Sooner or later a file in the folder is broken — a half-finished upload, an export that crashed. What the loop does next is a setting, and both choices have a cost you need to know about. This level runs the 5.2 pipeline over a folder where one file is corrupt, both ways, and then adds a deliberate pause with Sleep.
Goal
Know exactly what Stop and Skip do with a broken file, and how to tell from the output that something was skipped.
You'll need
- The exercise files for this level: a
flaky/folder with the same 12 files as 5.2 — exceptsales_2025-03-07, which is a truncated Excel file. - About 20 minutes.
Steps
1. Point the loop at the flaky folder
Duplicate your 5.2 pipeline (or rebuild it) as w5-4-flaky: set the File List's Directory to
flaky/ and the CSV Write's Filename to flaky_sales.csv. Keep On Error on
Stop loop on first failure. Delete any old flaky_sales.csv, then Execute.

Iterate node failed at row 8: Connector error: Magic File detection failed:
Excel error: Zip error: invalid Zip archive: Could not find EOCD
"Row 8" is the eighth file the File List returned — not the eighth day. The message tells you
which reader gave up and why: this .xlsx isn't a complete zip archive.
Now look at flaky_sales.csv. It exists, with the rows of the files processed before the
broken one. Stop stops; it doesn't undo what earlier rounds already wrote.
2. Skip and continue
Click the Iterate and set On Error to Skip failed and continue loop.

Delete flaky_sales.csv again and Execute. The run is green.
Count what you got: 259 rows from 11 distinct source_file values — one file short of the
12 in the folder, and 29 rows missing. The Console says which round was skipped and why:
⚠ 1 of 12 rounds failed and were skipped (on error: continue):
• row 8: Connector error: Magic File detection failed: Excel error: Zip error: invalid Zip archive: …
3. Make the skip visible when nobody is watching
The Console warning is there when you run from the editor. A scheduled run has no Console, and with Skip its status is still green — so the pipeline should check itself. The simplest check is a comparison you can do by eye — files listed versus files in the output:
SELECT COUNT(DISTINCT source_file) AS files_loaded, COUNT(*) AS rows_loaded
FROM input
Point a CSV Source at flaky_sales.csv, add that SQL and a Console, and compare files_loaded
with what the File List reported. 11 against 12 is your alarm.
4. Pace the loop with Sleep
Some loops call a service that limits how often you can hit it. From
ORCHESTRATE ▸ Flow Control, drop a Sleep between the Iterate and read_file, and set it to
2 seconds.
Execute and time it: every round now waits 2 seconds before reading. Three files take about 6 seconds; the whole folder, about 24. Sleep pauses every round because it sits after the Iterate, inside the loop.
Expected result
- Stop: the run fails on the corrupt
sales_2025-03-07.xlsx; the output keeps the rows written before it. - Skip: the run succeeds with 259 rows from 11 files.
- With Sleep at 2 seconds, the run takes about 2 seconds per file.
Common pitfalls
Choosing Skip because it's green. Green with Skip means "some rounds may have failed". The editor's Console lists them; a scheduled run only has its status and row count. Use Skip together with a check like the one in step 3.
Re-running after a Stop and getting duplicates. The first run appended some rows before stopping; the second appends everything again. Delete the output first, or write each run to its own dated file.
What you built
An informed choice between failing loud and failing quiet — and a check that makes the quiet one safe. Next level puts it all together and starts deleting files, where these choices matter most.
Check yourself
One question, drawn from a larger bank. Get it right to pass.
How many rows does the Skip run load from the flaky folder?