Scientific honesty requires reporting everything that might show you wrong, not only what supports you.
Feynman's 1974 Caltech address. His example was Millikan's oil-drop experiment, whose successors crept toward the true value slowly because nobody wanted to report a number far from Millikan's.
In practice
- A researcher reports the three analyses that did not support the finding alongside the one that did.
- An engineer's design review lists the two failure modes they could not rule out, which is the section reviewers read first.
- Subject
- Science & Epistemology
- Essay
- Rules for weighing claims