remdb 0.2.6__py3-none-any.whl → 0.3.103__py3-none-any.whl

rem/cli/commands/experiments.py

@@ -5,11 +5,21 @@ Experiments use ExperimentConfig (rem/models/core/experiment.py) for configurati
 and support Git+S3 hybrid storage. Includes dataset, prompt, and trace management.
 
 Directory Structure:
-    .experiments/{experiment-name}/
-    ├── experiment.yaml          # ExperimentConfig
-    ├── README.md                # Auto-generated docs
-    ├── datasets/                # Optional: small datasets
-    └── results/                 # Optional: metrics summaries
+    experiments/{experiment-name}/
+    ├── experiment.yaml          # ExperimentConfig (metadata, agent ref, evaluator ref)
+    ├── README.md                # Auto-generated documentation
+    ├── ground-truth/            # Evaluation datasets (Q&A pairs)
+    │   ├── dataset.csv          # Input/output pairs for evaluation
+    │   └── dataset.yaml         # Alternative YAML format
+    ├── seed-data/              # Data to seed REM before running experiments
+    │   └── data.yaml           # Users, resources, moments in REM format
+    └── results/                # Experiment results and metrics
+        └── {run-timestamp}/    # Each run gets its own timestamped folder
+            ├── metrics.json    # Summary metrics
+            └── run_info.json   # Run metadata (eval framework URLs, etc)
+
+Environment Variables:
+    EXPERIMENTS_HOME: Override default experiment directory (default: "experiments")
 
 Commands:
     # Experiment lifecycle
@@ -60,7 +70,7 @@ def experiments():
 @click.option("--results-location", type=click.Choice(["git", "s3", "hybrid"]), default="git",
               help="Where to store results")
 @click.option("--tags", help="Comma-separated tags (e.g., 'production,cv-parser')")
-@click.option("--base-path", default=".experiments", help="Base directory for experiments")
+@click.option("--base-path", help="Base directory for experiments (default: EXPERIMENTS_HOME or 'experiments')")
 def create(
     name: str,
     agent: str,
@@ -69,12 +79,17 @@ def create(
     dataset_location: str,
     results_location: str,
     tags: Optional[str],
-    base_path: str,
+    base_path: Optional[str],
 ):
     """Create a new experiment configuration.
 
     Creates directory structure and generates experiment.yaml and README.md.
 
+    The experiment directory will contain:
+    - ground-truth/: Q&A pairs for evaluation
+    - seed-data/: REM data (users, resources, moments) to load before running
+    - results/: Timestamped run results
+
     Examples:
         # Small experiment (Git-only)
         rem experiments create hello-world-validation \\
@@ -90,6 +105,9 @@ def create(
             --dataset-location s3 \\
             --results-location hybrid \\
             --tags "production,cv-parser,weekly"
+
+        # Custom location
+        EXPERIMENTS_HOME=/path/to/experiments rem experiments create my-test --agent my-agent
     """
     from rem.models.core.experiment import (
         ExperimentConfig,
@@ -99,15 +117,19 @@ def create(
         ResultsConfig,
         ExperimentStatus,
     )
+    import os
 
     try:
+        # Resolve base path: CLI arg > EXPERIMENTS_HOME env var > default "experiments"
+        if base_path is None:
+            base_path = os.getenv("EXPERIMENTS_HOME", "experiments")
         # Build dataset reference
         if dataset_location == "git":
             dataset_ref = DatasetReference(
                 location=DatasetLocation.GIT,
-                path="datasets/ground_truth.csv",
+                path="ground-truth/dataset.csv",
                 format="csv",
-                description="Ground truth dataset for evaluation"
+                description="Ground truth Q&A dataset for evaluation"
             )
         else:  # s3 or hybrid
             dataset_ref = DatasetReference(
@@ -168,26 +190,167 @@ def create(
         config_path = config.save(base_path)
         readme_path = config.save_readme(base_path)
 
-        # Create datasets directory
-        datasets_dir = config.get_experiment_dir(base_path) / "datasets"
-        datasets_dir.mkdir(parents=True, exist_ok=True)
+        # Create new directory structure
+        exp_dir = config.get_experiment_dir(base_path)
+
+        # Create ground-truth directory
+        ground_truth_dir = exp_dir / "ground-truth"
+        ground_truth_dir.mkdir(parents=True, exist_ok=True)
+
+        # Create seed-data directory
+        seed_data_dir = exp_dir / "seed-data"
+        seed_data_dir.mkdir(parents=True, exist_ok=True)
 
