Forward-deployed / AI Knowledge Graph
Flowable

Decision governance — the per-table checklist

Fill one card per deployed decision table. A table that can’t fill this card shouldn’t be in production.

Card: <table key>

1. Ownership

  • [ ] Accountable owner (one name): ____________
  • [ ] UNIQUE-violation / no-match alerts route to the owner, not ops
  • [ ] Backup owner named for leave/attrition

2. Change ritual

  • [ ] Changes proposed as table diffs (row-level), never console edits
  • [ ] Approval: owner + one reviewer minimum; committee sign-off where mandated
  • [ ] Deployed as a new version via the same CI as process models
  • [ ] Production consoles are read-only for everyone

3. Validation (gates the deploy)

  • [ ] Golden cases file exists and passes (input → expected output pairs)
  • [ ] Structural checks: overlap (UNIQUE), completeness, cell types
  • [ ] Impact replay: last N days’ real inputs through old vs new version, diff reviewed (“flips X% of cases from A to B”)

4. Audit

  • [ ] History records the table version behind every decision (retention per Phase 9 policy)
  • [ ] Regulated flows pin in-flight instances to their starting version (Phase 8)
  • [ ] Approved-policy document ↔ deployed-version mapping is maintained

Scope check (before any of the above)

  • [ ] This is policy the business tunes (threshold/band/grid/eligibility), not an algorithm, validation, or anything needing loops, calls, or state — those stay in code.