Replication Part A/B validation

all_solutions rows: 3000
unique task IDs: 500 (expected 500)
task range: 11-510
pass_mbpp=true: 1914
pass_mbpp=false: 1086
gpt-5.4: rows=1000, pass=602, fail=398
claude-sonnet-4-6: rows=1000, pass=708, fail=292
gemini-2.5-pro: rows=1000, pass=604, fail=396
passing rows with complete 4 Sonar metrics: 1914
failing rows with all metric fields blank: 1086
Gemini raw API responses: 1000
Gemini HTTP 200: 1000
Gemini returned model gemini-2.5-pro: 1000
Gemini finishReason=stop: 1000
current-run Sonar report-task files: 1914
SonarQube serverVersion=26.4.0.121862: 1914
