In environmental criminal justice, defendants who restore the damaged ecosystem often receive more lenient sentences. Restoration in judgment documents, however, comes from two institutionally distinct sources: pre-judgment voluntary restoration initiated by the defendant, and court-ordered restoration imposed in the operative verdict. Whether the two are associated with sentencing leniency to a similar degree, and why courts respond as they do, carries direct implications for environmental governance, since the two sources point to different policy levers: encouraging early acceptance of responsibility versus strengthening post-judgment supervision. Drawing on 92,365 first-instance environmental criminal judgments delivered in China between 2014 and 2020, this study estimates the association of each restoration type with probation, the principal observable form of sentencing leniency. Linear probability models include province, year, month, offence-type, and court-hierarchy-level fixed effects and control for the main statutory sentencing factors; logit and probit estimation, propensity score matching, entropy balancing, Oster sensitivity analysis, and a specification curve across 108 model configurations confirm the pattern. Voluntary restoration is associated with a probation probability approximately 10 percentage points higher, whereas the court-ordered association is markedly weaker and statistically unstable. We then evaluate three candidate mechanisms for the voluntary–probation association: a signalling mechanism, in which restoration conveys the defendant’s remorse and accountability; an ecological-recovery mechanism, in which leniency rewards verified recovery; and an operational mechanism, in which probation enables defendants to complete restoration tasks. Within voluntary cases, verified-effective restoration is not more strongly associated with probation than ordinary in-kind or special-fund restoration, and in-kind restoration is not more strongly associated than fund-based restoration—patterns favouring the signalling mechanism. The findings suggest that strengthening environmental governance through judicial channels requires not only encouraging early, verifiable restoration through predictable sentencing incentives, but also developing independent ecological-performance verification and safeguards against pay-for-leniency.