Code Quality: What Makes Good Code?
High-quality code functions correctly, is easy to read and maintain, and handles various edge cases effectively.
Contributions
Every accepted correction to this page is recorded with the exact change, so readers can see how the page improved over time.
-
Corrected an untraceable IBM defect-cost statistic, an entirely unverified University of Passau study, a mislabeled Microsoft code-review sample size, an oversimplified account of the Knight Capital incident's actual technical cause, an unverified Pivotal Labs TDD statistic, an unverified Stack Overflow survey statistic, and an unverified DORA 'elite tier' claim from a report year that did not have an elite tier.
What the page claimedArticle attributed a widely-repeated but untraceable '100:1 defect cost ratio' to a real-sounding 'IBM Systems Sciences Institute' study that traces only to uncited internal training material, not a verifiable publication. It cited an entirely unverified 2016 University of Passau study on code smells and TDD that could not be verified. It described Bacchelli & Bird's real ICSE 2013 paper as analyzing '570 code reviews' when the real paper classified 570 individual review comments, not reviews. It described the real Knight Capital incident's mechanism inaccurately as simply 'old code decommissioned years earlier' being reactivated, when the actual cause was a deployment that missed one of eight servers combined with reuse of an old activation flag for a new purpose, which inadvertently re-triggered dormant code only on the unpatched server. It attributed unverified 40%/15-25% statistics to specific named engineers (Corey Haines, Gary Bernhardt) at Pivotal Labs with no findable source. It attributed unverified 40%/1.4x statistics to the real Stack Overflow Developer Survey with no findable source in the 2023 results. And it cited an unverified '1.8x more likely to be elite tier' DORA finding from the 2022 State of DevOps Report - a report that, per DORA's own methodology that year, did not use an 'elite' tier at all (it was merged into 'High').
What was correctedIBM statistic reframed as a commonly-repeated but unverifiable industry estimate rather than a real replicating study, alongside the real Boehm order-of-magnitude finding. The unverified Passau study replaced with a general, accurate statement about TDD research findings. The Bacchelli & Bird figure corrected to '570 individual review comments.' The Knight Capital passage rewritten to describe the real mechanism (missed eighth server, reused activation flag). The Pivotal Labs and Stack Overflow unverified statistics removed, replaced with accurate general statements. The DORA claim rewritten to describe the real, repeatedly-confirmed dose-dependent relationship without the specific unverified number and impossible tier reference.
Why: Independent verification found 3 of the article's tail-section claims (Michael Feathers' 'code without tests' definition, the Software Engineering at Google readability-certification program, and Thomas McCabe's 1976 complexity threshold of 10) were fully accurate and left untouched, while the remaining claims ranged from real-but-imprecise to entirely unverified - a mixed pattern rather than wholesale fabrication.
View the full record →