ADR 0006: Data Quality and Governance for Knowledge Layer

Proposed Status: Proposed Date: 2026-05-06 Domain: information Level: system Authors: Genesis
data-qualitygovernanceknowledge-layerdata-stewardship

Status

Proposed

Context

ADR 0004 (LLM Wiki) and ADR 0005 (RAG and Agent Memory) establish three knowledge patterns for the Integral stack.

The article identifies a critical gap that most teams underinvest in:

> "The governance layer underneath all of them — data quality, freshness, access control — is what most teams underinvest in. > Stale or ungoverned inputs degrade all three simultaneously."

This ADR establishes the governance requirements that apply across all three knowledge patterns.

Decision

Implement four governance requirements across the knowledge layer.

Requirement 1: Source Provenance

Required for all wiki pages. Every concept page must cite at least one source page. No synthesized claim without a backing source. wiki lint enforces this. Pages with missing sourceIds fail validation.

Requirement 2: Freshness Tracking

Required for all concept pages. Every page carries updatedAt. It is checked by wiki lint for staleness. Pages older than 30 days without update trigger a stale-page warning.

Requirement 3: Contradiction Detection

Enforced by wiki lint. When two concept pages make conflicting claims, wiki lint surfaces the contradiction. A human must adjudicate. The resolution gets recorded as a new claim with status: resolved and evidence pointing to the adjudication source.

Requirement 4: Ingest Quality Gates

Required before wiki commit. Before any source page is committed to the vault:

  • Source must be cited in at least one concept page.
  • Concept page must pass wiki lint with 0 errors.
  • Source must have valid provenance frontmatter.
Source URL or file path.

Options Considered

OptionAssessment
Trust LLM ingest qualityRejected — errors compound in LLM Wiki.

No quality gate means bad synthesis baked permanently. |

Manual review every ingestRejected — unsustainable at scale.
Creates bottleneck. |
Automated lint + provenance + staleness checksAdopted — wiki lint enforces all four requirements.

Positive

  • All three knowledge patterns share the same governance foundation.
  • Quality degradation caught before propagation.
  • Clear audit trail: source → concept → claim → evidence.
  • Wikilink integrity maintained via wiki lint.

Negative

  • Ingest workflow is slower.
Requires lint plus provenance before commit.
  • Contradiction adjudication requires human time.

Risks

  • If wiki lint is not run after every ingest, governance requirements slip.
  • Staleness warnings can pile up if vault is not actively maintained.

Implementation

These requirements are enforced by openclaw wiki lint in the standard workflow:

openclaw wiki ingest 
openclaw wiki compile
openclaw wiki lint  # fails if: missing sourceIds, missing provenance, stale page, contradiction

Wiki vault mode isolated means all governance is self-contained. No external dependency.

References

  • Visrow (2026). RAG vs. Agent Memory vs. LLM Wiki.
  • ADR 0004 — LLM Wiki Pattern for Integral Knowledge Commons (M10)
  • ADR 0005 — RAG and Agent Memory in the Integral Stack