Data Governance

The data governance is where the data comes from, where it goes, and who decides (L372) — the lineage (the data's path: the source L372, the transformations L372, the destinations L372), the quality (the data's fitness: the completeness L372, the accuracy L372, the freshness L335, L372), and the ownership (who decides: the data owner L372, the steward L372, the access L262, L372) — the enterprise's (L380) data (L372), governed (L372).

The one-sentence interview answer: "The data governance is the data's rulebook (L372). The lineage (L372): the data's path (L372) — the source (L372): where it comes from (L372); the transformations (L372): what happens to it (L372); and the destinations (L372): where it goes (L372) — the RAG (L349), the training (L365), the analytics (L372). The quality (L372): the data's fitness (L372) — the completeness (L372), the accuracy (L372), and the freshness (L335) — the golden set's (L342) quality (L372). The ownership (L372): who decides (L372) — the data owner (L372): the business (L360) accountable (L372); the steward (L372): the day-to-day (L372); and the access (L262): the least privilege (L314). The AI's data (L372): the training (L365) — the consent (L312) and the provenance (L372); the RAG (L349) — the sources (L316) and the quality (L338); and the prompts (L312) — the PII (L313) and the retention (L322). The AI shape (L173): the enterprise (L380) — the data (L313) governed (L372): the lineage (L372), the quality (L372), and the ownership (L372) — the AI's (L173) data (L372), rulebook'd (L372)."

Think of the data governance as the library's collection policy. The policy (the governance, L372) governs the books (the data, L313): the provenance (the lineage, L372) — where each book (L372) came from (the source, L372) and where it's shelved (the destination, L372); the condition (the quality, L372) — the pages (the completeness, L372) and the printings (the freshness, L335); and the curators (the ownership, L372) — who decides (L372) what's acquired (L372) and who may borrow (the access, L262). The library works because the provenance is recorded, the condition is checked, and the curators decide (L372).

   the collection policy (the governance, L372)
   ┌────────────────────────────────────────────────────────┐
   │ the provenance (the lineage, L372) — the sources, the  │
   │ shelves (L372)                                         │
   │ the condition (the quality, L372) — the pages, the     │
   │ printings (L335)                                       │
   │ the curators (the ownership, L372) — the access        │
   │ (L262)                                                 │
   └────────────────────────────────────────────────────────┘

The mental model is the collection policy: the provenance, the condition, and the curators (L372).

   the source (L372)
        │  e.g. the help center (L265)
        ▼
   ┌────────────────────── THE LINEAGE (L372) ──────────────────────────┐
   │  the transformations (L372): the parse (L177), the chunk (L178)   │
   │  the destinations (L372): the RAG (L349), the training (L365)     │
   └──────────────────────────┬───────────────────────────────────────┘
                              ▼
   ┌────────────────────── THE QUALITY (L372) ──────────────────────────┐
   │  the completeness (L372) · the accuracy (L372) · the freshness    │
   │  (L335) — the golden set (L342)                                   │
   └──────────────────────────┬───────────────────────────────────────┘
                              ▼
   ┌────────────────────── THE OWNERSHIP (L372) ────────────────────────┐
   │  the data owner (L372) · the steward (L372) · the access (L262)   │
   └──────────────────────────────────────────────────────────────────┘

The flow is the dataset: source → lineage → quality → ownership (L372).

• The lineage (L372). The data's path (L372): the source (L372), the transformations (L372), the destinations (L372).
• The quality (L372). The data's fitness (L372): the completeness (L372), the accuracy (L372), the freshness (L335).
• The ownership (L372). Who decides (L372): the data owner (L372), the steward (L372), the access (L262).
• The AI's data (L372). The training (L365), the RAG (L349), the prompts (L312).

• An enterprise engagement (L379). The data (L313) governed (L372) — the lineage (L372) and the ownership (L372).
• A RAG platform (L349). The sources (L316) — the lineage (L372) and the quality (L338).
• A training pipeline (L365). The dataset (L372) — the provenance (L372) and the consent (L312).
• A regulated AI (L371). The GDPR (L371) — the lineage (L372) and the deletion (L312).
• Anything enterprise (L380). The rulebook (L372) — the lineage, the quality, the ownership (L372).

The through-line: the rulebook is the data's — the lineage, the quality, and the ownership (L372).

Say it in four moves:

1. The lineage. "The data's path — the source, the transformations, the destinations (L372)."
2. The quality. "The completeness, the accuracy, the freshness (L372)."
3. The ownership. "Who decides — the owner, the steward, the access (L372)."
4. The AI's. "The training (L365), the RAG (L349), the prompts (L312)."

• The lineage is the trust (L372). The data's path (L372) — the provenance (L372) — the audit's (L322) and the compliance's (L371) evidence (L372).
• The quality is the AI's (L372). The golden set (L342) — the retrieval (L338) and the groundedness (L337) — the data's (L372) fitness (L372).
• The ownership is the accountability (L372). The data owner (L372) — the business (L360) decides (L372) — the steward (L372) maintains (L372).
• The access is the least privilege (L262). The data (L313) — the L262 scope (L262) — the isolation (L320), data-shaped (L372).
• The consent is the training's (L312). The user's (L162) data (L313) — the training (L365) — the consent (L312) and the provenance (L372).

• The un-governed data (L372). The data (L313) used (L372) without the lineage (L372) — the provenance (L372) unknown (L372).
• The quality-blind (L372). The golden set (L342) without the quality (L372) — the retrieval (L338) and the groundedness (L337) suffer (L372).
• The owner-less (L372). The data (L313) without the owner (L372) — the decisions (L372) unmade (L372).
• The wide access (L262). The data (L313) open (L262) — the isolation (L320) — the least privilege (L314).
• The consent-less training (L365). The user's (L162) data (L313) fine-tuned (L365) — the consent (L312) missing (L372).

