Add offline policy builder and sacred eval gate
This commit is contained in:
parent
94eae8ceba
commit
6bba85fe9a
3 changed files with 141 additions and 0 deletions
69
syncpatch/run-sacred-evals
Normal file
69
syncpatch/run-sacred-evals
Normal file
|
|
@ -0,0 +1,69 @@
|
|||
#!/usr/bin/env python3
|
||||
from __future__ import annotations
|
||||
|
||||
import json
|
||||
import subprocess
|
||||
from pathlib import Path
|
||||
|
||||
ROOT = Path('/home/openclaw/.openclaw/workspace')
|
||||
CONFIG = ROOT / 'configs' / 'sacred_eval_gate.json'
|
||||
RESULT = ROOT / 'evals' / 'results' / 'regression_results.json'
|
||||
RUNNER = ROOT / 'evals' / 'run_regression.py'
|
||||
|
||||
|
||||
def load_config() -> dict:
|
||||
try:
|
||||
return json.loads(CONFIG.read_text(encoding='utf-8'))
|
||||
except Exception:
|
||||
return {
|
||||
'min_route_accuracy_percent': 95.0,
|
||||
'min_memory_signal_accuracy_percent': 90.0,
|
||||
'max_timeouts': 0,
|
||||
'require_pass': True,
|
||||
}
|
||||
|
||||
|
||||
def run_regression() -> int:
|
||||
proc = subprocess.run(['python3', str(RUNNER)], capture_output=True, text=True)
|
||||
return proc.returncode
|
||||
|
||||
|
||||
def main() -> int:
|
||||
cfg = load_config()
|
||||
rc = run_regression()
|
||||
if rc != 0:
|
||||
print(json.dumps({'ok': False, 'decision': 'hold', 'reason': 'runner_failed', 'runner_rc': rc}, ensure_ascii=False))
|
||||
return 1
|
||||
data = json.loads(RESULT.read_text(encoding='utf-8'))
|
||||
route = float(data.get('route_accuracy_percent', 0.0) or 0.0)
|
||||
memory = float(data.get('memory_signal_accuracy_percent', 0.0) or 0.0)
|
||||
timeouts = int(data.get('timeouts', 0) or 0)
|
||||
passed = bool(data.get('pass'))
|
||||
ok = True
|
||||
reasons = []
|
||||
if route < float(cfg.get('min_route_accuracy_percent', 95.0)):
|
||||
ok = False
|
||||
reasons.append('route_accuracy_below_threshold')
|
||||
if memory < float(cfg.get('min_memory_signal_accuracy_percent', 90.0)):
|
||||
ok = False
|
||||
reasons.append('memory_accuracy_below_threshold')
|
||||
if timeouts > int(cfg.get('max_timeouts', 0)):
|
||||
ok = False
|
||||
reasons.append('timeouts_above_threshold')
|
||||
if bool(cfg.get('require_pass', True)) and not passed:
|
||||
ok = False
|
||||
reasons.append('regression_not_pass')
|
||||
print(json.dumps({
|
||||
'ok': ok,
|
||||
'decision': 'promote' if ok else 'hold',
|
||||
'reasons': reasons,
|
||||
'route_accuracy_percent': route,
|
||||
'memory_signal_accuracy_percent': memory,
|
||||
'timeouts': timeouts,
|
||||
'pass': passed,
|
||||
}, ensure_ascii=False))
|
||||
return 0 if ok else 1
|
||||
|
||||
|
||||
if __name__ == '__main__':
|
||||
raise SystemExit(main())
|
||||
Loading…
Add table
Add a link
Reference in a new issue