The original title is about a readability test for a compression format scoring 0/24. Let me rewrite this for a mobile feed.

Short summary

The author tested ctxfold's CSV compression format for LLM readability and scored 0/24—models couldn't find rows or reconstruct values from folded CSV. The root cause: CSV has no syntax to lift, so compression creates indirection through prefix headers that models can't reliably resolve. The fix was documentation: CSV folding is now pipeline-mode only, decompressed before reaching the model.

• •Folded CSV scored 0/24 on GPT-4o-mini readability test; GPT-4o managed only 6-9/24
• •Models read values directly, not reconstruction rules—header indirection kills readability
• •CSV folding moved to pipeline-mode: compress for transit, decompress before prompt building

Generated with AI, which can make mistakes.