JSIFS Methodology Group · 2026-02-11
In programmes where the effect of interest sits near the noise floor, the dominant determinant of the result is the measurement chain rather than the phenomenon. This is not a criticism of the teams involved; it is a structural property of working at the limit of an instrument.
The consequence is a sequencing rule. Hypothesis work performed before artifact elimination allocates scarce reasoning effort to explaining a signal that may not survive a control. The reverse order — enumerate the artifact attack surface, rank by reproducing power, define eliminating controls, then reason about mechanism — costs less and produces a defensible record either way.
A complete attack surface covers at least four families: instrument (calibration drift, saturation, non-linearity, cross-talk), environment (thermal, vibration, electromagnetic, pressure, ground motion), procedure (tare, mounting, cabling, operator variation), and analysis (leakage, selection, post-hoc thresholds, rounding). Each entry needs a mechanism statement, an honest judgement of whether it could reproduce the full magnitude, and a specific eliminating control.
The output of this work is not a verdict. It is a residual-risk statement: what remains unexplained after the controls that were actually performed. Absence of a found artifact is not positive evidence for a novel effect, and any document that reads otherwise should be revised before it circulates.