You inherit a nightly pipeline. The system that feeds it changed its export last week, and nobody told you. This level hands you that pipeline and that export. It has three problems, and they show up one at a time — each fix reveals the next. The last one doesn't fail at all.
Goal
w7-4-nightly-orders running green and writing the right data: one row per day, 365 rows,
2,682 paid orders in total.
You'll need
- The exercise files for this level:
w7-4-nightly-orders.json(the pipeline) andorders_export.csv(the new export). - Levels 7.1–7.3: Monitor, a schedule, an alert with a min rows threshold.
- About 25 minutes.
The pipeline reads the export, counts paid orders and units per day in a SQL node, and writes
paid_per_day.csv.
Steps
1. Import, schedule, subscribe
Import w7-4-nightly-orders.json (right-click MY PIPELINES ▸ Import Pipelines, as in 1.2).
Schedule it daily, and subscribe yourself twice: Failures only, and Custom thresholds… with
min rows 1. Then ▶ Run now — and treat what follows as a night on call.
2. Problem one
The run fails. The error, in Monitor under the name and in your email:
Connector error: CSV file not found: /replace/with/your/folder/orders_export.csv
A path that only existed on someone else's machine. Fix both paths — the source's File Path and the target's output — and run again.
3. Problem two
It fails again, differently:

Schema error: No field named status.
Valid fields are input."order_id;customer_id;sku;quantity;order_date;status".
Read the "valid fields": there's exactly one, and its name is the whole header line with semicolons
in it. That's World 2's wrong-delimiter signature, seen from a SQL node. The export switched from ,
to ;. Set the source's Delimiter to ; and run again.
4. Problem three — the green one
The run succeeds. Monitor shows Completed. No failure email.
But the min rows alert fires:

Rows written: 0. Open paid_per_day.csv: a header and nothing else. Nothing crashed; the SQL simply
matched no rows. Preview the source and look at the status column: Paid, Pending, Refunded —
the new export capitalizes them. WHERE status = 'paid' compares text exactly, so it matches nothing.
Make the comparison robust to the next surprise, not just this one:
WHERE LOWER(status) = 'paid'
Run again.
Expected result
- The run completes, and neither alert fires.
paid_per_day.csvhas 365 rows (one per day of 2025) plus the header, starting2025-01-01, 6, 10.paid_ordersadds up to 2,682 — the same paid-order count you've seen since World 3.
Common pitfalls
Fixing problem two by editing the CSV.
The export will arrive with ; again tomorrow. Fix the pipeline to read what the source actually
sends.
Stopping at green. Problem three is the most common real-world failure there is: a changed value, a filter that matches nothing, a green run. Without the min rows alert, you'd find out when someone asks why the report is empty.
LOWER everywhere.
It's the right fix for a status code. For names or free text, think before you fold case — LOWER
makes Cardoso and cardoso the same person.
What you built
A method more than a pipeline: read the error, not just the status; let the error's details point at the cause; and don't trust a green run until the output agrees. That's World 7 — pipelines you operate, not just pipelines you build.
Check yourself
One question, drawn from a larger bank. Get it right to pass.
Which alert catches the third problem?