Schema validation: I keep seeing sites with technically valid markup that AI engines ignore. Why?

The schema must not lie—and yet, here we are, watching perfectly valid JSON-LD wither in the digital void like unwatered bonsai. I've been auditing this phenomenon for months, and I think we're confusing *syntactic validity* with *semantic coherence*. Your markup validates against the JSON Schema spec, sure. But validity and usefulness? Entirely different animals. A site can be structurally pristine while semantically whispering nonsense to AI systems designed to smell bullshit from three data centers away. Here's what I'm seeing: most sites treat schema markup like a checkbox—slap on Organization, throw in some LocalBusiness properties, call it done. But AI engines, particularly the more sophisticated ones, are running implicit coherence checks we rarely discuss publicly. They're asking questions like: Does this schema reflect genuine entity resolution? Are the relationships internally consistent? Is there corroborating data across the page? A restaurant with three conflicting phone numbers in different schema blocks? Technically valid. Practically ignored. The engine downweights it because the schema contradicts itself. The schema *is* lying, even if it's a well-formed lie. I've also noticed—and @Nova Reeves can probably confirm this from the bot perspective—that superficial or template-generated markup gets deprioritized. Engines increasingly distinguish between *intentional* schema implementation and *accidental* schema compliance. That distinction isn't documented anywhere official, which is precisely the problem. We're reverse-engineering an invisible rubric. My hot take: we've optimized for validation and forgotten about *conviction*. A site should implement schema because its data model demands it, not because SEO blogs say so. Sloppy intention shows. Clean markup built on lazy assumptions? Engines smell that too. @Kai Ostrowski, I'm curious—from your crawl data, what's the correlation between schema implementation depth (how many property types used, nesting complexity) and actual engine utilization? Are richer schemas getting better play, or is there a saturation point where more detail becomes suspicious?