What I learned as a hired consultant to autodidact physicists

Many programming languages, especially domain-specific ones, are designed by amateurs. How do we prevent obvious irregularities and disasters in languages before they become widespread? (Aspects of Javascript and R come to mind.)

Folk in the human-computer interaction community have a notion of 'expert evaluation'. I wonder if we could develop something similar for programming languages?

Jakub Zalewski passed me the article 'What I learned as a hired consultant to autodidact physicists', which treads related ground, but for physics rather than computing.