Runner: G2 finds the FUNCTION statement after local classes; budget floor applied to all tasks
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_014aUaQeLnwbb1zTpN7kHeat
This commit is contained in:
35
tasks_gen/eval/G0160/empirical.json
Normal file
35
tasks_gen/eval/G0160/empirical.json
Normal file
@@ -0,0 +1,35 @@
|
||||
{
|
||||
"_superseded": {
|
||||
"deepseek-v4.1-flash:cloud@2026-10-03": {
|
||||
"score": 0,
|
||||
"parts": {
|
||||
"total": 0,
|
||||
"failed_gates": [
|
||||
"G1_active",
|
||||
"G2_contract",
|
||||
"G3_hidden_runs"
|
||||
]
|
||||
},
|
||||
"hidden": "0/0",
|
||||
"tool_calls": 0,
|
||||
"seconds": 479.5,
|
||||
"run_dir": "runs/emp/13004_G0160_llm_deepseek-v4.1-flash_cloud"
|
||||
}
|
||||
},
|
||||
"deepseek-v4.1-flash:cloud": {
|
||||
"score": 90.0,
|
||||
"parts": {
|
||||
"correctness": 40.0,
|
||||
"craft": 20.0,
|
||||
"clean": 15.0,
|
||||
"own_tests": 15.0,
|
||||
"discipline": 0.0,
|
||||
"total": 90.0,
|
||||
"note": "own_tests: faulty-reference part not implemented in skeleton"
|
||||
},
|
||||
"hidden": "6/6",
|
||||
"tool_calls": 60,
|
||||
"seconds": 506.3,
|
||||
"run_dir": "runs/emp/16005_G0160_llm_deepseek-v4.1-flash_cloud"
|
||||
}
|
||||
}
|
||||
Reference in New Issue
Block a user