# Nebius Partner Ops — Take-Home Work Sample

**Role:** Manager, Partner Operations Systems and Analytics · **Author:** Dico Angelo · **Date:** 2026-04-15

## What this is

A full simulated take-home exercise for the Nebius Partner Ops Systems & Analytics role — built and executed against a synthetic-but-realistic 4-CSV dataset representing a live HubSpot→Salesforce cutover. Every artifact in `deliverables/` is production-grade (not outline).

## How to review

Start with the dashboard. Everything else supports it.

| # | File | What it shows |
|---|------|---------------|
| — | [`03-dashboard.html`](03-dashboard.html) | Leadership-grade exec dashboard. Open in browser. |
| 1 | [`01-data-model.md`](01-data-model.md) | Target SF data model (ERD at `../diagrams/ERD-salesforce-data-model.png`) |
| 2 | [`02-findings.md`](02-findings.md) | KPIs + 7 data-quality findings, ranked by $ exposure |
| 3 | [`04-governance-memo.md`](04-governance-memo.md) | 1-page cutover attribution freeze strategy |
| 4 | [`05-roadmap.md`](05-roadmap.md) | 30-60-90 mapped to 7 JD pillars + Week-1 quick-win |

## How it was built

- **`../data/generate.py`** — deterministic synthetic dataset (seed = `gh_jid 4805888101`), 4 CSVs with injected messiness representative of a live cutover
- **`../analysis/run.py`** — pure-stdlib Python analysis → `kpis.json` + `findings.json`
- **`../diagrams/`** — ERD generated via ResearchGravity refined pipeline (Planner → Stylist → Critic), compressed to webp

## Headline numbers

- **$11.59B** total pipeline · **61.3%** partner-involved · **40.3%** partner-sourced
- **$1.58B** exposed pipeline with no deal-reg (biggest finding — closed by 1 validation rule)
- **7** duplicate partner records · **48** tier-mismatched partners · **2** ghost-attribution opps
- Week-1 quick-win: validation rule blocks `Stage=Qualify` without Approved deal-reg

## The thesis

Every rule in this package is downstream of one governance gate: **`Deal_Registration__c.Origin_Flag__c`, locked at `Stage=Propose`, gated by a T-14 freeze window.** Name the gate before the rule. That's the difference between a partner-ops program that scales with 547% YoY growth and one that dies at the next parallel-run.