Frontend benchmark is replaced with the 70-case agent core. Old 276 cases are gone.

[codewithshinde]

Frontend benchmark is replaced with the 70-case agent core. Old 276 cases are gone.

What we do

• suites/frontend/cases/

  • feature.jsonl # 20
  • bugfix.jsonl # 20
  • docs.jsonl # 10
  • retrieval.jsonl # 10
  • testing.jsonl # 10

• All mode: agent, variant: 1 only

• Graded with bench/grade.mjs + npm run build / typecheck / npm test

• Seeded bugs in fixtures (typos, off-by-one, wrong roles, etc.)

• Same command: pnpm benchmark:frontend