#The One-Page Cheatsheet

The page to read in the car park. Nothing new here — only what you most want loaded when you walk in.

#Posture

You are a consultant meeting a client, not a candidate being examined. Ask before you answer. Decide, then say what it costs. Be the most honest person in the room.

#The four frameworks

Design — CLAIM: Context (ask 3–5 questions) → Levers (the 2–3 decisions that decide it) → Architecture (draw) → Implications (what you gave up) → Measure (how you'd know).

Debug — Symptom → Scope → Evidence → Hypothesis → Test → Fix → Prevent. Never propose a fix before naming the evidence.

Behavioural — STAR-L: keep S and T short, most airtime on A ("I decided…"), R has a number, L is honest.

Objections — Feel → Felt → Found → Fact. Concede what's true first. The fact is mandatory.

#Opening moves

• Any design question: "Before I design anything — what's the freshness requirement, who operates this in a year, and what's the regulatory position?"
• Any performance question: "First: what changed? Then: where is the time going?"
• Any comparison question: concede what's genuinely true about the competitor, name the dimension that decides it for this customer, recommend, offer coexistence.
• Anything you don't know: "I don't know that specifically. Here's how I'd reason about it, and here's how I'd verify." Then reason well and move on.

#The numbers you'll reach for

State these as heuristics. False precision is worse than "order of magnitude, roughly X".

#Six things that signal real experience

1. "Before I tune anything — what changed?"
2. "Max versus median task time in the slowest stage."
3. "I'd pin the data with time travel and re-run, to separate a data problem from a code problem."
4. "Adding nodes to a skewed job makes it more expensive and exactly as slow."
5. "The fix is one thing; the monitor that catches it next time is the deliverable."
6. "Governance that can be bypassed isn't governance."

#Six things not to do

1. Design before asking.
2. List features instead of naming a decision and its trade-off.
3. Bluff. (The only disqualifying failure.)
4. Use old product names — DLT, Workflows, online tables.
5. Defend a wrong answer. Say "you're right, I had that backwards" and move on.
6. Talk for eleven minutes without a breath.

#Currency, in one line each

Lakehouse//RT / Reyden — millisecond SQL on governed Delta/Iceberg, no serving-store copy. Beta, read-only. Lakebase / LTAP — serverless Postgres on lake storage; operational writes landing directly in open columnar formats, removing the CDC pipeline. LTAP announced. Iceberg v3 GA — deletion vectors, row lineage, VARIANT; Delta and Iceberg can share Parquet files. Unity AI Gateway — runtime governance of models/agents/tools: hard spend caps, routing, action policies, guardrails, tracing. Unity Catalog Metrics / Domains / Glossary — the governed semantic layer; the fix for inconsistent KPIs and the substrate agents need. Omnigent — open-source meta-harness above agent frameworks. Genie One / Ontology — agentic analytics with a continuously-built enterprise context layer.

Say them in service of an architecture point. Never recite the list.

#The three sentences to have ready

Your thesis on AI:

"Everyone has the same models. The differentiator is governed context and runtime control — which is a data architecture problem, not a model problem."

Your thesis on cost:

"I don't want to make the platform cheaper, I want to make it attributable. Once spend is tagged to product teams, they optimise without anyone policing them."

Your thesis on the role:

"The job is making architecture repeatable across engagements, so the tenth customer gets the benefit of the first nine — while staying close enough to the metal to be credible when something is on fire."

#Last things

• Ask your questions. Having none is a scored failure.
• Close with: "Is there anything about my background that gives you pause? I'd rather address it now."
• Breathe. You have done the work.