Validate evaluation result format.

Heatmap: Dimension scores for generated figures with reference figure row at top.

Bar chart: Compare overall_score of generated figures with reference figure baseline.

* Build provider-specific options from environment variables * Supports various AI SDK providers with their unique configuration options * * Enviro

Convert code (mxgraphxml, HTML, etc.) to text2image prompt using LLM. This function uses an LLM to analyze the given code and generate a det

Create an LLM client from SDK config. Args: config: AutoFigure SDK Config object purpose: 'generation' or 'methodology'

* Auto-detect provider based on available API keys * Returns the provider if exactly one is configured, otherwise null

Evaluate all generated figures against a reference figure and generate visualizations. Returns: Dictionary containing re

Evaluate a single academic figure using VLM scoring. Args: figure_path: Path to the image file. figure_id: F

Extract methodology from a file (PDF or Markdown). Args: file_path: Path to the paper file Returns:

Extract methodology from raw text content. Args: text: Paper text content Returns: Extracted method

Extract core methodology from paper content using LLM. Uses SDK's LLMClient for API calls.

Get reference figure paths. Returns custom references if provided, otherwise returns built-in defaults.

* Check if auto-resubmit should happen based on tool errors. * Does NOT handle retry count or quota - those are handled by the caller.

Main function example.

Converts mxGraph XML to PNG using Playwright (headless browser) with draw.io embed. This approach mirrors how next-ai-draw-io exports diagra

Perform anonymous pairwise comparison of two figures. Args: figure1_path: Path to first figure. figure2_path

Perform pairwise comparisons in parallel. Args: comparison_tasks: List of comparison tasks, format: [(figure1_path, figu

Register AutoFigure routes with Flask app.

Reset AUTOFIGURE_CONFIG to default values to prevent cross-session pollution. This MUST be called at the start of each session's layout gener

No-op function for backward compatibility. With the new SDK-based approach, each request creates fresh LLMClient/ImageEnhancer instances,

* Compare two strings by character frequency (order-agnostic)

Validate configuration and return list of errors. Returns empty list if configuration is valid.

* Validate that required API keys are present for the selected provider