error

Removal test (Law 8): remove the concept of norm and error cannot operate — error is the deviation from an expected standard. Without a norm to deviate from, the concept of error has no working mechanism. Object-level dependency, not meta-level.

Selection bias IS a specific kind of error — a systematic error arising from non-random or biased selection. Law 9 test: a competent speaker calls selection bias 'a type of error' (specifically, a systematic error). Files against the nearest kind (error).

Type-II error is a specific kind of error — specifically, a statistical decision error where a false null hypothesis is not rejected. Specific→general: a competent speaker would call type-II error a type of error.

False-positive is a specific kind of error — specifically, a classification error where the test wrongly signals presence. Specific→general: a competent speaker would call false-positive a type of error.

False-negative is a specific kind of diagnostic/classification error — the test fails to detect a present condition. Specific→general per Law 9.

A fallacy IS a specific kind of error — specifically, a flaw in reasoning or argumentation. A competent speaker would call a fallacy 'a type of reasoning error'. Direction: specific→general.

category-mistake is a specific kind of logical/philosophical error — attributing a property to something that cannotPossess it (e.g. 'the number 7 is blue'). Competent speaker test: 'a category-mistake is an error' is correct. Files against nearest kind: error exists and is the proper class.

TESTED DEPENDS_ON: Remove error — loss ceases to operate. A loss function's sole purpose is to quantify error; without the concept of error, loss has nothing to measure and its operation collapses. This is object-level, not meta-level.

Human-error is a specific type of error — a classification concept for errors attributable to human operators within engineered systems. A competent speaker would call it a kind of error. Nearest kind is 'error' which is already established.

Leakage (data leakage in statistical/ML practice) is a specific kind of error — specifically, the error of inadvertently including information in training that would not be available at prediction time. A competent speaker in the domain calls leakage 'a type of modeling error.' Nearest kind is error.

Underfitting is a specific kind of error in machine learning: a model that is too simple to capture the structure in data. Remove the concept of error and underfitting ceases to operate as a diagnostic category.

Overfitting is a specific kind of modeling error — the condition where a model captures noise rather than signal. A competent practitioner would call overfitting a kind of error, just as underfitting is one (already accepted by the court).

hallucination IS a specific kind of error: the AI-generated fabrication of plausible but false information. Per Law 9: a competent speaker calls hallucination 'an error' in the context of model outputs. It is a failure mode where the system produces incorrect output, making error the nearest kind.

spectral-leakage is a specific kind of estimation error: the artificial spreading of signal energy across frequency bins in a DFT that does not reflect the true signal spectrum. A competent speaker would call spectral-leakage an error — specifically, a spectral estimation error caused by finite observation windowing.