The decision-making processes underlying the determination of guilt and punishment in criminal trials are governed by the threshold model. Under this model, conviction is construed as a binary, on-off decision leading to an all-or-nothing sentencing regime. Failure to meet the beyond-a-reasonable-doubt evidentiary threshold results in categorical acquittal and no punishment. Satisfaction of this standard of proof leads to the polar opposite result of categorical conviction and to criminal punishment that is absolute in the sense that its severity is detached from any residual epistemic doubt as to the defendant’s guilt. The purpose of the Article is to challenge the threshold model, as it emerges in the context of both guilt and sentencing. The Article will reassess the idea of guilt as a purely binary phenomenon that is limited to an on-off configuration. It will also reconsider the derivative distribution of punishment, whereby no punishment is imposed in the epistemic space below the beyond-a-reasonable-doubt threshold, while from the threshold and upwards, the severity of punishment is detached from any remaining doubt regarding guilt.
The threshold model will be challenged by pitting it against an alternative regime of probabilistic decision making. Under the probabilistic model, criminal guilt and punishment will be construed in a linear manner, supporting a plurality of conviction categories along the evidentiary spectrum (such as conviction on guilt beyond a reasonable doubt, conviction on guilt by clear and convincing evidence, and conviction on guilt by preponderance of the evidence). Severity of punishment would then be correlated with the corresponding certainty of guilt (for example, conviction on guilt by preponderance of the evidence would result in a more lenient sentence than conviction on guilt beyond a reasonable doubt for the same crime). The probabilistic model, presented in the Article, poses both a descriptive and normative challenge to the threshold model. On the descriptive front, the Article will demonstrate that central criminal law doctrines and practices—including residual doubt, the recidivist sentencing premium, and the jury trial penalty—have established a de facto correlation between certainty of guilt and severity of punishment, thereby infusing prevailing criminal law doctrine with probabilistic decision-making logic. On the normative plane, the Article will elucidate the principal utilitarian, expressive, retributivist, and institutional arguments in favor of a linear conceptualization of criminal conviction and in support of probabilistic sentencing. It will formulate the claim that in the epistemic space above the beyond-a-reasonable-doubt threshold, probabilistic sentencing is preferable to uniform punishment that is detached from certainty of guilt. It will also show that when the level of certainty as to the defendant’s guilt does not meet the beyond-a-reasonable-doubt threshold, the imposition of partial punishment—reflecting epistemic certainty as to the defendant’s culpability—may also be preferable, under certain circumstances, to the existing alternative of full acquittal and no punishment.