S01 · incident simulation

A competitor price feed that passes every health check

People, data and outputs in this scenario are simulated. Each stage is recorded when you commit; the reveal comes after.

4stages
35 minabout
0stages committed

The brief

You are the on-call ML engineer for the revenue-management (RM) team of a synthetic airline (simulated company). A daily job recommends a fare level (L, M or H) for each of 19 flights, using the booking pace of each departure and a competitor price per flight that arrives each morning as one file per route from a data vendor. The recommendations are published at 06:00 and analysts may override them.

The ingest job runs three health checks: the files arrive before 05:30, the merged table has 19 rows with no nulls, and every price lies between 20 and 2,000. All three have been green every day this month. Competitor prices below come from the course's synthetic airline database; the vendor, its messages and its logs are simulated.

Stage 1 of 4

Saturday 2024-05-11, 08:05. A pricing analyst (simulated) writes: "ZX104 has shown the same competitor gap for what feels like a week and the recommendation keeps saying H. ZX105 and ZX106 look the same. The other routes look normal to me."

The dashboard shows all three ingest checks green for every day since 2024-05-01.

What do you do first?

Compare each flight's competitor price with its own history (consecutive identical days against normal movement), see which of the 19 flights are affected, and check the ingest log for the last real refresh before changing anything.Replace the competitor price of the three flights with its 7-day average right away, so the recommendations stop using a number you do not trust while you investigate, and note the override in the run log for the vendor.Treat it as a market fact: competitors sometimes hold fares for a while, the health checks are green, nothing in the pipeline has changed, and the recommendations still look commercially sensible.Switch off automated recommendations for the whole network until the vendor confirms in writing that the feed is healthy, so that no further recommendation rests on an unverified input.