Cyclostratigraphy’s near 100% success rate in statistical cycle identification suggests confirmation bias; absence of cyclicity is not regarded as a possible outcome. Vaughan et al 2011 (VBS) showed that the usual methods of estimating confidence levels (CLs) admit numerous false cycle detections, but in subsequent debate it is asserted that the corrections recommended by VBS do not apply in cyclostratigraphy because they lead to rejection of the expected orbital periods. Is there a deeper problem? VBS particularly criticised universal failure to correct CLs for the unavoidably multiple nature of significance tests of power spectra. However, the multiple-test problem is compounded by assumptions of unlimited freedom to vary procedures to allow for properties of individual datasets. Statistical analysis in cyclostratigraphy operates in a large variable-space, both of target hypotheses (many orbital cycles and combinations thereof), and of procedures (many pre- and syn-processing options). Each of the many data-contingent choices made before and during spectral analysis and significance-testing implies the existence of alternatives: in effect, the reported analysis is only one of many. Given that multiple experiments will eventually achieve a positive result purely by chance, unadjusted significance thresholds will result in large numbers of spurious cycle identifications, a possible explanation for observed success rates. Additional multiplicity is implied by the practice of treating CLs as a guide, rather than as a definitive signal:noise discriminator; treating CLs as movable (or even optional) negates the concept that the particular dataset is just one realisation of many permitted by the noise model; without pre-selection of a CL the statistics are meaningless. Suggestions for practical improvements include: better hypothesis formulation (with attention to the prior probability of signal preservation in an unreliable recording medium); more care in discriminating between the exploratory (hypothesis-setting) and confirmatory (hypothesis-testing) modes of data analysis; advance definition of analytical protocols; and publication of all results whether positive or negative. Reference: Vaughan et al 2011: doi:10.1029/2011PA002195.