AgentSkillsCN

quality-backfill-outcomes

根据notion_status_events推断已推送信号的TP/FP标签，并将其存储于signal_quality_metrics中。

SKILL.md
--- frontmatter
name: quality-backfill-outcomes
description: Infer TP/FP labels for pushed signals from notion_status_events and store
  them in signal_quality_metrics.
user-invocable: true
allowed-tools:
- Bash
- Read

quality-backfill-outcomes

When to use

  • •You want automatic TP/FP labels from CRM outcomes (Passed/Funded) within a time window.
  • •You need to populate labels before running FP pattern mining.

Inputs

  • •signals DB path.
  • •days_to_count window (default 30 days from push).
  • •Optional: since_days to limit scanning.

Workflow

  1. •Ensure you have status events (run quality sync-status-events at least once).
  2. •Run outcome backfill: python -m ops.cli quality backfill-outcomes --days-to-count 30.
  3. •Validate: sample a few signals and confirm label_source='notion_status_event' and notion_status matches the triggering outcome.

Outputs

  • •JSON summary: scanned, labeled, fp, tp, skipped_no_events.

Guardrails

  • •Manual labels are preserved unless --override-manual is set.
  • •This is conservative by default: only Passed => FP and Funded => TP (not Source).

References

  • •references/reference.md
  • •docs/QUALITY_OPS_ARCHITECTURE.md