Capstone: Present & certify

Final step: present the foundation the way a senior data engineer presents to a mixed technical/business audience — not a code tour, a decision brief: what Harbor Lane can now trust, what it cost, what's watched, what's still gray.

The presentation (10 minutes, five beats)

1. The before picture — three concrete wrong numbers the old setup produced (inflated customers, understated repeat rate, ambiguous revenue), each with its dollar-or-decision consequence.
2. The architecture in one diagram — raw → staging → marts → metrics, with the gate and the resolver drawn as first-class components, not afterthoughts.
3. The trust story — what is guaranteed now: idempotent loads, the check families and their SLAs, golden identity, certified definitions. Guarantees, not features.
4. The AI readiness claim — the specific sentence this course has been building toward: 'An AI agent pointed at this warehouse reads honest schemas, queries resolved entities, and computes only certified definitions.' Say what would still make you nervous, too.
5. The roadmap — the three things you'd do next quarter (e.g., review-queue for gray-zone matches, hourly freshness on orders, dim_date when weekday analysis becomes routine) with a one-line justification each.

Certification rubric

• Design doc (25%) — grain declarations correct; idempotency pattern justified; wholesale/consumer separation reasoned; edge-case rulings recorded, not dodged.
• Build (35%) — the five-command sequence runs green from scratch; re-runs are idempotent; the three planted incidents were caught by the gate, named, and repaired; no regression in pre-capstone metrics.
• Validation & docs (25%) — ten questions answered with correct numbers and cited definitions; LINEAGE.md and QUALITY.md complete and honest.
• Presentation (15%) — a decision brief, not a demo; the before/after deltas quantified; limits stated plainly.

Passing earns the Certified AI Data Foundations Specialist credential (ID format EDOVA-DF-2026-XXXX, independently verifiable at edova.ai/verify). It certifies the full loop: modeling data for AI, moving it idempotently, defending its quality, resolving its entities, and governing its definitions.