[OPIK-3210] [FE] Add span-level online scoring rules support

[thiagohora]

Details

This PR implements frontend support for Span-Level Online Scoring Rules (OPIK-3210). The backend implementation is already complete, and this PR adds the frontend UI components and logic to support creating and managing span-level automation rules.

Key Changes:

• Scope Selector: Added "Span" option to the scope selector in rule creation dialog (gated by feature toggle)
• Filter Builder: Implemented span filter builder with all span-specific fields:
  • Basic fields: ID, Name, Type, Input, Output, Duration
  • Usage fields: Total tokens, Prompt tokens, Completion tokens
  • Cost and error fields: Estimated cost, Errors
  • Provider fields: Model, Provider
  • Dictionary fields: Metadata, Tags, Feedback scores, Custom filter (input/output paths)
• Field Binding: Added autocomplete support for binding span attributes (input/output/metadata paths) to scorer inputs
• Operators: Added is_empty and is_not_empty operators for dictionary filters (only available in rule filter context)
• Templates: Ensured only custom LLM-as-judge template is available for span scope (no built-in templates)
• Tests: Added comprehensive backend tests for IS_EMPTY and IS_NOT_EMPTY operators on span filters
• Bug Fix: Fixed OUTPUT_JSON field extraction to properly handle nested keys (was missing key extraction logic)

Technical Implementation:

• Extended EVALUATORS_RULE_TYPE and EVALUATORS_RULE_SCOPE enums to include span types
• Created SPAN_FILTER_COLUMNS with all span-specific filterable fields
• Updated filter normalization to handle backend input_json/output_json fields
• Added rule-specific operators for dictionary filters (is_empty/is_not_empty)
• Updated variable binding autocomplete to use span paths when scope is span

https://github.com/user-attachments/assets/c411031f-5ace-4d97-80ed-61ff314f5bd3

Change checklist

• [x] User facing
• [ ] Documentation update

Issues

• Resolves OPIK-3210

Testing

• All existing tests pass (70 tests)
• Added 9 new tests for IS_EMPTY and IS_NOT_EMPTY operators on dictionary fields:
  • INPUT_JSON with missing/existing keys
  • OUTPUT_JSON with missing/existing keys
  • METADATA with missing/existing keys
  • Edge cases: null values, empty strings, null parent objects
• Manual testing:
  • Created span-level rule with filters
  • Verified filter builder shows all span fields
  • Verified autocomplete works for input/output/metadata paths
  • Verified is_empty/is_not_empty operators work correctly
  • Verified only custom LLM-as-judge template is available for spans

Documentation

• No documentation updates required (UI feature)

[CometActions]

✅ Test environment is now available!

Access Information

• URL: https://pr-4269.dev.comet.com
• Cluster: comet-ml-development
• Namespace: pr-4269
• Version: 1.9.32-4269-merge-577
• Application logs: View in Grafana

The deployment has completed successfully and the version has been verified.

[github-actions[bot]]

SDK E2E Tests Results

105 tests   104 ✅  5m 44s ⏱️   1 suites    0 💤   1 files      1 ❌

For more details on these failures, see this check.

Results for commit 74dd74dd.

[github-actions[bot]]

Backend Tests Results

  349 files   -  1    349 suites   - 1   55m 18s ⏱️ + 2m 25s 5 858 tests + 1  5 851 ✅ + 1  7 💤 ±0  0 ❌ ±0  5 847 runs  +58  5 840 ✅ +58  7 💤 ±0  0 ❌ ±0 

Results for commit 4c84845c. ± Comparison against base commit 6dd57670.

:recycle: This comment has been updated with latest results.

[github-actions[bot]]

🔄 Test environment deployment started

Building images for PR #4269...

You can monitor the build progress here.

[CometActions]

✅ Test environment is now available!

Access Information

• URL: https://pr-4269.dev.comet.com
• Cluster: comet-ml-development
• Namespace: pr-4269
• Version: 1.9.32-4269-merge-597
• Application logs: View in Grafana

The deployment has completed successfully and the version has been verified.