Model Evaluator

Model Evaluator tracks two linked systems: what happens after AI classifications enter deployed decisions, and how the institutions that evaluate AI form, gain credibility, and shape which findings count. A methodology lane keeps both systems tied to evidence before their signals are scored.

What we track

• Consequence: how deployed classifications propagate through platform rules, monetization, catalog admission, account standing, recommendations, statutory obligations, and high-stakes decisions.
• Formation: who evaluates AI systems, how evaluator institutions are funded or qualified, what their methods claim, and where credibility or coverage gaps remain.
• Methodology: disciplines for separating model behavior from harness failure, evidence from interpretation, and coordination from authority.
• Who bears downstream costs, which institutions acquire decision weight, and where the honest state remains UNKNOWN.

Why it matters

Benchmark leaderboards measure model capability on curated evaluation sets. Deployment reveals which classifications actually alter access, standing, distribution, or cost. Evaluation institutions add a second layer: their reports can influence releases, procurement, regulation, and public interpretation even when the authority behind those reports is still forming.

Model Evaluator studies both layers without treating either one as benchmark authority. The question is not only whether a model scored well, but what followed from its classification and why a particular evaluation was allowed to carry weight.

Three lanes

The site publishes in three lanes: methodology (disciplines for reading evaluation reports), consequence (case studies of deployed classifiers), and formation (analysis of the emerging evaluation market).

Latest

• Coordination surface is not a gate — separating shared-method adoption from admission authority before scoring an evaluation-market event.
• The composite-instrument discriminator — how many actor classes a finding crosses, read before it is scored.
• The four-question falsifier — a standing gate for reading AI evaluation reports before scoring what they mean.

What's coming

Case studies of deployed-classifier consequences, analysis of evaluator-market formation, and reusable methods for reading the evidence connecting them. Draft presence does not imply publication; each piece crosses its own source and editorial gates.