back gracefully when the complete distribution or an individual score histogram is unavailable.
• Adds regression coverage through the public report API and direct histogram rendering tests.

Diagram

%%{init: {'theme': 'neutral'}}%%
flowchart LR
    A[Judge score summary] --> B{Distribution available?}
    B -- No --> C[MissingValue fallback]
    B -- Yes --> D{Histogram available?}
    D -- No --> C
    D -- Yes --> E[Category-count mapping]
    C --> F[Empty mapping]
    E --> G[Histogram table]
    F --> H[No data row]
    G --> I[Rendered report section]
    H --> I

Loading

_{Reviews (2) · Last reviewed commit: "Merge branch 'main' into fix/issue-905-s..."}

[chethanuk]

I have read the DCO document and I hereby sign the DCO.

How do I sign it?

[nabinchha]

I have read the DCO document and I hereby sign the DCO.

How do I sign it?

Instructions are here: #919 (comment)

[chethanuk]

I have read the DCO document and I hereby sign the DCO.

[chethanuk]

recheck

[nabinchha]

Thanks for putting this together, @chethanuk!

Summary

This change makes judge-profile report generation tolerate the empty, missing, and incomplete histogram shapes accepted by the schema. The crash handling matches the PR's stated intent, but the new empty-state row needs one rendering correction before merge.

Findings

Warnings — Worth addressing

packages/data-designer-config/src/data_designer/config/utils/visualization.py:271 — Empty-state markup is rendered literally in saved reports

• What: The placeholder is added as the plain string "[dim]no data[/dim]", while generate_analysis_report() deliberately renders with markup=False. Through the public DatasetProfilerResults.to_report() path, the saved HTML therefore contains and displays the literal text [dim]no data[/dim] rather than a dimmed no data label.
• Why: The PR fixes the exceptions, but every repaired empty/missing-histogram case now produces visibly malformed report output on the exact path under test. The current tests miss this because they only check the report size and the value-column cells.
• Suggestion: Pass a styled renderable instead, e.g. table.add_row(Text("no data", style="dim"), "") (Text is already imported), and extend the saved-report assertion to verify "no data" is present without the Rich control tags.

What Looks Good

• Putting default=0 on the shared histogram maximum handles empty input at the narrowest reusable boundary.
• The MissingValue union and .get() fallback align report rendering with the shapes the Pydantic models already accept, without tightening the serialized schema.
• The parametrized public-path tests cover empty, missing, incomplete, and populated cases, and the full config suite remains green.

Verdict

Needs changes — render the empty-state label with a Text object and lock the saved output down with an assertion so the repaired path does not expose Rich markup tags.

---

This review was generated by an AI assistant.