C.17:4.1 - Novelty: unlike which admitted set?

Novelty describes supported difference from a finite comparison corpus under one declared similarity Method. When that Method returns a calibrated similarity result on [0,1] for each like-for-like comparison, that source result keeps its declared Similarity Scale. The following transformation defines a bounded value on a corresponding declared Novelty Scale, whose meaning is difference from the corpus and whose positive polarity increases as maximum similarity falls:

Novelty = 1 - max similarity(bearer, corpus member)

A declared normalization to [0,1] may be part of the Method; state its source Scale and transformation. If the Method uses another similarity or distance Scale, state the lawful coordinate construction and resulting Scale instead of reusing this formula. Subtracting an unrestricted result from one does not make it bounded.

When the Method compares representations or observations rather than the evaluated bearers themselves, name each bearer and the value actually compared, together with the describing, projection, measurement, or other stated relation that lets the comparison support the bearer-level claim. Use a compatible basis for corpus members, or state the mapping and relevant loss. A direct like-for-like comparison of epistemes needs no extra representation relation.

A robust top-k variant is allowed when declared. The result identifies the Novelty Characteristic and Scale editions, corpus and inclusion rule, source editions, comparison window, Method, model or encoder edition, distance definition, invariances, calibration, uncertainty, ClaimScope, evidence, and intended use. Changing any load-bearing element creates a different comparison basis or result edition. Those identifiers make the result reproducible; they do not by themselves show that the value is robust. When the value materially affects a comparison or pool treatment, use diagnostics suited to the claim. For example, inspect the nearest corpus members and their distances, repeat the reading with a plausible alternative corpus or similarity Method and report the sensitivity, and remove a claimed invariance to see whether it materially changes the result. These are bounded diagnostic examples, not one mandatory algorithm. If no robustness check was performed, report the supported value and uncertainty without calling it robust.

Novelty is neither timeless originality nor a property detached from its comparison basis. A label such as Novelty@context is not an executable input and must not substitute for the result chain.