         # Create results directory if Git-based
         if results_location == "git":
-            results_dir = config.get_experiment_dir(base_path) / "results"
+            results_dir = exp_dir / "results"
             results_dir.mkdir(parents=True, exist_ok=True)
 
+        # Create placeholder files with documentation
+        ground_truth_readme = ground_truth_dir / "README.md"
+        ground_truth_readme.write_text("""# Ground Truth Dataset
+
+This directory contains Q&A pairs for evaluating the agent.
+
+## Format
+
+**CSV format** (`dataset.csv`):
+```csv
+input,expected_output,metadata
+"What is the capital of France?","Paris","{\"difficulty\": \"easy\"}"
+```
+
+**YAML format** (`dataset.yaml`):
+```yaml
+- input: "What is the capital of France?"
+  expected_output: "Paris"
+  metadata:
+    difficulty: easy
+```
+
+## Generating Ground Truth
+
+### Using AI Assistants
+
+AI coding assistants (like Claude, GPT-4, etc.) can help generate comprehensive ground-truth datasets:
+
+1. **Generate from existing examples**: Show the assistant examples from your domain and ask it to create similar Q&A pairs
+2. **Create challenging questions**: Ask the assistant to act as a judge and generate HARD questions that test edge cases
+3. **Vary difficulty levels**: Request a mix of easy, medium, and hard questions with appropriate metadata tags
+
+Example prompt:
+```
+Based on these example documents about [your domain], generate 20 Q&A pairs
+for evaluating an agent. Include:
+- 5 easy factual questions
+- 10 medium questions requiring reasoning
+- 5 hard questions with edge cases
+Format as CSV with difficulty and category metadata.
+```
+
+### Ground Truth as Judge
+
+**Important**: Keep ground-truth data **separate** from the agent being tested:
+- Ground truth should be hidden from the agent during evaluation
+- The agent should only see the `input` field
+- The evaluator compares agent output against `expected_output`
+- This ensures unbiased evaluation
+
+### Quality Guidelines
+
+1. **Diverse Coverage**: Include various question types and difficulty levels
+2. **Domain-Specific**: Use terminology and scenarios from your actual use case
+3. **Metadata Tags**: Add difficulty, category, priority for analysis
+4. **SME Review**: Have domain experts validate expected outputs
+
+## Usage
+
+These datasets can be:
+- Loaded into evaluation frameworks (Arize Phoenix, etc.)
+- Used for regression testing
+- Converted to different formats as needed
+
+The experiment runner will automatically use this data for evaluation.
+""")
+
+        seed_data_readme = seed_data_dir / "README.md"
+        seed_data_readme.write_text("""# Seed Data
+
+This directory contains REM data to load before running the experiment.
+
+## Format
+
+Use standard REM YAML format:
+
+```yaml
+users:
+  - id: test-user-001
+    user_id: experiment-test
+    email: test@example.com
+
+resources:
+  - id: resource-001
+    user_id: experiment-test
+    label: example-document
+    content: "Document content here..."
+
+moments:
+  - id: moment-001
+    user_id: experiment-test
+    label: example-meeting
+    starts_timestamp: "2024-01-15T14:00:00"
+```
+
+## Generating Seed Data
+
+### Using AI Assistants
+
+AI coding assistants can help generate realistic seed data for your experiments:
+
+1. **From existing datasets**: Reference examples from the `datasets/` directory
+2. **Domain-specific scenarios**: Describe your use case and ask for appropriate test data
+3. **Anonymized versions**: Ask to create fictional data based on real patterns
+
+Example prompt:
+```
+Based on the recruitment dataset examples in datasets/domains/recruitment/,
+generate seed data for testing a CV parser agent. Include:
+- 3 test users
+- 5 CV documents (resources) with varied experience levels
+- 2 interview moment entries
+Use fictional names and anonymize all content.
+```
+
+### Best Practices
+
+1. **Minimal**: Only include data necessary for the ground-truth questions to be answerable
+2. **Anonymized**: Always use fictional names, companies, and content
+3. **Relevant**: Seed data should provide context for evaluation questions
+4. **Versioned**: Track changes to seed data in Git for reproducibility
+
+## Usage
+
+Load this data before running experiments:
+```bash
+rem db load --file seed-data/data.yaml --user-id experiment-test
+```
+
+This ensures your agent has the necessary context for evaluation.
+""")
+
         click.echo(f"\n✓ Created experiment: {name}")
         click.echo(f"  Configuration: {config_path}")
         click.echo(f"  Documentation: {readme_path}")
-        click.echo(f"  Datasets: {datasets_dir}")
+        click.echo(f"  Ground Truth: {ground_truth_dir}")
+        click.echo(f"  Seed Data: {seed_data_dir}")
         if results_location == "git":
             click.echo(f"  Results: {results_dir}")
         click.echo(f"\nNext steps:")
-        click.echo(f"  1. Add dataset to {datasets_dir}/")
-        click.echo(f"  2. Review configuration: {config_path}")
-        click.echo(f"  3. Run experiment: rem experiments run {name}")
-        click.echo(f"  4. Commit to Git: git add .experiments/{name}/ && git commit")
+        click.echo(f"  1. Add ground truth Q&A to {ground_truth_dir}/dataset.csv")
+        click.echo(f"  2. Add seed data to {seed_data_dir}/data.yaml (optional)")
+        click.echo(f"  3. Review configuration: {config_path}")
+        click.echo(f"  4. Run experiment: rem experiments run {name}")
+        click.echo(f"  5. Commit to Git: git add {base_path}/{name}/ && git commit")
 
