He was arrested, bailed, and freed again. Twice, that gap ended in a killing.

The story

Simon Levy, 40, has been told by a judge at the Old Bailey that he will never be released, after being convicted of murdering grandmother Carmenza Valencia-Trujillo in March 2025 and mother-of-four Sheryl Wilkins five months later. He had already raped a third woman in the same car park where Ms Wilkins's body was later found, and was jailed in 2021 for groping two women, before sexually assaulting a prison guard in 2022, being released the next year, then sexually assaulting a woman in October 2023 and allegedly raping another in May 2024. He was recalled to prison for the May 2024 assault, then released again that August. Arrested in November 2024 over four sexual assaults on the Tube, he was freed on bail at two magistrates' court hearings in May and June 2025, failed to attend three crown court hearings, and no arrest warrant was requested. Less than two weeks after an August 11 hearing, he murdered Ms Wilkins. Between the two murders he attacked six more women. The Metropolitan Police, British Transport Police and Crown Prosecution Service have all admitted failings.

The reframe, with the manuscript

Read in sequence, the individual decisions in Simon Levy's case are not hard to understand. A magistrate grants bail on the information in front of them at that hearing. A prosecutor manages a caseload and does not request a warrant for a missed hearing that, on its own, looks administrative rather than urgent. An officer processes a recall, then a release, according to whatever the file says that week. Each of these is a small, defensible act of discretion, made by someone doing a version of their job as it is normally done.

None of those individual acts required bad faith. That is exactly what makes the outcome so hard to sit with, and exactly what the manuscript's account of discretionary systems describes.

Discretion exists at multiple stages. Officers decide where to patrol. Supervisors set operational priorities. Suspicion thresholds vary by environment. Reporting practices differ across units and neighbourhoods. Each step introduces a small degree of variation.

Levy's case is not the demographic pattern the manuscript illustrates with stop and search data, but the underlying mechanism is the same one, applied to a single individual across time instead of across a population at one moment. Bail granted here. A hearing missed there. A warrant not requested. A recall followed by a release. None of these, examined alone, is the decision that killed Sheryl Wilkins. All of them, together, were.

Once these patterns become visible, discretion stops looking like a neutral tool and begins to resemble a quiet form of power.

That is the uncomfortable part of what the Met, British Transport Police and CPS have now admitted. None of the three institutions is claiming a single officer or prosecutor broke a rule. They are admitting something closer to what this book describes: that a system built on distributed judgement calls, each one reasonable in isolation, can produce an outcome that no individual decision-maker chose and that every institution involved now has to answer for collectively. Laura, the woman attacked by Levy in 2018, told the BBC she felt "really let down by the police" and "really disappointed." That is a fair description of what happens when discretion is exercised correctly at every visible stage and still fails, structurally, to stop a pattern that was already there in the file.

The three institutions naming their own failings is, in one sense, the system working. It is also the point at which the manuscript's account becomes hardest to argue with: an institution that can identify its failings clearly only after two women are dead was not lacking the information beforehand. It was lacking a mechanism that treated a missed hearing, a recall, and a pattern of prior attacks as connected facts rather than separate, individually manageable events.