KnowledgeForge：把已解决的 ITSM 工单变成持续进化的知识库

你是 ITSM 知识库编辑。请根据已解决事件生成候选知识文章。

必须遵守：
1. 不补充工单中没有证据支持的根因或命令。
2. 删除个人信息、凭据、客户标识和内部主机名。
3. 区分“已验证步骤”和“待确认建议”。
4. 输出 JSON，字段为 title、symptoms、scope、cause、resolution、verification、warnings。
5. 如果信息不足以形成文章，将 publishable 设为 false，并列出 missing_information。

事件内容：
{{SANITIZED_TICKET}}

import re
from dataclasses import dataclass


@dataclass
class Article:
    tenant_id: str
    title: str
    symptoms: str
    resolution: str
    verification: str


def tokens(text: str) -> set[str]:
    return set(re.findall(r"[a-z0-9_./-]+", text.lower()))


def similarity(left: Article, right: Article) -> float:
    a = tokens(f"{left.title} {left.symptoms} {left.resolution}")
    b = tokens(f"{right.title} {right.symptoms} {right.resolution}")
    return len(a & b) / len(a | b) if a or b else 0.0


def quality_score(article: Article) -> int:
    score = 0
    score += 25 if len(article.title) >= 12 else 0
    score += 25 if len(article.symptoms) >= 30 else 0
    score += 30 if len(article.resolution) >= 50 else 0
    score += 20 if len(article.verification) >= 25 else 0
    return score


def decide(candidate: Article, library: list[Article]) -> dict:
    # Tenant filtering must happen before similarity comparison.
    same_tenant = [a for a in library if a.tenant_id == candidate.tenant_id]
    nearest = max(same_tenant, key=lambda a: similarity(candidate, a), default=None)
    nearest_score = similarity(candidate, nearest) if nearest else 0.0
    quality = quality_score(candidate)

    if quality < 70:
        action = "request_more_information"
    elif nearest_score >= 0.80:
        action = "deduplicate_or_merge"
    elif nearest_score >= 0.45:
        action = "human_review"
    else:
        action = "publish_candidate"

    return {
        "tenant_id": candidate.tenant_id,
        "quality_score": quality,
        "nearest_title": nearest.title if nearest else None,
        "similarity": round(nearest_score, 3),
        "action": action,
    }


library = [
    Article(
        tenant_id="tenant-a",
        title="Restart a stalled invoice worker",
        symptoms="Invoices remain queued and the worker health check times out.",
        resolution="Drain the queue, restart the invoice worker, and confirm that new jobs are consumed.",
        verification="Submit a test invoice and confirm completed status.",
    )
]

candidate = Article(
    tenant_id="tenant-a",
    title="Recover an unresponsive invoice worker",
    symptoms="Invoice jobs stay queued while worker health checks repeatedly time out.",
    resolution="Pause intake, drain queued jobs, restart the invoice worker, and resume intake after health checks pass.",
    verification="Submit one test invoice and verify that it reaches completed status without retrying.",
)

print(decide(candidate, library))

python3 knowledge_gate.py