C.29.1:6 - Bias-Annotation

Bias toward convenient summaries. Aggregates are easy to display and compute. Test them against the receiving question by constructing cases with the same aggregate and different answers. In the thermal case, the mean is sufficient for mean evolution and insufficient for the maximum.

Bias toward familiar structure. A recognized matrix, graph or algebra can make a proposed correspondence feel established. Recover the represented participants and operation conditions before applying its theorem. The route graph needs permission state when permission changes continuation.

Bias toward exact answers. Insisting on reconstruction of every source detail can hide a cheaper sufficient answer. Derive a bound over all compatible cases and compare it with the decision threshold. Conversely, report unresolved cases when that bound spans the threshold.

Bias from machine-generated fluency. An AI can produce a plausible mapping and a correct algebraic simplification while omitting the premise that connects them. Ask a contributor for the domain, operation comparison and returned conclusion. Use subject expertise to assess an unfamiliar proof; a persuasive explanation does not settle its premises.