← sources

Failed a 6h async migration take-home for “too much rigour, not usable after cutover”. Fair?

source: reddit (reddit:cscareerquestions)published 2026-10-29 08:32ingested 2026-10-30 02:36

retained text

It was an async assignment with a hard 6-hour limit: no one to ask during it, just the written brief. The brief said up front that the scope was intentionally bigger than 6 hours: nobody was expected to finish, just to “migrate as much as you can defend”. The task: migrate an Informatica + Oracle ETL (4 mappings, 20 objects) to Databricks and prove parity.

What I did:

* Loaded and verified all source tables (0 diffs)
* Migrated 16 of 19 outputs, diffed row by row at tolerance 0: 11 identical, 5 with measured, explained diffs
* One mapping fully worked from source to target: the contract header, proven identical on both run dates
* Didn’t migrate the main SCD2 dimension: the delivered code couldn’t produce it (wrong system code, 7 broken joins, unknown hash). Tested downstream views on a copy of the legacy dimension instead, clearly labelled

Every decision went into assumptions, client questions and decisions docs, so it was all traceable. I didn’t skip over it.

Feedback: they liked the rigorous validation, the clear scope decisions and the questions for the client. But they rejected it because apart from the contract header, nothing would work once Oracle is switched off, and because there was too much process for the time box.

The brief’s criteria were prioritisation, parity, communication, AI use and coverage. Nothing about cutover.

Honestly, I’m fuming. It feels like I blew a great opportunity *because* I was honest about what I could and couldn’t prove.

In 6 hours, with no one to ask, would you have built the SCD on guessed logic just to get an end-to-end flow? Or is proving what you can, honestly, the more mature call?

document fetched 2026-10-30 00:23 · canonical https://reddit.com/r/cscareerquestions/comments/1xiecs3/failed_a_6h_async_migration_takehome_for_too_much