100 Day of MLOps: 47 Day 47: Debug a Failing Great Expectations Checkpoint

Shadow · June 28, 2026, 6:07am

I’m working on a Great Expectations (GX) lab task. I updated the fraud_schema suite and fixed the failing rule by changing the amount lower bound to handle negative values in the drifted dataset.

I ran the script:

python3 /root/code/dataquality/fix_drift.py

The output shows:

Checkpoint drift_check result: success=True

Data Docs also shows:

Status: Succeeded
4/4 expectations passed
Everything is green

Problem

Even though everything looks correct, the lab is still marked as FAILED and there is no error message explaining why.

The observed minimum value is -372.22. I used -400 as the lower bound (buffer), but I was told either -372.22 or -400 should work.

What might be wrong

I suspect one of these issues:

The grader expects the exact value -372.22, not -400
The autograder checks the saved JSON file, not runtime success or Data Docs
A hidden validation step is failing that is not shown in Data Docs
The suite update is not being correctly picked up by the grader

Question

Why would the lab still fail even when:

success=True
Data Docs is green
All expectations pass

What am I missing?

Alistair_KodeKloud · June 28, 2026, 9:31am

You can refer to this solutiion

github.com

santoshkal/100DaysOfMLOps/blob/main/day-47.md

# Task: Debug a Failing Great Expectations Checkpoint

The xFusionCorp Industries ML platform team extended the `fraud_schema` suite to a second batch — `data/transactions_drifted.csv`, a week's worth of real production rows. The `drift_check` checkpoint runs the existing suite against this file and fails on its first run. Your task is to use Data Docs to diagnose which expectation failed and why, widen the offending bound in the fix-script, re-run the checkpoint, and confirm Data Docs goes green.


1. Open the Data Docs button (port `8081`). The landing page lists two past validation runs under fraud_schema:

  - `default` – Green (against the clean transactions.csv).
  - `drift_check` – red (against the drifted file).


Click into the red `drift_check` run. The failing expectation's row names the column and bound; the Observed Values section shows the actual batch statistics—enough to pick a widened bound.

2. Open `/root/code/dataquality/fix_drift.py` in the VS Code editor. The four expectations are already in the script; one of them carries the bound Data Docs flagged. Widen it so the expectation admits the observed values with a little headroom. Do not delete other expectations; the fix is a widening, not a deletion.

3. Run the script:

```
   python3 /root/code/dataquality/fix_drift.py
```

This file has been truncated. show original