What's this about?
IdeaAMBIG tests whether models can recognize research-method descriptions that sound plausible but omit information a competent implementer needs. It contains 163 real gaps and 497 controlled synthetic ones.
What exactly is “codification readiness”?
It means a specification is precise enough to implement without inventing unsupported assumptions. A good high-level idea can still be unready for code.
Why is this worth noting?
The best of 13 models recovered only 9.6% of real defects, but clarification success reached 80.6% once the defect was supplied; providing the gold resolution raised downstream readiness from 14% to 98%. Finding the missing question, not asking it, is the bottleneck.