Skip to main content

Results View

Browse and analyze experiment outputs in a structured tree interface.

Overview​

The Results view displays all experiment runs from the experiments/ directory, organized in a hierarchical tree structure. It provides quick access to experiment artifacts and analysis tools.

Tree Structure​

Results
├─ Configure Evaluation… (opens configs/evaluation.yaml)
└─ Experiments
   └─ customer_support_test_20251104_143610
      ├─ Parse Results (action)
      ├─ Evaluate Results (action)
      ├─ per_trace_analysis/ (folder, appears after parsing)
      │  ├─ 01_trace_abc123.md
      │  └─ 02_trace_def456.md
      ├─ summary.json
      ├─ traces.jsonl
      ├─ observations.jsonl
      ├─ errors.json
      └─ logs.json

Features​

Experiments Folder​

All experiment runs are grouped under a collapsible Experiments folder:

  • Shows up to 15 most recent experiments
  • Sorted newest to oldest by folder name
  • Each experiment displays:
    • Name: From summary.json or folder name
    • Timestamp: Formatted as YYYY-MM-DD HH:MM:SS
    • Statistics: Success rate and run count (if available)

Example label: customer_support_test with description 2025-11-04 14:36:10 • 10 runs • 100% success

Parse Results Action​

Click Parse Results inside an experiment folder to generate per-trace Markdown:

  1. Prompts for output directory (default: per_trace_analysis)
  2. Checks if output already exists
  3. Confirms overwrite if needed
  4. Runs fluxloop parse experiment <folder> --output <dir>
  5. Refreshes view to show new per_trace_analysis/ folder

Evaluate Results Action​

Generates evaluator reports defined in configs/evaluation.yaml:

  1. Prompts for output directory (default: evaluation)
  2. Confirms overwrite if directory exists
  3. Runs fluxloop evaluate experiment <folder> --output <dir>
  4. Refreshes the tree to show the new evaluation/ folder

The evaluation/ folder contains:

  • summary.json
  • per_trace.jsonl
  • report.md
  • report.html (Phase 2 interactive report)

Per-Trace Analysis Folder​

After parsing, a per_trace_analysis/ folder appears:

  • Contains one Markdown file per trace
  • Files are named: <iteration>_<trace_id>.md
  • Recursively browsable for nested outputs
  • Click any file to open in editor

Artifact Files​

Standard experiment output files:

  • summary.json
  • traces.jsonl
  • observations.jsonl
  • errors.json
  • logs.json
  • per_trace_analysis/ (appears after parsing)
  • evaluation/ (appears after running Evaluate Results)

Files only appear if present in the experiment directory.

Actions​

Configure Evaluation​

Top-level button that opens configs/evaluation.yaml for editing evaluator definitions.

Parse Results​

Appears inside each experiment folder. Executes:

fluxloop parse experiment <experiment_path> --output <dir> [--overwrite]

Evaluate Results​

Runs the evaluator pipeline:

fluxloop evaluate experiment <experiment_path> --output evaluation [--overwrite]

Outputs land in the evaluation/ folder under the selected experiment.

Open Files​

Click any artifact file to open it in the appropriate editor:

  • JSON files: Syntax-highlighted JSON editor
  • JSONL files: Line-by-line text view
  • Markdown files: Markdown editor with preview support

File Watchers​

The Results view automatically refreshes when:

  • New experiments are created in experiments/
  • Existing experiment files are modified
  • Parse operations complete

Tips​

  • Quick Parse: Use Parse Results directly from the tree instead of Command Palette
  • Timestamp Sorting: Most recent experiments appear at the top
  • Multi-Parse: Parse different experiments with different output directories for comparison
  • Nested Analysis: per_trace_analysis/ can contain subdirectories; all are browsable