A reporting layer
over evaluation
infrastructure.
Evaluation Cards is a collection of reported model–benchmark results, organized under a five-level rollout hierarchy and four interpretive signals computed over the joined record.
Interpretive signals
Four signals computed over each (model, benchmark, metric-path) record and aggregated to the corpus level. Per-record instances appear on every model and benchmark page.
of results, by model, benchmark, split and metric, disclose every setup item that applies. The rest cannot be independently re-run.
78% leave at least one setup item undisclosed. Most often undisclosed: Output length (77%), Decoding (76%).
mean across 134,652 reported score triples.
Observed range: 7% to 93%.
of reported score triples have reports from more than one party.
92% third-party, 8% first-party of 134,652 unique triples.
of setup-eligible groups diverge across variants (475 of 2,761).
Cross-party divergence: 21%.
Benchmark families
All 93 →Mercor ACE
1 reported benchmark across this family.
AgentHarm
1 reported benchmark across this family.
How Inference Compute Shapes Frontier LLM Evaluation
7 reported benchmarks across this family.
aiXamine
311 reported benchmarks across this family.
Alpaca-EVAL-V1
1 reported benchmark across this family.
Alpaca-EVAL-V2
1 reported benchmark across this family.
Every score resolves to an explicit path through this hierarchy, so aggregate claims drill down to the evidence supporting them.