Record per-task completion statistics in task_completions.jsonl. Only called for sessions that completed without an API error.

Reset the session sandbox (for new sessions/days).

Run a single agent

Run simulation for date range Args: init_date: Start date end_date: End date

Exhaust mode: attempt every available GDPVal task, retrying API errors up to max_task_failures times per task. Date advances by one w

Save corrected data to new economic_real_value directory

Append the estimate to the JSONL output file

Save the generated meta-prompt to a JSON file

Save information to the agent's memory file. The memory is stored in markdown format and is persistent across sessions. Args: co

* Sends templated email via SES to primary and admin recipients * @param {Object} formData - Contact form data * @returns {Promise<void>}

* Send email via SES

Set the global logger instance

Set up terminal output log file for a specific date This log captures the exact terminal output with all formatting

Show practical use cases enabled by new format

Show a clear summary of improvements

Start LiveBench MCP services using direct Python execution

Test basic token and API tracking

Test querying by date and task_id

Test evaluation score threshold for payments

Test edge cases for payment threshold

Test meta-prompt generation for a single category

Test that TaskManager properly tracks and exhausts tasks

Test that all packages are available in the E2B template

Verify the token_costs.jsonl file format

Test task exhaustion with filters applied

Upload reference files to the active sandbox session. Returns remote paths inside the sandbox.

* Validate input fields

* Validates reCAPTCHA token with Google * @param {string} token - reCAPTCHA response token * @returns {Promise<boolean>} - true if valid, false othe

Validate that costs are properly separated by channel

Validate all code integration points

Validate querying capabilities

Validate the 0.6 threshold logic comprehensively

* Verify reCAPTCHA token with Google

* Verify reCAPTCHA token with Google

Watch agent data files for changes and broadcast updates This runs as a background task

Args: provider: A nanobot LLMProvider (or TrackedProvider wrapper) that supports async chat().

Initialize logger Args: signature: Agent signature/name data_path: Base path for agent data

Initialize Work Evaluator Args: max_payment: Maximum payment for perfect work data_path: Path to agent data

Initialize LLM Evaluator Args: meta_prompts_dir: Path to directory containing evaluation meta-prompts model:

Initialize Task Manager with flexible task loading Args: task_source_type: Type of task source ("parquet", "jsonl", or "