# bird-bench / toxicology__204 - taskset: [bird-bench](https://harnessreport.com/tasks/bird-bench.md) - difficulty: hard - category: database - language: - runnable from the site: no - agent timeout: 3600s ## Results by harness _none yet_ ## Instruction ``` You are given a natural language question and a SQLite database. Your task is to write a SQL query that answers the question. Database file: `/app/db.sqlite` You may use `sqlite3` or Python to inspect the schema and data. Do not modify the database. Question: What are the average structural characteristics of carcinogenic molecules among the first 100 molecules, including the number of atoms, carbon atoms, bonds, double bonds, connected atoms, and the ratio of carbon-to-carbon connections? Evidence: first 100 molecules refers to molecule_id between 'TR000' and 'TR099'; carcinogenic molecules refers to label = '+'; only molecules containing carbon or hydrogen are considered Schema: CREATE TABLE `atom` ( `atom_id` TEXT NOT NULL, `molecule_id` TEXT DEFAULT NULL, `element` TEXT DEFAULT NULL, PRIMARY KEY (`atom_id`), FOREIGN KEY (`molecule_id`) REFERENCES `molecule` (`molecule_id`) ) CREATE TABLE `bond` ( `bond_id` TEXT NOT NULL, `molecule_id` TEXT DEFAULT NULL, `bond_type` TEXT DEFAULT NULL, PRIMARY KEY (`bond_id`), FOREIGN KEY (`molecule_id`) REFERENCES `molecule` (`molecule_id`) ) CREATE TABLE `connected` ( `atom_id` TEXT NOT NULL, `atom_id2` TEXT NOT NULL, `bond_id` TEXT DEFAULT NULL, PRIMARY KEY (`atom_id`,`atom_id2`), FOREIGN KEY (`atom_id`) REFERENCES `atom` (`atom_id`) ON DELETE CASCADE ON UPDATE CASCADE, FOREIGN KEY (`atom_id2`) REFERENCES `atom` (`atom_id`) ON DELETE CASCADE ON UPDATE CASCADE, FOREIGN KEY (`bond_id`) REFERENCES `bond` (`bond_id`) ON DELETE CASCADE ON UPDATE CASCADE ) CREATE TABLE `molecule` ( `molecule_id` TEXT NOT NULL, `label` TEXT DEFAULT NULL, PRIMARY KEY (`molecule_id`) ) Output: Write ONLY the SQL query to `/app/answer.sql`. Do not include code fences, comments, or explanations. ``` --- Harness Report runs agent harnesses from their GitHub repos on Harbor tasks and records every model call. Every page is also `.md` and `.json`; index: https://harnessreport.com/llms.txt · MCP: https://harnessreport.com/mcp