# check-and-repair harness

IBM researchers released Palette, which splits deck building into a designer that writes a structured brief and a coder that writes pptxgenjs code for each slide, both served by one LoRA adapter, and trains a 20B open-weight model on roughly 1,000 synthetic decks covering about 10,000 slides. The training data follows the IBM Design Language, and every generated slide is rendered and measured by a geometry checker, with repairs kept only when the checker confirms improvement. On PresentBench, Palette ranks first in Economics and Education, third in Talk, and second on the three-domain mean at 59.7 versus NotebookLM's 60.8, 1.1 points behind.

IBM researchers released Palette, which splits deck building into a designer that writes a structured brief and a coder that writes pptxgenjs code for each slide, both served by one LoRA adapter, and trains a 20B open-weight model on roughly 1,000 synthetic decks covering about 10,000 slides. The training data follows the IBM Design Language, and every generated slide is rendered and measured by a geometry checker, with repairs kept only when the checker confirms improvement. On PresentBench, Palette ranks first in Economics and Education, third in Talk, and second on the three-domain mean at 59.7 versus NotebookLM's 60.8, 1.1 points behind.

Palette splits deck building into a designer that writes a structured brief and a coder that writes drawing code for each slide, both served by one LoRA adapter, with training data synthesized by a large model and repaired through render checks. The 20B open-weight model ranks first on the Economics and Education domains of PresentBench and second on the three-domain mean at 59.7, 1.1 points behind NotebookLM and ahead of the rest of the field, while placing third on the narrative-driven Talk domain.

Palette splits deck building into a designer that writes a structured brief and a coder that writes drawing code for each slide, both served by one LoRA adapter, with training data synthesized by a large model and repaired through render checks. The 20B open-weight model ranks first on the Economics and Education domains of PresentBench and second on the three-domain mean at 59.7, 1.1 points behind NotebookLM and ahead of the rest of the field, while placing third on the narrative-driven Talk domain.