AI & developer tools
ALX
Research into compiling human intent into a structure an agent can actually execute.
The question
Can a smaller, structured representation preserve the obligations in a human request without losing the source meaning or inviting the model to improvise requirements?
The approach
The research moved from compact prompting to source-preserving execution contracts, bounded repair, and explicit evaluation. Prompt-only and runtime-assisted results are judged separately.
Where it stands
The compact v3 prompt failed its general validation gate. The v3.1.9 compiled-contract runtime later earned conditional validation for the tested 8B-and-larger models, latency, repair, and scope conditions.
The current boundary
The runtime remains opt-in and in shadow use. Conditional evidence is not a general claim of better reasoning or universal compression.