autoresearch-blog

You can refine a blog draft iteratively by running up to N rounds of evaluation toward a target score. The process uses a rubric to judge each version, deciding whether to keep or discard it, and persists the experiment state to support resume operations.

Judge-driven evaluation uses a rubric to score blog drafts and decide whether to keep or discard each version. This mechanism drives the iteration process by applying structured decision points to technical analysis or opinion blogs in Chinese or English.

Yes, you can track experiment state per article to isolate experiments and avoid cross-contamination. The process maintains per-article active_experiment.json files and records baseline, keep, and discard versions on disk to support audit and rollback.

To resume a paused blog refinement experiment, use the continue operation. The system persists experiment state on disk, allowing you to pick up exactly where the previous budgeted iteration left off without losing progress.

Yes, iterative blog refinement supports both Chinese and English drafts. The judge-driven evaluation and keep or discard decisions apply to technical analysis and opinion blogs written in either language.

Budget-limited blog post refinement restricts the number of iterations to a set budget, stopping once the target score is reached or the iteration limit is hit. It focuses solely on refining a single blog draft rather than processing multiple articles simultaneously.