Confirmation Bias

Confirmation Bias is the tendency to seek out, weight, and remember evidence that supports a belief you already hold, while discounting or forgetting evidence against it. It is not a failure of intelligence — highly capable reasoners are often better at it, because they're more skilled at constructing a case for a conclusion they've already reached.

The bias shows up most dangerously at the search stage, before any weighing happens at all. A debugger who suspects the database drops one query into a console, sees a slow number, and stops — never running the query that would have implicated the network instead. The problem isn't that they misjudged the evidence; it's that they only went looking for evidence of one kind. This is why Falsifiability matters as a discipline rather than a philosophy-class abstraction: the useful question isn't "what would confirm my theory" but "what observation would force me to abandon it," and then going and making that observation.

Confirmation bias compounds with Anchoring Effect — the first hypothesis anyone proposes in an incident channel becomes the lens everyone else's search runs through — and with Motte and Bailey, where a belief under threat retreats to a defensible version rather than being tested against the strong version anyone actually acted on. Code Review has its own flavor: a reviewer who already trusts an author reads the diff looking for reasons it's fine, and a reviewer who doesn't trust them reads the same diff looking for reasons it's broken. Neither is reading the diff.

The practical countermeasure is cheap and specific: before accepting an explanation, write down one observation that would disprove it, then go check that observation before checking the ones that would confirm it. This single reordering — falsify first — does more work than any amount of "staying open-minded."