Model card: evaluation link, simplified live-web figure and table

README.md

@@ -86,7 +86,7 @@ parsing and no answer outside the options you list.
 
 ![WebJev-35B-A3B turns a state and typed questions into option probabilities](assets/overview.png)
 
-## Evaluation
+## [Evaluation](https://github.com/lexmount/WebJev/tree/main/evaluation)
 
 All numbers are measured with the released BF16 weights served by vLLM, at temperature 1.0. The jev-1.13 numbers were
 measured through its official API on the same inputs.
@@ -97,8 +97,7 @@ WebJev serves as the decision component of the same browser agent on 125 real-we
 26 from WebGym and 24 from WebVoyager. Each run has a budget of 900 seconds and 60 actions. A deterministic grader
 checks the final page state and answer.
 
-- **Success rate** = solved ÷ evaluable tasks. Tasks lost to browser infrastructure or grader errors are excluded.
-- **Strict rate** counts all 125 tasks.
+Success rate = solved ÷ evaluable tasks; tasks lost to browser infrastructure or grader errors are excluded.
 
 | Task set | jev-1.13 | **WebJev-35B-A3B** |
 |---|---:|---:|
@@ -106,10 +105,6 @@ checks the final page state and answer.
 | Online-Mind2Web | 16.90% (12/71) | **38.36% (28/73)** |
 | WebGym | 8.00% (2/25) | **32.00% (8/25)** |
 | WebVoyager | 25.00% (6/24) | **45.83% (11/24)** |
-| Strict, all 125 tasks | 16.00% | **37.60%** |
-
-The 95% Wilson intervals of the overall rates do not overlap: 30–47% for WebJev against 11–24% for
-jev-1.13. Live websites change between runs, and one task is worth about 0.8 points.
 
 ### General structured decisions