Spaces:
Sleeping
Sleeping
File size: 8,925 Bytes
fa1fead | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 | """
evaluation/deployment_gate.py
WHY THIS EXISTS: Prevents deploying a SmartDocs version that fails Hindi users.
Reads the evaluation_report.json produced by ragas_evaluator.py and enforces
hard thresholds. Exit code 1 blocks the Railway / CI deploy pipeline.
Usage in CI:
python evaluation/deployment_gate.py
# Exit 0 = deploy approved
# Exit 1 = deploy blocked (reason printed)
LAW 16: faithfulness_hindi < 0.85 OR language_accuracy < 0.92 → BLOCK DEPLOYMENT.
"""
from __future__ import annotations
import json
import logging
import sys
from dataclasses import dataclass
from datetime import datetime, timezone
from pathlib import Path
from typing import Any
logger = logging.getLogger(__name__)
# ---------------------------------------------------------------------------
# Deployment thresholds — single source of truth
# ---------------------------------------------------------------------------
THRESHOLDS: dict[str, tuple[str, float, str]] = {
# (metric_path, minimum_value, human_readable_name)
"faithfulness_en": ("english.faithfulness", 0.90, "English faithfulness"),
"faithfulness_hi": ("hindi.faithfulness", 0.85, "Hindi faithfulness"),
"relevancy_en": ("english.answer_relevancy", 0.85, "English answer relevancy"),
"relevancy_hi": ("hindi.answer_relevancy", 0.83, "Hindi answer relevancy"),
"precision_en": ("english.context_precision", 0.80, "English context precision"),
"precision_hi": ("hindi.context_precision", 0.80, "Hindi context precision"),
"recall_en": ("english.context_recall", 0.75, "English context recall"),
"recall_hi": ("hindi.context_recall", 0.75, "Hindi context recall"),
"hallucination_rate": ("hallucination_rate", 0.10, "Hallucination rate", True), # lower is better
"faithfulness_ratio": ("faithfulness_ratio", 0.97, "Hindi/English faithfulness ratio"),
}
# Metrics where LOWER is better (inverted threshold check)
LOWER_IS_BETTER = {"hallucination_rate"}
# ---------------------------------------------------------------------------
# Gate result dataclass
# ---------------------------------------------------------------------------
@dataclass
class GateResult:
approved: bool
blocking_failures: list[dict] # [{"metric": str, "value": float, "threshold": float}]
warnings: list[str]
report_timestamp: str
evaluated_at: str
def print_report(self) -> None:
sep = "=" * 64
print(f"\n{sep}")
print("SMARTDOCS DEPLOYMENT GATE")
print(f"Report timestamp: {self.report_timestamp}")
print(f"Evaluated at: {self.evaluated_at}")
print(sep)
if self.approved:
print("\n✅ DEPLOYMENT APPROVED\n")
print(" All metrics above threshold.")
else:
print("\n🚫 DEPLOYMENT BLOCKED\n")
for failure in self.blocking_failures:
direction = ">" if failure["metric"] not in LOWER_IS_BETTER else "<"
print(
f" âœ-- {failure['name']:<35} "
f"actual={failure['value']:.4f} "
f"threshold{direction}{failure['threshold']:.4f}"
)
if self.warnings:
print("\n Warnings:")
for w in self.warnings:
print(f" âš {w}")
print(sep)
def to_dict(self) -> dict[str, Any]:
return {
"approved": self.approved,
"blocking_failures": self.blocking_failures,
"warnings": self.warnings,
"report_timestamp": self.report_timestamp,
"evaluated_at": self.evaluated_at,
}
# ---------------------------------------------------------------------------
# Helper: safe nested key access
# ---------------------------------------------------------------------------
def _get_nested(data: dict, dotted_path: str) -> float | None:
"""
Resolves a dot-separated key path against a nested dict.
Returns None if any key is missing.
