Fix where the brief and the exercise data disagree

[santiaro90]

A dry run through the exercise — fresh clone, Docker up, init.sh, then actually solving both challenges — turned up five places where the brief promises something the repo doesn't contain, and one thing candidates are never told: that they have 60 minutes.

Nothing here makes the exercise easier. Two of the changes make it harder.

What was off, and what changed

Gap	Change
Session length never stated	60 minutes, in the intro, and the "you won't finish" line now says so because of it
"thousands/decimal separators" — no grouping separator existed in any amount	mo_18 is now "1.050,00 kr". A parser that strips commas produces 1.05000 and fails loudly instead of silently returning the wrong number
"Some real people plausibly exist in both systems" — exactly one exact-match email, nothing to weigh	Two cells changed: one Meridian address now matches a TTG account only after case and padding are normalized, and one address is now shared by two people who are obviously not the same person
customer_id described as TTG's identity, but only pages carries it and the money keys on account_id	Said plainly, in both the README bullet and sources.yml
bin/ingest mid-session was unannounced, unlike bin/add-partner	Announced, in the same shape: your interviewer will run it and ask you to rebuild
sources.yml listed USD as a Meridian currency; no USD order exists	Corrected, and stated as a fact about Meridian's trade

Two smaller ones: vendor product names are gone from the source descriptions, since the exercise shouldn't imply anyone's stack; and dbt deps no longer leaves an untracked dbt_packages/ in the candidate's tree mid-session.

What I deliberately didn't do

Explain the trap in bin/ingest. The incremental batches turn the merge log into a chain and re-send a row that already exists. Single-hop resolution silently resolves people to the wrong id, and a duplicate breaks uniqueness. That is the sharpest thing in Challenge 1 and it stays unexplained — the README says the step is coming and sources.yml says the log is "appended to over time, like any other source", which is what a real engineer would get. Working out what that implies is the exercise.

Same reasoning on identity: sources.yml now says an address in both systems is "a hint, not a key", and stops there. The candidate finds the shared address themselves or they don't.

Add rows or tables. The dataset is small enough to read end to end inside the session, which is most of why the exercise works. Three cells changed; nothing grew.

Verified

Loaded the edited data and built a full working solution on top of it — staging, currency conversion off the seed, facts at both grains, identity dimension, tests. 42 nodes green.

• 1.050,00 kr parses to 1050.00; the naive strip-commas version errors rather than returning 105000.
• All three identity shapes now appear: the same person twice inside Meridian, a cross-system match that only resolves after normalization, and one address held by two different people.
• bin/add-partner mid-session plus a rebuild on an untouched tree still picks up the new brand, so Challenge 2's acceptance criterion is unaffected.

[Copilot]

Pull request overview

Aligns the exercise brief with its fixtures and runtime behavior.

Changes:

• Documents the 60-minute limit and mid-session ingestion.
• Clarifies identity, currency, and source semantics.
• Adds representative identity and currency edge cases.

Reviewed changes

Copilot reviewed 4 out of 5 changed files in this pull request and generated 1 comment.

Show a summary per file

File	Description
README.md	Updates exercise guidance and data examples.
dbt/models/sources.yml	Clarifies source semantics and currencies.
data/initial/meridian_orders.csv	Adds European grouping separators.
data/initial/meridian_customers.csv	Adds cross-system identity edge cases.
.gitignore	Ignores generated dbt artifacts.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.