     except Exception as e:
         logger.error(f"Failed to create experiment: {e}")
@@ -201,11 +364,11 @@ def create(
 
 
 @experiments.command("list")
-@click.option("--base-path", default=".experiments", help="Base directory for experiments")
+@click.option("--base-path", help="Base directory for experiments (default: EXPERIMENTS_HOME or 'experiments')")
 @click.option("--status", help="Filter by status (draft, ready, completed, etc.)")
 @click.option("--tags", help="Filter by tags (comma-separated)")
 def list_experiments(
-    base_path: str,
+    base_path: Optional[str],
     status: Optional[str],
     tags: Optional[str],
 ):
@@ -217,8 +380,13 @@ def list_experiments(
         rem experiments list --tags production,cv-parser
     """
     from rem.models.core.experiment import ExperimentConfig, ExperimentStatus
+    import os
 
     try:
+        # Resolve base path
+        if base_path is None:
+            base_path = os.getenv("EXPERIMENTS_HOME", "experiments")
+
         experiments_dir = Path(base_path)
         if not experiments_dir.exists():
             click.echo(f"No experiments directory found at {base_path}")
@@ -279,16 +447,21 @@ def list_experiments(
 
 @experiments.command("show")
 @click.argument("name")
-@click.option("--base-path", default=".experiments", help="Base directory for experiments")
-def show(name: str, base_path: str):
+@click.option("--base-path", help="Base directory for experiments (default: EXPERIMENTS_HOME or 'experiments')")
+def show(name: str, base_path: Optional[str]):
     """Show experiment details.
 
     Examples:
         rem experiments show hello-world-validation
     """
     from rem.models.core.experiment import ExperimentConfig
+    import os
 
     try:
+        # Resolve base path
+        if base_path is None:
+            base_path = os.getenv("EXPERIMENTS_HOME", "experiments")
+
         config_path = Path(base_path) / name / "experiment.yaml"
         if not config_path.exists():
             click.echo(f"Experiment not found: {name}")
@@ -348,7 +521,7 @@ def show(name: str, base_path: str):
 
 @experiments.command("run")
 @click.argument("name")
-@click.option("--base-path", default=".experiments", help="Base directory for experiments")
+@click.option("--base-path", help="Base directory for experiments (default: EXPERIMENTS_HOME or 'experiments')")
 @click.option("--version", help="Git tag version to load (e.g., 'experiments/my-exp/v1.0.0')")
 @click.option("--dry-run", is_flag=True, help="Test on small subset without saving")
 @click.option("--update-prompts", is_flag=True, help="Update prompts in Phoenix before running")
@@ -356,7 +529,7 @@ def show(name: str, base_path: str):
 @click.option("--phoenix-api-key", help="Phoenix API key (overrides PHOENIX_API_KEY env var)")
 def run(
     name: str,
-    base_path: str,
+    base_path: Optional[str],
     version: Optional[str],
     dry_run: bool,
     update_prompts: bool,
@@ -405,10 +578,14 @@ def run(
     from rem.services.git import GitService
     from rem.services.phoenix import PhoenixClient
     from rem.agentic.providers.phoenix import create_evaluator_from_schema
-    from datetime import datetime
-    import pandas as pd
+    from rem.utils.date_utils import utc_now, to_iso, format_timestamp_for_experiment
+    import os
 
     try:
+        # Resolve base path
+        if base_path is None:
+            base_path = os.getenv("EXPERIMENTS_HOME", "experiments")
+
         # Load experiment configuration
         if version:
             # Load from Git at specific version
@@ -437,36 +614,22 @@ def run(
             click.echo(f"  Mode: DRY RUN (no data will be saved)")
         click.echo()
 