Example: _get_nested(report, "english.faithfulness") → 0.91
"""
parts = dotted_path.split(".")
obj: Any = data
for part in parts:
if not isinstance(obj, dict) or part not in obj:
return None
obj = obj[part]
try:
return float(obj)
except (TypeError, ValueError):
return None
# ---------------------------------------------------------------------------
# Gate logic
# ---------------------------------------------------------------------------
def check_gate(report: dict) -> GateResult:
"""
Applies all deployment thresholds to an evaluation report dict.
Args:
report: Dict produced by EvaluationReport.to_dict() in ragas_evaluator.py.
Returns:
GateResult with approved=True if all thresholds pass.
"""
blocking_failures: list[dict] = []
warnings: list[str] = []
for metric_key, threshold_spec in THRESHOLDS.items():
# Handle both 4-tuple (with lower_is_better flag) and 3-tuple
if len(threshold_spec) == 4:
path, threshold, name, _ = threshold_spec
else:
path, threshold, name = threshold_spec[:3]
value = _get_nested(report, path)
if value is None:
warnings.append(f"Metric '{metric_key}' not found in report — skipped")
continue
if metric_key in LOWER_IS_BETTER:
failed = value >= threshold
else:
failed = value < threshold
if failed:
blocking_failures.append({
"metric": metric_key,
"name": name,
"path": path,
"value": value,
"threshold": threshold,
})
# Special check: if both en and hi faithfulness are present, assert ratio
en_faith = _get_nested(report, "english.faithfulness")
hi_faith = _get_nested(report, "hindi.faithfulness")
if en_faith and hi_faith and en_faith > 0:
ratio = hi_faith / en_faith
if ratio < 0.97:
blocking_failures.append({
"metric": "faithfulness_ratio_computed",
"name": "Computed Hindi/English faithfulness ratio",
"path": "computed",
"value": ratio,
"threshold": 0.97,
})
return GateResult(
approved=len(blocking_failures) == 0,
blocking_failures=blocking_failures,
warnings=warnings,
report_timestamp=report.get("timestamp", "unknown"),
evaluated_at=datetime.now(timezone.utc).isoformat(),
)
def load_and_check(report_path: Path) -> GateResult:
"""
Loads evaluation report from disk and runs the gate check.
Args:
report_path: Path to evaluation_report.json.
Returns:
GateResult.
Raises:
FileNotFoundError: If report does not exist.
json.JSONDecodeError: If report is malformed.
"""
if not report_path.exists():
raise FileNotFoundError(
f"Evaluation report not found at {report_path}. "
"Run: python evaluation/ragas_evaluator.py first."
)
with open(report_path, "r", encoding="utf-8") as f:
report = json.load(f)
return check_gate(report)
def write_gate_result(result: GateResult, output_path: Path) -> None:
"""Writes GateResult to JSON for CI artifact storage."""
output_path.parent.mkdir(parents=True, exist_ok=True)
with open(output_path, "w", encoding="utf-8") as f:
json.dump(result.to_dict(), f, indent=2)
logger.info("Gate result written to %s", output_path)
# ---------------------------------------------------------------------------
# CLI entry point
# ---------------------------------------------------------------------------
def main() -> int:
"""
CLI entry point.
Exit code 0 = deployment approved.
Exit code 1 = deployment blocked.
Exit code 2 = report not found (treat as blocked).
"""
logging.basicConfig(
level=logging.INFO,
format="%(asctime)s %(levelname)s — %(message)s",
stream=sys.stdout,
)
report_path = Path(__file__).parent / "evaluation_report.json"
gate_output_path = Path(__file__).parent / "gate_result.json"
try:
result = load_and_check(report_path)
except FileNotFoundError as exc:
print(f"\n🚫 DEPLOYMENT BLOCKED — {exc}\n")
return 2
except json.JSONDecodeError as exc:
print(f"\n🚫 DEPLOYMENT BLOCKED — Malformed evaluation report: {exc}\n")
return 2
result.print_report()
write_gate_result(result, gate_output_path)
return 0 if result.approved else 1
if __name__ == "__main__":
sys.exit(main())
|