G.3:6 - Bias‑Annotation

CHR authoring is where many biases become “baked in” as measurement choices. Typical risks:

• Proxy bias: a convenient observable substitutes for the intended construct. Mitigation: require ObservableOf + ReferencePlane + micro‑examples; force explicit “what is being measured” rather than relying on labels.
• Population and protocol shift: a change in the sampling regime or protocol can change the interpretation or validity of a reported characteristic value, or change the characteristic’s meaning when that regime or protocol is part of its definition. Mitigation: explicit validity windows and freshness/decay expectations; edition pins for protocol definitions; RSCR triggers on freshness/decay events and evidence surface edits.
• Ordinal misuse bias: ordinal ratings treated as interval/ratio by convenience. Mitigation: publish scale type + admissible transforms; legality matrix + guard macros; reject coordinate upgrades without proof hooks.
• Cross-tradition meaning bias: an imported expression erases its source-local meaning or makes a changed bearer, scope, window, reference plane, evidence basis, or intended use disappear. Mitigation: name those values, cite exact F.17 cells and an F.9 relation only when it obtains, and keep any downstream bounded-use claim explicit under C.2.1 and F.9, with evidence-bearing reliance governed by A.10 and any applicable B.3 assurance requirement. Loss remains visible through the applicable G.Core penalty rule rather than silently altering Part F or Part G semantics.
• Metric gaming bias (QD and evaluation): changing descriptors/distances changes what “diverse” means. Mitigation: edition‑pin metric definitions and make role declarations explicit (wiring via C.18 and C.19).