-        # Load agent schema from Git or filesystem
+        # Load agent schema using centralized schema loader
         agent_name = config.agent_schema_ref.name
         agent_version = config.agent_schema_ref.version
 
         click.echo(f"Loading agent schema: {agent_name} (version: {agent_version or 'latest'})")
 
-        # Try Git first, fallback to filesystem
-        agent_schema = None
-        try:
-            git_svc = GitService()
-            agent_schema = git_svc.load_schema(agent_name, version=agent_version)
-            click.echo(f"✓ Loaded agent schema from Git")
-        except Exception as e:
-            logger.debug(f"Git not available, trying filesystem: {e}")
-
-            # Fallback to local filesystem
-            from rem.services.fs import FS
-            fs = FS()
+        from rem.utils.schema_loader import load_agent_schema
 
-            schema_path = f"schemas/agents/{agent_name}.yaml"
-            try:
-                agent_schema = fs.read(schema_path)
-                click.echo(f"✓ Loaded agent schema from filesystem")
-            except Exception as fs_error:
-                logger.error(f"Failed to load agent schema: Git: {e}, FS: {fs_error}")
-                click.echo(f"Error: Could not load agent schema '{agent_name}'")
-                click.echo(f"  Tried Git: {e}")
-                click.echo(f"  Tried filesystem: {schema_path}")
-                click.echo(f"  Make sure the schema exists")
-                raise click.Abort()
+        try:
+            agent_schema = load_agent_schema(agent_name)
+            click.echo(f"✓ Loaded agent schema: {agent_name}")
+        except FileNotFoundError as e:
+            logger.error(f"Failed to load agent schema: {e}")
+            click.echo(f"Error: Could not load agent schema '{agent_name}'")
+            click.echo(f"  {e}")
+            raise click.Abort()
 
         # Create agent function from schema
         from rem.agentic.providers.pydantic_ai import create_agent
@@ -505,73 +668,85 @@ def run(
                 return {"output": serialized}
             return serialized if isinstance(serialized, dict) else {"output": str(serialized)}
 
-        # Load evaluator schema
+        # Load evaluator schema using centralized schema loader
         evaluator_name = config.evaluator_schema_ref.name
         evaluator_version = config.evaluator_schema_ref.version
 
-        # Resolve evaluator path (evaluators are organized by agent name)
-        evaluator_schema_path = f"rem/schemas/evaluators/{agent_name}/{evaluator_name}.yaml"
-
         click.echo(f"Loading evaluator: {evaluator_name} for agent {agent_name}")
 
-        try:
-            evaluator_fn = create_evaluator_from_schema(
-                evaluator_schema_path=evaluator_schema_path,
-                model_name=None,  # Use default from schema
-            )
-            click.echo(f"✓ Loaded evaluator schema")
-        except Exception as e:
-            logger.warning(f"Failed to load evaluator: {e}")
-            click.echo(f"Error: Could not load evaluator schema")
-            click.echo(f"  Path: {evaluator_schema_path}")
-            click.echo(f"  Make sure the schema exists")
+        # Try multiple evaluator path patterns (agent-specific, then generic)
+        evaluator_paths_to_try = [
+            f"{agent_name}/{evaluator_name}",  # e.g., hello-world/default
+            f"{agent_name}-{evaluator_name}",  # e.g., hello-world-default
+            evaluator_name,                     # e.g., default (generic)
+        ]
+
+        evaluator_fn = None
+        evaluator_load_error = None
+
+        for evaluator_path in evaluator_paths_to_try:
+            try:
+                evaluator_fn = create_evaluator_from_schema(
+                    evaluator_schema_path=evaluator_path,
+                    model_name=None,  # Use default from schema
+                )
+                click.echo(f"✓ Loaded evaluator schema: {evaluator_path}")
+                break
+            except FileNotFoundError as e:
+                evaluator_load_error = e
+                logger.debug(f"Evaluator not found at {evaluator_path}: {e}")
+                continue
+            except Exception as e:
+                evaluator_load_error = e
+                logger.warning(f"Failed to load evaluator from {evaluator_path}: {e}")
+                continue
+
+        if evaluator_fn is None:
+            click.echo(f"Error: Could not load evaluator schema '{evaluator_name}'")
+            click.echo(f"  Tried paths: {evaluator_paths_to_try}")
+            if evaluator_load_error:
+                click.echo(f"  Last error: {evaluator_load_error}")
             raise click.Abort()
 
