diff --git a/apps/web-roo-code/src/app/evals/methodology/methodology-content.tsx b/apps/web-roo-code/src/app/evals/methodology/methodology-content.tsx
index 18d9bcef3a..0285be5970 100644
--- a/apps/web-roo-code/src/app/evals/methodology/methodology-content.tsx
+++ b/apps/web-roo-code/src/app/evals/methodology/methodology-content.tsx
@@ -5,8 +5,6 @@ import {
ArrowRight,
FlaskConical,
Code,
- GitBranch,
- Building2,
AlertTriangle,
BarChart3,
Terminal,
@@ -17,6 +15,7 @@ import {
DollarSign,
Zap,
Trophy,
+ Scale,
} from "lucide-react"
import Link from "next/link"
@@ -476,126 +475,112 @@ export function MethodologyContent() {
Engineer Roles
-
+ Each role represents a different engineering seniority level. We test models against + exercises matched to that role's complexity, then score using role-specific weights. +
+ - {/* Role cards */} -
+ Different Roles, Different Weights
++ Each role has its own scoring weights. A model that's great for simple tasks might + not rank for architecture decisions. +
+ +Matched Exercises
++ Budget roles get simpler exercises. Complex roles get harder ones. The difficulty and + scoring shift together so recommendations stay relevant. +
+- Easy + Medium exercises. Boilerplate, simple bug fixes, test generation. Scoring - emphasizes{" "} - cost efficiency. -
- {/* Weight breakdown */} -- Scoring Weights +
Budget Roles
++ Cost and speed matter most. Simpler exercises where many models succeed, so + efficiency breaks the tie.
-- Medium exercises. Feature development, debugging, code review. Balanced scoring with - emphasis on{" "} - success rate + quality - . -
- {/* Weight breakdown */} -- Scoring Weights +
Complex Roles
++ Reasoning quality and success rate matter most. Harder exercises where only the + best models deliver.
-- Hard exercises. Architecture, ambiguous requirements, system design. Scoring - prioritizes{" "} - - reasoning quality + correctness - - . -
- {/* Weight breakdown */} -- Scoring Weights -
-