Your Research Code Works.. Until It Doesn’t: Rethinking Reproducibility
summary
Claims that reproducibility failures accumulate as Reproducibility Debt rather than erupting suddenly and that small shortcuts (missing deps, undocumented preprocessing, duplicated datasets) create longterm risk. Shows a taxonomy across data code docs infra people, cites a case where a new hire spends six months chasing broken runs after a predecessor leaves on three weeks notice, and embeds that knowledge into an RX prototype that scores risks probabilistically to make issues visible and manageable.
AI-written summary · about
chapters (14)
- 0:00Welcome
- 1:07Research code stops working
- 3:14Reproducibility as challenge
- 4:40Small issues accumulate
- 7:14Reproducibility depth concept
- 9:21Reproducibility debt causes
- 12:14Blame culture in reproducibility
- 15:28Reproducibility across lifecycle
- 17:35Reproducibility issues taxonomy
- 21:53Data in reproducibility failures
- 23:18Managing reproducibility risks
- 24:21Making risks visible early
- 26:09RX tool for risk management
- 26:56Q&A
related talks
35mYou got your engineering in my Data Science: Addressing the reproducibility crisis
11mTechnical Debt Crisis! When Non-Engineers Write Code
39mRight Code, Right Place, Right Time
33mThe Turing Way: A how to guide for reproducible research
35mRedundancy Driven Development - Navigating disruptive codebase changes.
39mInheriting code, and I don't mean classes
40mThe Hidden Cost of Data Debt
30mA data scientist's guide to code reviews