-        # Load dataset
+        # Load dataset using Polars
+        import polars as pl
+
         click.echo(f"Loading dataset: {list(config.datasets.keys())[0]}")
         dataset_ref = list(config.datasets.values())[0]
 
         if dataset_ref.location.value == "git":
-            # Load from Git
+            # Load from Git (local filesystem)
             dataset_path = Path(base_path) / name / dataset_ref.path
             if not dataset_path.exists():
                 click.echo(f"Error: Dataset not found: {dataset_path}")
                 raise click.Abort()
 
             if dataset_ref.format == "csv":
-                dataset_df = pd.read_csv(dataset_path)
+                dataset_df = pl.read_csv(dataset_path)
             elif dataset_ref.format == "parquet":
-                dataset_df = pd.read_parquet(dataset_path)
+                dataset_df = pl.read_parquet(dataset_path)
             elif dataset_ref.format == "jsonl":
-                dataset_df = pd.read_json(dataset_path, lines=True)
+                dataset_df = pl.read_ndjson(dataset_path)
             else:
                 click.echo(f"Error: Format '{dataset_ref.format}' not yet supported")
                 raise click.Abort()
         elif dataset_ref.location.value in ["s3", "hybrid"]:
             # Load from S3 using FS provider
             from rem.services.fs import FS
+            from io import BytesIO
 
             fs = FS()
 
             try:
                 if dataset_ref.format == "csv":
                     content = fs.read(dataset_ref.path)
-                    from io import StringIO
-                    dataset_df = pd.read_csv(StringIO(content))
+                    dataset_df = pl.read_csv(BytesIO(content.encode() if isinstance(content, str) else content))
                 elif dataset_ref.format == "parquet":
-                    # For parquet, we need binary read
-                    import tempfile
-                    with tempfile.NamedTemporaryFile(suffix=".parquet", delete=False) as tmp:
-                        tmp_path = tmp.name
-                        # Download via FS
-                        content_bytes = fs.read(dataset_ref.path)
-                        tmp.write(content_bytes)
-                    dataset_df = pd.read_parquet(tmp_path)
-                    Path(tmp_path).unlink()  # Clean up temp file
+                    content_bytes = fs.read(dataset_ref.path)
+                    dataset_df = pl.read_parquet(BytesIO(content_bytes if isinstance(content_bytes, bytes) else content_bytes.encode()))
                 elif dataset_ref.format == "jsonl":
                     content = fs.read(dataset_ref.path)
-                    from io import StringIO
-                    dataset_df = pd.read_json(StringIO(content), lines=True)
+                    dataset_df = pl.read_ndjson(BytesIO(content.encode() if isinstance(content, str) else content))
                 else:
                     click.echo(f"Error: Format '{dataset_ref.format}' not yet supported")
                     raise click.Abort()
@@ -615,13 +790,13 @@ def run(
 
             client = PhoenixClient(config=phoenix_config)
 
-            experiment_name = f"{config.name}-{datetime.now().strftime('%Y%m%d-%H%M%S')}"
+            experiment_name = f"{config.name}-{format_timestamp_for_experiment()}"
 
             click.echo(f"\n⏳ Running experiment: {experiment_name}")
             click.echo(f"   This may take several minutes...")
 
             experiment = client.run_experiment(
-                dataset=dataset_df,  # type: ignore[arg-type]
+                dataset=dataset_df,
                 task=task_fn,
                 evaluators=[evaluator_fn],
                 experiment_name=experiment_name,
@@ -631,12 +806,15 @@ def run(
                     "evaluator": config.evaluator_schema_ref.name,
                     "experiment_config": config.name,
                     **config.metadata
-                }
+                },
+                # Smart column detection for DataFrame -> Phoenix Dataset conversion
+                input_keys=["input"] if "input" in dataset_df.columns else None,
+                output_keys=["expected_output"] if "expected_output" in dataset_df.columns else None,
             )
 