• Record the lineage (L372) — the source, the transformations, the destinations (L372).
• Measure the quality (L372) — the completeness, the accuracy, the freshness (L335).
• Assign the ownership (L372) — the owner (L372) and the steward (L372).
• Scope the access (L262) — the least privilege (L314).
• Consent the training (L312) — the provenance (L372).

Q: Walk me through the data governance.

A: The data's rulebook (L372). The lineage — the source, the transformations, the destinations (L372). The quality — the completeness, the accuracy, the freshness (L372). The ownership — who decides (L372). And the AI's — the training (L365), the RAG (L349), the prompts (L312).

Q: What's the lineage?

A: The data's path (L372): the source (L372) — where it comes from; the transformations (L372) — the parse (L177), the chunk (L178), the embed (L181); and the destinations (L372) — the RAG (L349), the training (L365), the analytics (L372). The lineage (L372) is the provenance (L372) — the audit's (L322) and the compliance's (L371) evidence (L372).

Q: How does the data's quality affect the AI?

A: Directly (L372): the AI's (L173) quality (L341) is the data's (L372) quality (L372) — the golden set (L342) built from the governed (L372) data (L372); the retrieval (L338) and the groundedness (L337) on the quality (L372); and the training's (L365) output on the data's (L372) fitness (L372). The governance (L372) is the AI's (L173) foundation (L372).

Q: Who decides?

A: The ownership (L372): the data owner (L372) — the business (L360) accountable for the data's (L313) use (L372); the steward (L372) — the day-to-day quality (L372); and the access (L262) — the least privilege (L314) on the data (L313). The decisions (L372) — what's collected (L372), what's used (L372), what's deleted (L312) — made (L372).

• What's the lineage (L372)?
• How does the data's quality affect the AI (L372)?
• Who decides (L372)?
• What's the ownership (L372)?
• What's the consent (L312)?

The senior read: the right column is the AI's foundation (L372).

// The data governance (L372) — the lineage, the quality, the ownership (L372).
// 1 · THE LINEAGE (L372) — the data's path (L372).
const lineage = {
  dataset: 'help-center-docs',
  source:  's3://help-center/',               // the source (L372, L265)
  transforms: [
    { step: 'parse',   by: 'ingestion-worker' },   // L177
    { step: 'chunk',   by: 'ingestion-worker' },   // L178
    { step: 'embed',   by: 'embedding-job' },      // L181
  ],
  destinations: ['rag:help-base', 'analytics'],    // L349, L372
};

// 2 · THE QUALITY (L372) — the fitness (L372).
const quality = {
  completeness: 0.98,                       // the pages (L372)
  accuracy:     0.95,                        // the verified (L372)
  freshness:    'daily',                     // the update (L335)
};

// 3 · THE OWNERSHIP (L372) — who decides (L372).
const ownership = {
  owner:   'support-org',                    // the business (L360)
  steward: 'data-team',                      // the day-to-day (L372)
  access:  ['read:rag', 'write:ingestion'],  // the least privilege (L262)
};

// 4 · THE AUDIT (L322) — the lineage recorded (L372).
await audit.log({ dataset, lineage, quality, at: Date.now() });  // L322

What the reader must SEE — the rulebook, applied:

  source + transforms + destinations → the lineage (L372)
  completeness + accuracy + freshness → the quality (L335)
  owner + steward + access            → the ownership (L262)
  audit.log the lineage               → the evidence (L322)

  The lineage, the quality, the ownership (L372).

• The lineage is the audit's (L372). The recorded path (L372) — the compliance's (L371) evidence (L372).
• The quality is the AI's (L372). The golden set (L342) — the retrieval (L338) and the groundedness (L337) (L372).
• The ownership is the decision's (L372). The owner (L372) — the changes (L372) decided (L372).
• The access is the security's (L262). The least privilege (L314) — the isolation (L320), data-shaped (L372).

• The data governance = the data's rulebook (L372): the lineage, the quality, the ownership.
• The lineage: the source, the transformations, the destinations (L372).
• The quality: the completeness, the accuracy, the freshness (L335).
• The ownership: the owner (L372), the steward (L372), the access (L262).
• The AI's: the training (L365), the RAG (L349), the prompts (L312).

DATA GOVERNANCE = where the data comes from, where it goes, who decides

THE LINEAGE (L372)
  the source (L372) · the transformations (L372) — the parse (L177),
  the chunk (L178), the embed (L181)
  the destinations (L372) — the RAG (L349), the training (L365),
  the analytics (L372)
  the provenance (L372) — the audit's (L322) evidence (L372)

THE QUALITY (L372)
  the completeness (L372) · the accuracy (L372)
  the freshness (L335) — the golden set's (L342) quality (L372)

THE OWNERSHIP (L372)
  the data owner (L372) — the business (L360) accountable (L372)
  the steward (L372) — the day-to-day (L372)
  the access (L262) — the least privilege (L314)

THE AI'S (L372)
  the training (L365) — the consent (L312), the provenance (L372)
  the RAG (L349) — the sources (L316), the quality (L338)
  the prompts (L312) — the PII (L313), the retention (L322)

INTERVIEW, 4 MOVES
  1 lineage  "the source, the transformations, the destinations (L372)"
  2 quality  "the completeness, the accuracy, the freshness (L372)"
  3 ownership "the owner, the steward, the access (L372)"
  4 the AI's "the training, the RAG, the prompts (L372)"

You now hold the rulebook: the lineage, the quality, and the ownership — with the provenance recorded and the curators deciding. The library's collection is governed — and every book has its path (L372).

Next: the policies, the review boards, and the accountability for the model behavior — AI Governance (L373).