GPT Direction — Governance Input Control + Incremental Candidate Scan Addendum (2026-06-01)
GPT Direction — Governance Input Control + Incremental Candidate Scan Addendum
Date: 2026-06-01 Reviewer: GPT Council
Decision
The current One-Roof Governance technical design must be extended. It is not enough to classify objects already visible in PG. The input pipeline must also be governed so bad, incomplete, stale, duplicated, or unregistered inputs do not break later classification.
Addendum required in next macro
Add a new technical-design branch covering both:
- Incremental Governance Candidate Scan
- Governance Input Control / Input Quality Gate
Key principle
Do not modify the existing birth/khai-sinh gate unless proven necessary. Keep birth stable. Add independent post-birth candidate/invalidation logic and pre-classification input-quality controls.
Missing design topics to add
- Candidate layer between Birth/Registry and Governance Coverage.
- Dirty group / dirty source invalidation.
- Source snapshot + ruleset version tracking.
- Group_key design.
- Event-driven scan + incremental dirty-group scan + periodic full audit.
- Count>1 as candidacy trigger, not mandate.
- Birth-orphan vs governance-orphan precedence.
- Input completeness/quality gate.
- Input provenance and source trust level.
- Duplicate/conflict/stale-input detection.
- Schema/contract validation before classification.
- Registry visibility requirement before governance coverage scan.
- Late-arriving data and correction handling.
- Backfill/replay path.
- Anti-spam/coalescing/resource-budget controls.
- Fail-closed production/execution gate when candidate/coverage status is stale or unknown.
Rationale
If input is not controlled, classification can be technically correct but systemically wrong. The scanner must know what it is allowed to classify, from which source snapshot, under which ruleset, and whether the upstream input is complete and trustworthy enough.
Next action
When drafting the next agent prompt, include this direction and require Agent to audit the input pipeline broadly, not just design scanner output classification.