             # Update experiment status
             config.status = ExperimentStatus.COMPLETED
-            config.last_run_at = datetime.now()
+            config.last_run_at = utc_now()
             if not version:  # Only save if not loading from Git
                 config.save(base_path)
 
@@ -657,7 +835,7 @@ def run(
                         "agent": config.agent_schema_ref.name,
                         "evaluator": config.evaluator_schema_ref.name,
                         "dataset_size": len(dataset_df),
-                        "completed_at": datetime.now().isoformat(),
+                        "completed_at": to_iso(utc_now()),
                         "phoenix_url": getattr(experiment, "url", None),
                         "task_runs": len(exp_data.get("task_runs", [])),
                     }
@@ -837,20 +1015,24 @@ def dataset_add(
             --output-keys expected_label,expected_type
     """
     from rem.services.phoenix import PhoenixClient
-    import pandas as pd
+    import polars as pl
 
     try:
         client = PhoenixClient()
 
-        # Load CSV
-        df = pd.read_csv(from_csv)
+        # Load CSV with Polars
+        df = pl.read_csv(from_csv)
+        records = df.to_dicts()
 
         # Extract data
-        inputs = cast(list[dict[str, Any]], df[input_keys.split(",")].to_dict("records"))
-        outputs = cast(list[dict[str, Any]], df[output_keys.split(",")].to_dict("records"))
+        input_cols = input_keys.split(",")
+        output_cols = output_keys.split(",")
+        inputs = [{k: row.get(k) for k in input_cols} for row in records]
+        outputs = [{k: row.get(k) for k in output_cols} for row in records]
         metadata = None
         if metadata_keys:
-            metadata = cast(list[dict[str, Any]], df[metadata_keys.split(",")].to_dict("records"))
+            meta_cols = metadata_keys.split(",")
+            metadata = [{k: row.get(k) for k in meta_cols} for row in records]
 
         # Add to dataset
         dataset = client.add_examples_to_dataset(
@@ -1091,12 +1273,12 @@ def trace_list(
         rem experiments trace list --project rem-agents --days 7 --limit 50
     """
     from rem.services.phoenix import PhoenixClient
-    from datetime import datetime, timedelta
+    from rem.utils.date_utils import days_ago
 
     try:
         client = PhoenixClient()
 
-        start_time = datetime.now() - timedelta(days=days)
+        start_time = days_ago(days)
 
         traces_df = client.get_traces(
             project_name=project,

rem/cli/commands/process.py

@@ -192,15 +192,13 @@ def process_uri(uri: str, output: str, save: str | None):
 
 
 @click.command(name="files")
-@click.option("--tenant-id", required=True, help="Tenant ID")
-@click.option("--user-id", help="Filter by user ID")
+@click.option("--user-id", default=None, help="User ID (default: from settings)")
 @click.option("--status", type=click.Choice(["pending", "processing", "completed", "failed"]), help="Filter by status")
 @click.option("--extractor", help="Run files through custom extractor (e.g., cv-parser-v1)")
 @click.option("--limit", type=int, help="Max files to process")
 @click.option("--provider", help="Optional LLM provider override")
 @click.option("--model", help="Optional model override")
 def process_files(
-    tenant_id: str,
     user_id: Optional[str],
     status: Optional[str],
     extractor: Optional[str],
@@ -217,19 +215,22 @@ def process_files(
 
         \b
         # List completed files
-        rem process files --tenant-id acme-corp --status completed
+        rem process files --status completed
 
         \b
         # Extract from CV files
-        rem process files --tenant-id acme-corp --extractor cv-parser-v1 --limit 10
+        rem process files --extractor cv-parser-v1 --limit 10
 
         \b
         # Extract with provider override
-        rem process files --tenant-id acme-corp --extractor contract-analyzer-v1 \\
+        rem process files --extractor contract-analyzer-v1 \\
             --provider anthropic --model claude-sonnet-4-5
     """
+    from ...settings import settings
+    effective_user_id = user_id or settings.test.effective_user_id
+
     logger.warning("Not implemented yet")
-    logger.info(f"Would process files for tenant: {tenant_id}")
+    logger.info(f"Would process files for user: {effective_user_id}")
 
     if user_id:
         logger.info(f"Filter: user_id={user_id}")