It is February 7, 2026, afternoon.
During some free time while guiding AI work, I'll write a little log.
CZON Removed YAML Front Matter
CZON released version 0.8.6.
The main change is the removal of YAML Front Matter. Currently, the AI-powered metadata extraction feature is powerful enough, eliminating the need to manually add this information.
Previously, CZON would first assemble the YAML FrontMatter and then use AI to translate it into the target language. However, if the content format within the YAML was complex, such as containing quotes, the AI translation often failed to maintain the correct format.
The solution is to have the AI translate Metadata directly from JSON to JSON. The JSON format is widely trained on by various AIs, whereas YAML is not.
Furthermore, for scenarios where the original text should not be modified but metadata needs to be re-extracted, separating Metadata and content is beneficial. They are processed separately and merged into the final HTML during the rendering stage. For example, CZON may experimentally support some new Metadata fields in the future, but I don't want users to have to retranslate the entire article; users would only need to translate the Metadata.
When translating Metadata, a per-language, per-file translation approach is used, which might increase the number of AI calls slightly. However, for AI services billed by usage volume, this doesn't have a significant impact.
In other words, it is strongly discouraged to use AI services billed per call. By the way, I believe the translation function can utilize free-tier AI services, as this task is simple and requires almost no reasoning capability.
Additionally, CZON has added more Metadata fields and optimized some SEO-related content.
SandTable Refactoring
Also, I'm currently working hard to refactor SandTable (ST) so it can accept real historical market data for testing. However, some lingering technical debt still needs to be addressed.
Claude Opus 4.6 still performs mediocrely in long-range refactoring tasks. It still comes up with some inexplicable designs, and its judgment on compatibility remains somewhat inaccurate, requiring human planning.
It seems AI still hasn't overcome the dilemma of over-compatible design.
