Investigate a data discrepancy: a product tour
When two dashboards disagree, does this analyst find out why, with evidence?
You’re Ravi Shah
Candidate
Ravi Shah is asked why two dashboards show different September revenue. He works in a SQL workspace with a clock running. Then you see what the hiring team sees: the level for each skill, what the rules counted, and every step in order.
11 steps · about 2 minutes
Read the tour as text
Which number is right?
Ravi is asked why September MRR is $18,000.00 on the Billing dashboard and $30,500.00 on Finance’s. He has 40 minutes, and opening a view, asking a question or running a query each takes some of them.
Do they ask what the number means first?
Ravi reads the definitions, then asks Finance how it defines the month. Finance counts in Berlin calendar months. Both happen before his first query, and the rules count them.
Do they trace where each number comes from?
He opens the model behind the Billing number. It cuts each invoice into a month by its UTC date. Opening a model’s SQL is its own recorded step.
Do they test it against the data?
Ravi runs read-only SQL on the practice data: each invoice with its time in UTC and in Berlin. One invoice was issued late on the last night of August in UTC, which is already September in Berlin. Each query takes 2 minutes.
Do they back the cause with evidence?
Ravi runs a second query that adds up only the invoices that fall in September in Berlin but in August in UTC, then pins the claim that one boundary invoice carries the gap to it. A pin shows him no verdict, and nothing tells him whether the data confirmed it; the verdict is kept for the hiring team. The claims are neutral statements of what a query measures, not answers.
Can they reproduce the right number?
He writes the correction from the source tables, counting invoices by their Berlin date, and verifies it. In hiring the check he sees covers only the data he can see. In an assessment the correction is also checked on data he cannot see, so it has to hold for any date.
What do they send Finance?
Ravi sends the cause, the impact in cents and his confidence, with a message for Finance. The hiring team reads the message; no rule grades its wording.
What does the candidate see at the end?
In hiring, Ravi gets a submission receipt, not levels or feedback, so the answers cannot be passed on.
How is each level set?
In an assessment you publish now, rules set every level from what Ravi did, and no model reads the work. Framing is strong because he opened the definitions and asked Finance before his first query. Evidence gathering stops at meets: he confirmed the claim that explains the gap but never ruled out another explanation with a query, and strong needs both.
What did the rules count?
You see what the rules counted: the framing sources before his first query, each claim marked as a key claim or a ruled-out alternative with whether its last pin held, the correction, and the conclusion with his message. The expected cause and impact are listed beside them.
Who decides?
A person. The timeline lists every view, question, query with its SQL, pin with its verdict, correction and conclusion in plain words, and the actions a level cites are highlighted. You read it and decide.