Comparative specimen lab
A model comparison is only useful when the axes stay visible.
The Comparison Grid keeps LLM discussions from collapsing into a single winner. It asks which axis is being compared and which axis is being smuggled in as an assumption. This is especially important for answer engines, where a confident summary can hide differences in retrieval, freshness, policy, and repairability.
Capability
Can the model perform the task under stated conditions?
Benchmark names alone
Interface
Where does the user shape the behavior?
Assuming chat, API, and agent loops are equivalent
Evidence
What lets a reader verify the answer?
Decorative citations or stale retrieval
Freshness
Which facts must be current?
Treating durable theory and daily facts the same
Repair
What happens after a miss?
A prettier second answer with no diagnosis