Xcientist

Discover the concrete benchmark surfaces and entrypoints required for evaluation

Context-efficient file and terminal usage for experiment agents

Enable all required experiment conditions with runnable commands

Build an ordered executable code plan for controlled experiment enablement

Normalize idea components into explicit component-disable requirements

Create and verify the experiment runtime environment

Plan ordered prepare subtasks and end with reviewer-approved handoff artifacts

Deterministic protocol for repository, dataset, and benchmark acquisition

Execute unified science conditions through reviewer and hook gates

Build unified science condition plans from validated prepare/code evidence

Analyze and audit technical blog posts with engineering-focused scoring, source fidelity verification, and AI detection optimized for jargon-rich content

Refine and improve existing blog articles based on analysis recommendations, with intelligent judgment on which issues to fix

Read blog_analysis.md and project source files, then write a complete blog article with image placeholders and method files

Navigate the workspace, find and read project-related files, then write blog analysis to output directory

Introduce a structured alternative treatment for failure regimes so rare cases and recovery behavior are explicit without turning the method into generic control scaffolding.

Turn an open-loop process into a feedback-guided update mechanism that measures outcomes and adapts an existing rule online.

Replace a flat control or execution flow with a hierarchy that assigns responsibilities across explicit levels.

Commit to one concrete mechanism-level innovation by strengthening one weak existing component and keep scaffolding secondary to the task-solving path.

Add a cross-scale consistency mechanism that aligns representations or decisions across scales without making coordination overhead the main contribution.

Introduce speculative or optimistic execution plus explicit repair so the system can pursue fast paths without giving up recoverability.

Perform a localized block-level intervention with replace/rewire and prove gain with focused ablation.

Inject a theory-backed mechanism or invariant from another domain and wire it into the core objective or mechanism rule.