The question
In my recurrence-encoding study, I compared an explicit typed notation with a shorter version that has the same checked semantics. I also included shape templates and two Rust controls. The central question was simple: does a shorter representation make code generation cheaper?
Equivalent reference solutions in the shorthand used 42.35% fewer auxiliary tokens. Initial generated programs were shorter too. But a generation budget includes failed attempts and repairs, not only the final code.
What the runs showed
Across four hosted model configurations, the shorthand generally solved fewer tasks than the explicit notation. For three configurations, reported output per solved repetition was 1.32 to 2.60 times as high with shorthand. The fourth configuration was closer, and incomplete usage data prevented a complete cost ratio.
These are results under one frozen task protocol, not proof that compact syntax causes worse generation in general. Model familiarity, the available interface, and the repair budget all matter.
The practical lesson
Measure the whole path to a correct answer. A notation can compress reference code and still raise the cost of producing working code. If the goal is lower inference spend, count unsuccessful calls, feedback, and repairs alongside program length.