Loading ToolBench
A symbolic regression system that integrates LLM-guided configuration, physics-informed filtering, dimensional analysis, and equation discovery for scientific data analysis
This server has severe definition quality issues. While 13 tools are defined with names and basic descriptions, most lack proper input schemas, parameter descriptions, and output documentation. Tool names are inconsistent (mix of verb_noun and generic patterns). Descriptions are present but often vague (e.g., 'Executes symbolic regression search' without explaining when/why to use it). Critical: most tools show input schemas with only type and description fields, but many parameters lack descriptions entirely (e.g., 'X' and 'y' in run_pysr_search have no descriptions). No output schemas are documented. Error handling is absent, no guidance on retryability, user-fixable errors, or recovery paths. The codebase is research-oriented (PySR symbolic regression), not production-grade agent tooling.
Post-run analysis and visualization of symbolic regression results with π-aware feature handling.
Applies LLM agent's chosen action (REFINE_GRAMMAR, TUNE_HYPERPARAMETERS, etc.) to grammar and system state.
Sends a prompt to the configured Gemini model and returns the response text.
Computes generic statistics for any dataset including per-feature and target statistics.
Loads API key from .env and configures the Google Gemini model.
Main function to generate a PySR configuration by querying the LLM with knowledge base context and problem description.
No input schemas visible for most tools. Tools like configure_llm, apply_agent_action, and analyze_run show empty or minimal schemas.
Parameter descriptions missing or trivial. run_pysr_search has params 'X', 'y', 'translated_params' with only type info, no descriptions of what values are valid, expected ranges, or format.
No output schemas documented. Tools return complex objects (DataFrames, sympy expressions, dicts) but no schema describes what fields the LLM should expect.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 36 | 2026-07-28+ | v2 |
Loads data from Excel file and prepares features and target for symbolic regression.
Makes target dimensionless by dividing by a reference variable with matching dimension.
Converts PySR equation strings to sympy expressions, handling power notation, inverse functions, and engineered features.
Restores original variable names in sympy expressions by substituting Pi_* symbols with their original definitions.
Executes Phase 2 orchestration loop with iterative LLM agent guidance for grammar refinement and hyperparameter tuning.
Executes symbolic regression search using PySR with translated parameters.
Translates LLM-generated configuration JSON into executable PySR parameters.
No error handling or recovery guidance. Tools provide no indication of retryability, user-fixable errors, or what to do on failure.
Tool names lack clarity for LLM selection. 'run_pysr_search' vs 'run_phase2_orchestration' are ambiguous, does one call the other? 'apply_agent_action' is generic (what action?).