Evidence Engine
Grade the strength of evidence behind a claim across a body of studies — transparently, and show what would raise the grade.
How a pile of studies becomes a letter
There is no proprietary black box. The grade is four transparent steps over numbers you can see and change.
A randomized trial controls for confounders a petri dish cannot. Design is the biggest lever, so it scores ×1 to ×4.
Bigger studies count more — but with diminishing returns. Going 50 → 500 matters more than 5,000 → 50,000.
Each study's weight = design × size, signed by its result (supports +1, mixed 0, null −1). Direction = the share of decided weight that supports the claim. Consistency = how unanimous the decided studies are. Mass = 1 − e^(−weight/12), so more and bigger studies saturate toward certainty. The score is 100 × direction × mass — and a lone study is capped below an A until someone replicates it.
Same starting evidence, same supporting result — the RCT moves the grade 9× as far, because design weight dominates.
